This PDF file contains the front matter associated with SPIE Proceedings Volume10953, including the Title Page, Copyright information, Table of Contents, Introduction, and Conference Committee listing.

Photon counting detectors (PCDs) with their low noise, high spatial and contrast resolution, and dual energy imaging capabilities are shown to be prospective candidates for Cone-Beam CT (CBCT) vascular imaging. The ultra-high frame rate capability of such detectors can enable CBCT at scan speeds far greater than that of current CBCT scanners which can only go up to about 0.14 rev/s (50 degrees/sec) compared to current multislice CT scanners that can achieve 3-4 rev/s. XCounter’s Actaeon PCD with frame rates up to 1000 fps can achieve CBCT scans at multi-slice CT scanner rates. The CdTe Actaeon PCD with pixel pitch of 100 μm and dual energy direct acquisition capability has built-in electronic anti-charge sharing correction. CBCT of a moving wire was demonstrated with the Actaeon PCD at these higher speeds and shown to be able to eliminate the effect of acquisition system image motion blur. Also a patient-specific 3D printed vascular phantom with a stent deployed was imaged at high frame rates using the Actaeon for CBCT acquisition with reconstruction using an FDK algorithm and the maximum intensity projection images clearly showed individual strut detail. This feasibility study will lead to exploration of new possibilities of high frame-rate imaging with PCDs and their potential applications.

Purpose: Time-of-flight (TOF) been successfully implemented in whole body PET, significantly improving clinical performance. However, for dedicated brain PET systems, TOF has not been a priority due the relatively small size of the human head, where coincidence timing resolution (CTR) below 200 ps is necessary to arrive at substantial performance improvements. The Brain PET (BET) consortium is developing a dual-ended PET detector block concept with ultrafast CTR, high sensitivity and high spatial resolution (X, Y, depth-of-interaction, DOI) that provides a pathway to significantly improved brain PET. Methods: We have implemented analytical and Monte Carlo models of scintillation photons transport in scintillator segments with arbitrary trans-axial cross-section dimensions. Results: Timing performance is independent of trans-axial cross-section as long as there is a gap between the scintillator and reflector wrapping. Intimate contact between the wrapping with the scintillator decreases the percentage of total internally reflected photons, degrading CTR performance. Excellent CTR performance can be achieved using simple fixed voltage thresholding techniques to determine the arrival times at the top and bottom SiPM. The average of the top and bottom arrival time corresponds to the time of gamma ray absorption, while the difference in arrival time corresponds to DOI. A simple algorithm to use the difference in arrival time to compensate for gamma ray transit time and optical photon transit achieves performance within 20% of the Cramer-Rao lower bound. We established that the advanced silicon photomultiplier designs with high single photon detection efficiency (QE=80%) and high single photon timing resolution (SPTR) ~50 ps are critical for achieving ultrafast TOF-PET performance with CTR ~50 ps and ~4 mm DOI resolution.

The design, fabrication, and performance of pre-screening tool using bioelectrical tissue profile is described. A probe of 8 electrodes using active-probe-sensing (AP-sensing) module was linearity distortion in the frequency response bonded to the 8 sub-regions of the breast surface forming the Bioimpedance transfer measurement system. Each measurement channel acquired the data from 1/8 of breast according to each breast defined in 8 sub-regions with split current injection channel and response behavioural bioelectricity detection. For this tissue bioelectrical measurement system, the electrodes were placed on the breast surface and before the bioelectrical profile were measured, we investigated a closed loop technique to compensate for the effects and measure the channel-skin-contact impedance. The AP-sensing module and connecting pads could be placed in the measurement electrode-Bras according to different breast size, and all measurement sub-regions should be equivalent for each case, but could be different in scale of 1- 2cm in related to different breast size. Bilateral structure was applied to compare each breast tissue bioelectrical properties in related to tissue behavioural in different frequency. Bioelectrical measurement efficiency was evaluated by the use of bioelectrical plots equivalent to a theoretical plot of the pure tissue profile versus average intracellular and extracellular and admittance behavioural of breast tissue able to flow electric field using Cole-Cole structure tissue modelling as a well-known bioelectrical tissue profile. The bioelectrical tissue profile as a pre-screening tool using theoretical pure tissue profiles and experimental measurements were evaluated in related a conventional Bioimpedance spectroscopy instruments. Breast bioelectrical profile of different breast density categories and average measurement values were significant, according to exist of fibro-glandular tissues in the total breast volume. The different of the theoretical values corresponding to pure fatty and fibro-glandular breast tissue behaviour was slightly different with the experimental measurements. We demonstrated the bioelectrical profile of breast tissue and extracting bioelectrical features that comparing in a bilateral structure to apply bioelectrical features as a supplementary data in the machine learning algorithms and present correspond risk factor for susceptibility of breast cancer in future studies. The preliminary equivalent theoretical and experimental results, evaluate the possibility of this new, non-imaging and label-free quantitative technique.

A new method for the detection of ionizing radiation with the potential to improve the coincidence time resolution in positron emission tomography (PET) was investigated. This method is based on Pockels effect (i.e., linear electro-optic modulation effect) and pump-probe measurement with cadmium telluride (CdTe) and lithium niobate (LiNbO₃). In this work, the performance of the two detector materials were compared experimentally. CdTe detector material demonstrated a repeatable change in modulation signal level after laser diode illumination, while LiNbO₃ crystal gave no response to laser diode as the radiation source, suggesting the shorter carrier lifetime and lower carrier mobility found of LiNbO₃ material. The modulation signal induced by 511 KeV photons in LiNbO₃ and CdTe both can be detected through the new method. We found that the CdTe crystal could provide a higher sensitivity to 511 KeV photons than the LiNbO₃ crystal under the same bias voltage. In addition, the amplitude of modulation signal increased linearly with the bias voltage before saturation. The modulation signal strength in LiNbO₃ crystal was continued to increase after 2200 V due to its high resistivity which could reduce the dark current in detector and thus reduce the noise level during experiment, while the modulation signal of CdTe with low resistivity tended to be saturated at the bias voltage of higher than 1400 V. Therefore, further increasing the bias voltage for both detector crystals may hypothetically enhance the modulation strength and detection sensitivity of PET.

Diabetic retinopathy (DR), which is a major cause of blindness in the world is characterized by hard exudate lesions in the eyes as these lesions are one of the most prevalent and earliest symptoms of DR. In this paper, a fully automated method for hard exudate delineation is described that could assist ophthalmologists for timely diagnosis of DR before disease progress to a level beyond treatment. We used a dataset consist of 107 images to develop a U-Net-based method for hard exudate detection and segmentation. This network consists of shrinking and expansive streams in which shrinking path has the same structure as conventional convolutional networks. In expansive path, obtained features are merged with those from shrinking path with the proper resolution to generate multi-scale features and accomplish distinction between hard exudate and normal tissue in retinal images. The training images were augmented artificially to increase the number of samples in the dataset and avoid overfitting issues. Experimental results showed that our proposed method reported sensitivity, specificity, accuracy, and Dice similarity coefficient of 96.15%, 80.77%, 88.46%, and 67.23 ± 13.60% on 52 test images, respectively.

The visual quality measurement of optical coherence tomography (OCT) images is very important for the diagnosis of diseases in the later stage. This paper presented a novel OCT image quality assessment method. The concept of pairwise learning in learning to rank (LTR) is introduced to extract image features sensitive to OCT image quality levels. First, a simple multi-input network (Ranking-based OCT image features extraction network) is constructed by using the residual structure. Second, the ROFE Network is trained by pairwise images. Third, the trained ROFE Network is used to extract the ranking sensitive features of OCT images. Finally, support vector regression (SVR) model is used to get the objective quality scores of OCT images. In order to verify the effectiveness of the proposed method, 608 OCT images with subjective perceptual quality are collected, and a number of experiments are carried out. The experimental results show the proposed method has strong correlations with subjective quality evaluations.

Temporal and spatial details of vascular flow patterns and rates in neuro and cardio vascular pathology are difficult to evaluate even with so-called “real-time” 15 or 30 fps x-ray imaging using flat panel detectors (FPDs). Higher frame rates of 1000 fps with high spatial resolution of 100 um pixels can demonstrate vascular flow details previously unseen by any other means. A new ultra-high temporal and spatial resolution detector with the above capabilities, the XCounter’s Actaeon, is a direct conversion photon counting detector (PCD) with built-in electronic charge sharing correction and dual energy threshold settings. This PCD was used to image a 3D-printed realistic aneurysm flow phantom injected with both iodine and gas bubble contrast media. The flow patterns, including details of the vortex flow and velocities of individual gas bubbles, were recorded in a sequence of 1 ms images and compared with frames from standard 30 fps angiographic imaging where flow patterns were blurred and individual gas bubble movement could not be observed. Because of the low-energy threshold capability of the PCD, instrumentation noise was virtually negligible enabling quantum limited performance such that, at standard angiography dose rates for FPDs, the per frame noise quality for the 1 ms frames of the PCD sequences was acceptable. This resulted in a combination of temporal, spatial, and contrast resolution unseen previously.

Differentiating a solitary brain metastasis (METS) from glioblastoma multiforme (GBM) is an important yet difficult task using current MR imaging techniques. A final diagnosis is obtained by performing a stereotactic brain biopsy, which carries a small but not insignificant risk. Distinguishing between primary and secondary malignant neoplasms is critical for providing appropriate patient prognosis, management and treatment planning. Devising non-invasive means of distinguishing between the two would be clinically useful, as patients may forego a surgical biopsy. In this study, we propose a radiomics approach for enhanced characterization and classification of such tumors. The final dataset for this study consisted of pre-treatment MRI scans acquired from 52 patients (aged 61 +/- 7 years; 31M, 21F; 35 GBM, 17 METS, 3T MRI scanner) consisting of Contrast-Enhanced (CE) T1 and T2 FLAIR sequences. The extracted features were used with an Adaboost classifier with a 10-fold cross validation scheme. The best results (AUC=0.84) were obtained using Local Binary Patterns extracted from the CE T1 sequences and also a high-dimensional feature vector from the wavelet-transformed image at the lowest frequency (AUC=0.84). Combining all the features from both the sequences resulted in a classification performance of 0.71. Our results suggest that radiomics-based machine learning analysis can accurately differentiate glioblastomas from metastatic brain tumors. Improvements in classification of such tumors could potentially reduce the need for an invasive stereotactic brain biopsy.

Background: Alzheimer’s Disease (AD) is a progressive neurodegenerative disorder, in which pathological alterations are seen in both gray matter (GM) and white matter (WM). To date functional MRI (fMRI) studies of AD have been exclusively focused on GM, since blood oxygenation level dependent (BOLD) signals in WM are relatively weak and thus ignored in practice. Our recent work provides compelling evidence that BOLD fluctuations in brain WM are reliably detectable and reflect neural activities, offering the potential of investigating the functional connectivity in WM. Purpose: In this study, we aim to apply our fMRI analysis method to the investigation of functional alterations in WM during the progression of AD. Method: Raw resting state fMRI data of normal subjects and patients (total n=290, 5 diagnostic groups) were obtained from the Alzheimer’s Disease Neuroimaging Initiative database. Each fMRI image was parcellated into 82 GM regions and 48 WM bundles. Temporal correlation between each pair of GM and WM was calculated and the correlations of all pairs constituted a functional correlation matrix (FCM) for each subject. The FCMs were averaged within each diagnostic group, and differences in the averaged FCMs between the normal group and each disease group were sought. Result: Differences in functional correlations progressively enlarge as the disease evolves, and fornix and ventral entorhinal cortices exhibited most pronounced differences between the normal and disease groups. Conclusion: Functional connectivity in WM may serve as a novel neuroimaging biomarker for the progression of AD.

Phase fMRI refers to a technique of fMRI phase imaging, which acquires the fMRI phase data to accompany the fMRI magnitude data acquisition at no extra cost. Both fMRI phase and magnitude data are generated from the same magnetic field (source), but have different properties. Under a linear phase fMRI approximation, a phase image (unwrapped) represents brain internal magnetic field. Therefore, the fMRI phase data offers, in theory, a more direct and a more accurate depiction of brain functional mapping and functional connectivity, though this comes with additional noise signal as well. In this study, we report on functional connectivity computed from a cohort of fMRI phase data (from 600 subjects). We decomposed the group phase data by independent component analysis (pICA) and calculated the phase functional network connectivity (pFC) matrix by temporal correlations of pICA timecourses. Next, we statistically analyzed the significant connection patterns in pFC. In comparison with conventional magnitude fMRI (denoted by mICA and mFC), our phase fMRI study contributed new information on resting-state brain function connectivity as follows: 1) the thresholded pFC contains a smaller number of significant connections than does the thresholded mFC; and 2) the positive and negative connections in pPNC are more balanced than those in mFC. We seek to justify the phase-inferred brain function connectivity features in the sense of using the phase representation of the brain internal field map.

The different lengths and conduction velocities of axons connecting cortical regions of the brain yield information transmission delays which are believed to be fundamental to brain dynamics. A critical step in the estimation of axon conduction speed in vivo is the estimation of the inter hemispheric transfer time (IHTT). The IHTT is estimated using electroencephalography (EEG) by measuring the latency between the peaks of specific electrodes or by computing the lag to maximum correlation on contra lateral electrodes. These approaches do not take the subject’s anatomy into account and, due to the limited number of electrodes used, only partially leverage the information provided by EEG. Using the previous published Connectivity Informed Maximum Entropy on the Mean (CIMEM) method, we propose a new approach to estimate the IHTT. In CIMEM, a Bayesian network is built using the structural connectivity information between cortical regions. EEG signals are then used as evidence into this network to compute the posterior probability of a connection being active at a particular time. Here, we propose a new quantity which measures how much of the EEG signals are supported by connections, which is maximized when the correct conduction delays are used. Using simulations, we show that CIMEM provides a more accurate estimation of the IHTT compared to the peak latency and lag to maximum correlation methods.

Background: The dentate-rubro-thalamic-tract (DRTT) has recently been suggested as a new target in treatment of tremors for patients of essential tremor and Parkinson’s disease. Diffusion MRI (dMRI) tractography was reported to able to reconstruct DRTT non-invasively in human brain. However, the performance of dMRI tractography with different setups has not been rigorously evaluated, which serves as a necessary step to optimize the protocol of reconstruction. Purpose: In this study, we aim to assess the efficacy of the dMRI-based DRTT tractography by comparing with histological “ground truth” in the same brain. Method: A fixed squirrel monkey brain was scanned in 9.4T magnet with four shells (b=3000/6000/9000/12000s/mm² ) and 100 diffusion gradient directions for each shell. Probabilistic tractography was implemented with a set of variables (i.e., b value, number of gradient directions, step length and binarization threshold), and the dMRI-derived DRTT volume was quantitatively compared with histological DRTT volume, which was obtained from Myelin stain of the same brain. Moreover, the primary orientations estimated by dMRI were compared with histological fiber orientations along the skeleton of histological DRTT. Result: The sensitivity of dMRI measure increases as the number of gradient directions increases and decreases as the binarization threshold increases. As b value increases, primary orientations estimated from dMRI agree more with histological ones along the skeleton of histological DRTT. Conclusion: Our work provides a valuable assessment of DRTT tractography, which serves as a guideline for optimization of protocol starting from image acquisition, orientation estimation to tractography.

Quantitative analyses of three-dimensional (3D) micro structures in human lungs can provide detailed information to elucidate pulmonary disease progress. The pulmonary blood vessels, have so far been studied morphologically and functionally from various aspects. However, there are problems with low spatial resolution or limited field of view. In this study, we propose an extraction method for pulmonary blood vessels like to capillary beds. 3D microstructure of vascular systems is visualized by analyzing of high-resolution dual-energy CT images using large-field high-resolution x-ray micro-CT.

Objective: Hyperpolarized noble gas magnetic resonance imaging (MRI) provides valuable insights on lung function, and yet is not widely available, whereas thoracic x-ray computed tomography (CT) protocols are nearly universally accessible. Our aim was to develop a texture analysis pipeline to train and test machine learning classifiers, predicting MRI-based ventilation metrics from single-volume thoracic CT in patients with chronic obstructive pulmonary disease (COPD). Methods: MR ventilation maps were generated and registered to thoracic CT datasets. Images were segmented into volumes of interest (15x15x15mm), resulting in approximately 6,000 volumes-of-interest per subject participant. 85 firstorder and texture features were calculated to describe each volume, including a new texture feature based on the size and occurrence of CT clusters (we called the cluster volume matrix), which is similar to run-length-matrix. A Logistic Regression, Linear Support Vector Machine and Quadratic Support Vector Machine were trained using 5-fold crossvalidation on a cohort of seven subjects. The highest performing classification model was then applied to a test cohort of three subjects. Results: There was qualitative spatial agreement for the experimental MRI ventilation maps and the CT-predicted functional maps. The training set was classified with 71% accuracy, while the test set was classified with 66% accuracy and area under the curve (AUC) = 0.72. Conclusions: This proof-of-concept study demonstrated feasibility in a small group of patients with moderate classification accuracy. Novel insights will be used to optimize this approach with future application to a larger heterogeneous patient cohort.

Chronic obstructive pulmonary disease (COPD) affects 200 million people worldwide, and is projected by the World Health Organization to be the third leading cause of death world-wide by 2030. Few drugs are available to treat COPD, and none that lead to improvements in long-term survival. A major problem for drug discovery is a poor understanding of COPD pathogenesis. Animal models of COPD rely on demonstration of emphysema and airway wall thickening on histology, which generally require 6 months of daily cigarette smoke exposure. Functional changes however may develop sooner as the disease process begins in small airways. To identify changes in lung micro-structure and function during daily cigarette smoke exposures (1 or 3 months), we used respiratory-gated micro-computed tomography (micro-CT) and image-based measurements of lung and airway volume and gas content. Mice were imaged pre-exposure, exposed daily to tobacco cigarette smoke, and imaged again. Images representing peak inspiration and end expiration were reconstructed with 0.075 mm isotropic voxel spacing. Significant differences were observed between pre-exposure and post-exposure scans for the lung volume, and air content at peak inspiration and for tidal volume in the control mice. These results suggest that the lung capacity of the mice continued to develop over the exposure period in control mice. The 3-month smoke-exposed mice exhibited increased lung volumes compared to 1-month and control groups for both respiratory phases. In vivo respiratory-gated micro-CT imaging is an effective non-invasive means of monitoring the progression of respiratory disease as early as 1 month into a smoke-exposure study.

Objective: Our aim was to develop and evaluate multi-parametric response maps derived from pulmonary x-ray computed tomography (CT), ¹H and hyperpolarized ³He static ventilation and diffusion-weighted magnetic resonance imaging (MRI). These maps were generated to phenotype patients with chronic obstructive pulmonary disease (COPD) based on the presence of airways disease, air trapping, emphysema, alveolar distension, and ventilation defects. Methods: To generate thoracic imaging multi-parametric response maps (mPRM), multispectral ¹H, ³He and CT images were segmented and co-registered. ¹H and ³He MR images were segmented using a semi-automated segmentation algorithm, the diffusion weighted MR images were segmented using a threshold-based algorithm and CT images were segmented using Pulmonary Workstation 2.0 (VIDA Diagnostics, Coralville, IA). The volume-matched segmented ¹H/³He maps were registered using landmark rigid registration. The ³He maps/the diffusion weighted images were registered using an intensity-based rigid registration. CT-to-MRI co-registration was achieved using modality-independent neighborhood descriptor (MIND) deformable registration; inspiratory and expiratory CT were co-registered using an affine registration with a deformable step provided by the NiftyReg toolkit. The co-registered thoracic maps were used to generate multiparametric maps. Results: mPRM maps were generated for six different voxel classifications with increasing disease abnormality/severity as follows: 1) ventilated voxels with >-856HU/>-950HU and normal apparent diffusion coefficient (ADC) values, 2) ventilated voxels with <-856HU/<-950HU and abnormal ADC values, 3) ventilated voxels with >-856HU/>-950HU and normal ADC values, 4) ventilated voxels with <-856HU/<-950HU and abnormal ADC values, 5) unventilated voxels with >-856HU/>-950HU, and, 6) unventilated voxels with <-856HU/<-950HU. Conclusion: mPRM measurements were automated in a dedicated pipeline for MRI and CT measurements to phenotype COPD patients.

Automated segmentation and labeling of individual brain anatomical regions is challenging due to individual structural variability. Although, atlas-based segmentation has shown its potential for both tissue and structure segmentation, the inherent natural variability as well as disease-related changes in MR appearance is often inappropriately represented by a single atlas image. In order to have a more accurate representation, several atlases may be used for the segmentation task in a given neuroimaging study. In this paper, we present the MultisegPipeline, it uses multiple atlases that have been visually inspected and capture the expected variability in a neonatal population. The MultisegPipeline transfers the labeled regions from each atlas to the target image using deformable registration (ANTs or QuickSilver is available for this task). Additionally, the set of labels are merged using a label fusion technique that reduces the errors produced by the registration. The final output is a single label map that combines the results produced by all atlases into a consensus solution. In our study, the MultisegPipeline is used to segment brain MR images from 31 infants, a leave-one-out strategy was used to test our framework. The average dice score coefficient was 0.89.

This paper presents a novel unsupervised segmentation method for the 3D microstructure in micro-computed tomography (micro-CT) images. Micro-CT scanning of resected lung cancer specimens can capture detailed and surrounding anatomical structures of them. However, its segmentation is difficult. Recently, many unsupervised learning methods have become greatly improved, especially in their ability to learn generative models such as variational auto-encoders (VAEs) and generative adversarial networks (GANs). Meanwhile, however, most of the recent segmentation methods using deep neural networks continue to rely on supervised learning. Therefore, it is rather difficult for these segmentation methods to cope with the growing number of unlabeled micro-CT images. In this paper, we develop a generative model that can infer segmentation labels by extending α-GAN, a principled combination that iterates variational inference and adversarial learning. Our method consists of two phases. In the first phase, we train our model by iterating two steps: (1) inferring pairs of continuous and discrete latent variables of image patches randomly extracted from an unlabeled image and (2) generating image patches from the inferred pairs of latent variables. In the second phase, our trained model assigns labels to patches from a target image in order to obtain the segmented image. We evaluated our method using three micro-CT images of a lung cancer specimen. The aim was to automatically divide each image into three regions: invasive carcinoma, noninvasive carcinoma, and normal tissue. Our experiments show promising results both quantitatively and qualitatively.

Advent of advanced imaging technology and better neuro-interventional equipment have resulted in timely diagnosis and effective treatment for acute ischemic stroke (AIS) due to large vessel occlusion (LVO). However, objective clinicoradiologic correlate to identify appropriate candidates and their respective clinical outcome is largely unknown. The purpose of the study is to develop and test a new computer-aided detection algorithm to quantify region-specific AIS and “at risk” brain volumes prior to thrombectomy using CT perfusion imaging protocol. Fourteen patients with LVO related AIS and assessed radiologically for their eligibility to undergo mechanical thrombectomy was retrospectively analyzed for the study. First, the scheme automatically categorizes images into multiple series of scans acquired from a section of brain. Each image in series is labeled to a specified brain location. Next, image segmentation is performed to separate brain region from skull. The brain is then split into left and right hemispheres, followed by detecting amount of blood in each hemisphere. Last, comparison between amount of blood in each hemisphere over the series of scans is made to observe the wash-in and wash-out rate of blood to assess the extent of already damaged and “at risk” brain tissue. By integrating the scheme into a user graphic interface, the study builds a unique image feature analysis and visualization tool to observe and quantify the delayed or reduced blood flow (brain “at risk” to develop AIS) in the corresponding hemisphere, which has potential to assist radiologists to quickly visualize and more accurately assess the extent of AIS.

Enlarged ventricles are a marker of several brain diseases; however, they are also associated with normal aging. Better understanding of the distribution of ventricular sizes in a large population would be of great clinical value to robustly define imaging markers that distinguish health and disease. The AGES-Reykjavik study includes magnetic resonance imaging scans of 4811 individuals from an elderly Icelandic population. Automated brain segmentation algorithms are necessary to analyze such a large data set but state-of-the-art algorithms often require long processing times or depend on large manually annotated data sets when based on deep learning approaches. In an effort to increase robustness, decrease processing time, and avoid tedious manual delineations, we selected 60 subjects with a large range of ventricle sizes and generated training labels using an automated whole brain segmentation algorithm designed for brains with ventriculomegaly. Lesion labels were added to the training labels, which were subsequently used to train a patch-based three-dimensional U-net Convolutional Neural Network for very fast and robust labeling of the remaining subjects. Comparisons with ground truth manual labels demonstrate that the proposed method yields robust segmentation and labeling of the four main sub-compartments of the ventricular system.

For the initial shoulder preoperative diagnosis, it is essential to obtain a three-dimensional (3D) bone mask from medical images, e.g., magnetic resonance (MR). However, obtaining high-resolution and dense medical scans is both costly and time-consuming. In addition, the imaging parameters for each 3D scan may vary from time to time and thus increase the variance between images. Therefore, it is practical to consider the bone extraction on low-resolution data which may influence imaging contrast and make the segmentation work difficult. In this paper, we present a joint segmentation for the humerus and scapula bones on a small dataset with low-contrast and high-shape-variability 3D MR images. The proposed network has a deep end-to-end architecture to obtain the initial 3D bone masks. Because the existing scarce and inaccurate human-labeled ground truth, we design a self-reinforced learning strategy to increase performance. By comparing with the non-reinforced segmentation and a classical multi-atlas method with joint label fusion, the proposed approach obtains better results.

Purpose: We develop an Active Shape Model (ASM) framework for automated bone segmentation and anatomical landmark localization in weight-bearing Cone-Beam CT (CBCT). To achieve a robust shape model fit in narrow joint spaces of the foot (0.5 – 1 mm), a new approach for incorporating proximity constraints in ASM (coupled ASM, cASM) is proposed. Methods: In cASM, shape models of multiple adjacent foot bones are jointly fit to the CBCT volume. This coupling enables checking for proximity between the evolving shapes to avoid situations where a conventional single-bone ASM might erroneously fit to articular surfaces of neighbouring bones. We used 21 extremity CBCT scans of the weight-bearing foot to compare segmentation and landmark localization accuracy of ASM and cASM in leave-one-out validation. Each scan was used as a test image once; shape models of calcaneus, talus, navicular, and cuboid were built from manual surface segmentations of the remaining 20 scans. The models were augmented with seven anatomical landmarks used for common measurements of foot alignment. The landmarks were identified in the original CBCT volumes and mapped onto mean bone shape surfaces. ASM and cASM were run for 100 iterations, and the number of principal shape components was increased every 10 iterations. Automated landmark localization was achieved by applying known point correspondences between landmark vertices on the mean shape and vertices of the final active shape segmentation of the test image. Results: Root Mean Squared (RMS) error of bone surface segmentation improved from 3.6 mm with conventional ASM to 2.7 mm with cASM. Furthermore, cASM achieved convergence (no change in RMS error with iteration) after ~40 iterations of shape fitting, compared to ~60 iterations for ASM. Distance error in landmark localization was 25% to 55% lower (depending on the landmark) with cASM than with ASM. The importance of using a coupled model is underscored by the finding that cASM detected and corrected collisions between evolving shapes in 50% to 80% (depending on the bone) of shape model fits. Conclusion: The proposed cASM framework improves accuracy of shape model fits, especially in complexes of tightly interlocking, articulated joints. The approach enables automated anatomical analysis in volumetric imaging of the foot and ankle, where narrow joint spaces challenge conventional shape models.

The skin is the largest organ in our body. There is a high prevalence of skin diseases and a scarcity of dermatologists, the experts in diagnosing and managing skin diseases, making CAD (Computer Aided Diagnosis) of skin disease an important field of research. Many patients present with a skin lesion of concern, to determine if it is benign or malignant. Lesion diagnosis is currently performed by dermatologists taking a history and examining the lesion and the entire body surface with the aid of a dermatoscope. Automatic lesion segmentation and evaluation of the symmetry or asymmetry of structures and colors with the help of computers may classify a lesion as likely benign or as likely malignant. We have explored a deep learning program called Deep Extreme Cut (DEXTR) and used the Faster-RCNN-InceptionV2 network to determine extreme points (left-most, right-most, top and bottom pixels). We used the ISIC challenge-2017 images for the training set and received Jaccard index of 82.2% on the ISIC testing set 2017 and 85.8% on the PH2 dataset. The proposed method outperformed the winner algorithm of the competition by 5.7% for the Jaccard index.

With the ever growing occurrences of skin cancer and limited healthcare settings, a reliable computer assisted diagnostic system is needed to assist the dermatologists for lesion diagnosis. Skin lesion segmentation on dermo- scopic images can be an efficient tool to determine the differences between benign and malignant skin lesions. The dermoscopic images in the public skin lesion datasets are collected from various sources around the world. The color of lesions in dermoscopic images can be strongly dependent on the light source. In this work, we provide a new insight on the effect of color constancy algorithms on skin lesion segmentation with deep learning algorithm. We pre-process the ISIC Challenge Segmentation 2017 dataset using different color constancy algorithms and study the effect on a popular semantic segmentation algorithm, i.e. Fully Convolutional Networks. We evaluate the results with two evaluation metrics, i.e. Dice Similarity Coefficient and Jaccard Similarity Index. Overall, our experiments showed improvements in semantic segmentation of skin lesions when pre-processed with color constancy algorithms. Further, we investigate the effect of these algorithms on different types of lesions (Naevi, Melanoma and Seborrhoeic Keratosis). We found pre-processing with color constancy algorithms improved the segmentation results on Naevi and Seborrhoeic Keratosis, but not Melanoma. Future work will seek to investigate an adaptive color constancy algorithm that could improve the segmentation results.

PET-CT scans using 18F-FDG are increasingly used to detect cancer, but interpretation can be challenging due to non-specific uptake and complex anatomical structures nearby. To aide this process, we investigate the potential of automated detection of lesions in 18F-FDG scans using deep learning tools. A 5-layer convolutional neural network (CNN) with 2x2 kernels, rectified linear unit (ReLU) activations and two dense layers was trained to detect cancerous lesions in 2D axial image segments from PET scans. Pre-contoured scans from a retrospective cohort study of 480 oesophageal cancer patients were split 80:10:10 into training, validation and test sets. These were then used to generate a total of ~14000 45×45 pixel image segments, where tumor present segments were centered on the marked lesion, and tumor absent segments were randomly located outside the marked lesion. ROC curves generated from the training and validation datasets produced an average AUC of ~<95%.

This study presents a web-system repository: Data Storage for Computation and Integration (DSCI) for Osteoarthritis of the temporomandibular joint (TMJ OA). This environment aims to maintain and allow contributions to the database from multi-clinical centers and compute novel statistics for disease classification. For this purpose, imaging datasets stored in the DSCI consisted of three-dimensional (3D) surface meshes of condyles from CBCT, clinical markers and biological markers in healthy and TMJ OA subjects. A clusterpost package was included in the web platform to be able to execute the jobs in remote computing grids. The DSCI application allowed runs of statistical packages, such as the Multivariate Functional Shape Data Analysis to compute global correlations between covariates and the morphological variability, as well as local p-values in the 3D condylar morphology. In conclusion, the DSCI allows interactive advanced statistical tools for non-statistical experts.

Alzheimer’s disease (AD) is the most common form of dementia, and an accurate diagnosis confers many clinical research and patient care benefits. The current research-setting criteria needs to consider at least one supportive biomarker before diagnosing a subject with AD, and brain atrophy measured using structural magnetic resonance is one of them. Yet, brain atrophy is currently defined using only volumetric information which could obviate localized morphological variations. We measured hippocampal neuroanatomical asymmetry from MR images of 417 subjects as a surrogate measurement of brain atrophy, anticipating that it would have a better sensitivity than volumetric information regarding differences between healthy controls and subjects with AD. Asymmetry was defined in terms of the overlapping voxels between left and right hippocampi after a co-registration process. We found a significant difference (p-value = 0.007) in discrimination power between hippocampal volume and neuroanatomical asymmetry. This result suggests that neuroanatomical asymmetry should be further studied to determine whether it could replace the current brain atrophy biomarker.

Magnetic resonance spectroscopy (MRS) has been widely used for studying metabolic changes in rheumatic, neurodegenerative diseases and several other types of pathologies. Nevertheless, the accurate measurement of brain metabolite concentrations is still problematic and challenging, specially for multivoxel MR Spectroscopic Imaging (MRSI) data. There is a collection of artifacts and spectra are acquired from a region containing mixed tissues: white matter (WM), grey matter (GM) and cerebrospinal uid (CSF) composition. However, the studies are interested in analyzing metabolite changes in a particular brain tissue or structure. Therefore, our work proposes a pipeline for automatic selection of spectra of interest, a subset of spectra from MRSI acquisitions based on MRI content analysis and spectral quality metrics. The proposed pipeline helps to improve multivoxel spectroscopy analysis and estimates of metabolite concentrations, by eliminating spectra outside the tissue or structure of interest and identifying noisy spectra.

Introduction: Traumatic brain injury (TBI) contributes to nearly a third of injury-related deaths, is the fourth leading cause of death in the U.S., and costs the U.S. ~$60 billion annually. There are two types of TBI, focal and diffuse, each requiring drastically different treatments. The current clinical standard for monitoring severe TBI is through intracranial pressure (ICP) sensing; however, significant limitations in the ICP response have motivated investigation into more multi-modal monitoring approaches. Electrical impedance has been shown to be sensitive to pathological changes within tissue including ischemia and stroke lesions. We hypothesize that by correlating electrical impedance to intracranial volume (ICV) changes we will be able to identify onset of a focal injury and localize it within the intracerebral space, overcoming many of the current limitations in TBI monitoring. Methods: A saline phantom and porcine animal model were used with controlled volume inflation steps of a Fogarty catheter. Impedance was collected across 8 electrode sectors and spatial localization capabilities compared to inclusion location. Autologous blood was then injected to simulate an intracerebral hemorrhage and the same protocol applied. Results: The phantom successfully detected inclusion presence, volume change and location. The animal model detected inclusion change with moderate success in accurately specifying location. Conclusion: Electrical impedance was successfully able to detect changes in intracranial volume in both a phantom and animal model. Additionally, initial results show potential spatial localization capabilities enabling differentiation of focal events from diffuse injury in monitoring of traumatic brain injury.

Integration of imaging and non-imaging data has been a heated topic in biomedicine. While functional magnetic resonance imaging (fMRI) can serve as endo-phenotype for mental disorders, many recent researches have confirmed the essential role played by epigenetic factors in the progress of various mental diseases including Schizophrenia(SZ), which fosters an emerging branch imaging epigenetics. In this study, we focus on the integration of fMRI and DNA methylation to have a deeper understanding of SZ: we applied a model combining Lasso with Canonical Correlation Analysis (CCA) for joint DNA methylation and fMRI analysis of 184 subjects (80 patients,104 healthy controls). In the model, the regression term focuses on extracting the discriminative features associated with the disease, while the CCA term incorporates the co-expression among extracted features.We succeeded in

Efficient segmentation of the substantia nigra (SN) in midbrain cerebral images is a prerequisite for reliable quantification and evaluation of severity of Parkinson’s disease (PD). General-purpose edge-detection techniques aren’t sufficient to for accurate segmentation due to inconsistent shape and fuzzy boundaries. Additionally, the regional properties (such as grey level) of the SN and other cerebral structures are significantly similar, and thus misclassification of segmented regions is also expected. This paper presents an algorithm for localization and segmentation of the SN in neuromelanin-sensitive magnetic resonance imaging (MRI) of the midbrain. The localization is performed using a cross-correlation template matching model in which multiple templates were used to find a match with Cerebral Peduncle, a collective structure of the SN and cerebral crus in the midbrain. We adopted a new approach that uses the parametric equation of cardioid plane curve (a curve that resembles the general structure of cerebral peduncle) to generate multiple deformable templates for localization algorithm. The segmentation of the SN is then performed using the freeform active contour segmentation model in the localized region. A total of 60 slices (10 training, 50 testing), obtained from 19 scans of 10 healthy volunteers and 9 patients with PD, were acquired using a 3T MRI system. The localization algorithm succeeded in 99.8% of the cases, while the segmentation method outperformed with an average sensitivity= 0.83, specificity = 0.97, and Dice-score = 0.73.

Generating pseudo-CT images from MRI provides electron density maps for radiation therapy planning and saves additional CT scans. Fully convolutional neural networks were proposed for pseudo-CT generation. We investigated the influence of architectures and hyperparameters on the quality of the pseudo-CT images. We used fully convolutional neural networks to transform between registered MRI and CT volumes of the pelvic region: two UNet variants using transposed convolutions or bilinear upsampling, LinkNet using residual blocks and strided convolutions for downsampling, and we designed transnet to maintain tensor spatial dimensions equal to the image’s size. Different architectures revealed similar error metrics, although pseudo-CTs differ visually. Comparison of LinkNet and UNet showed that downsampling does not affect translation. Replacing transposed convolutions with bilinear upsampling improved the pseudo-CTs’ sharpness. Translation quality quickly saturates with the number of convolution layers; increasing the number of layers from 4 to 19 decreases the MAE from 44HU to 37HU. Varying the number of feature maps showed that good translation quality can be achieved with networks that are substantially narrower than those previously published. Generally, the pseudo-CT have MAE lower than 45HU, computed inside of the true CT’s body shape.

Dental fissure sealants were shown to be effective for prevention of fissure caries in individuals with high caries risk. However, their inhibitory effect on enamel at a distance from sealant margin has not been thoroughly investigated. Previous studies have confirmed effectiveness of optical coherence tomography (OCT) for imaging of sound and demineralized enamel. The purpose of this study was to assess the potential of cross-polarization OCT (CP-OCT) as a non-destructive tool to evaluate the efficacy of fissure sealants on inhibition of demineralization of adjacent enamel. In this study, fissure sealants were applied to bovine enamel preparations (0.5×3×1 mm³, width× length× depth) and specimen surface was covered with varnish with the exception of 0.5 mm of peripheral enamel around the fissures margins. Then, demineralized enamel lesions were produced by immersion of specimens in demineralization solution (pH 4.5) for 7 days and serial cross-sectional OCT images were obtained from each specimen. The OCT images were acquired using a high scan rate (30 kHz) continuous wavelength scanning diode laser centred near 1310 nm (Santec, Japan). In addition, stereomicroscope was used to observe enamel inhibition zones on histological sections after trimming specimens to the same OCT cross-section for confirmation. CP-OCT images showed subsurface enamel lesions adjacent to fissure sealants. Also, demineralized lesions were formed away from the fissure sealants in some specimens as enamel margins resisted demineralization, forming zones of inhibition at enamel-sealant interface. This study suggests that CP-OCT could be used to nondestructively evaluate demineralization inhibition effect of dental materials in enamel.

Diabetic macular edema (DME) is a leading cause of vision loss in diabetic patients. The underlying cause for the onset of DME is the degradation of the blood-retinal barrier, whose primary function is maintaining the extracellular fluid at an optimal range. Vascular endothelial growth factor (VEGF) has proven to be a catalyst in altering the permeability of the blood-retinal barrier, thereby initiating a cascade of events that ultimately results in a loss of visual acuity.1 The primary imaging techniques to recognize and diagnose DME are fluorescein angiography (FA) and spectral-domain optical coherence tomography (SD-OCT). Taking a multimodal approach of FA in combination with SD-OCT provides images of vasculature and other eye structures to help better identify key features such as level, location, and amount of leakage.2 First-line treatments for DME have now evolved to using anti-VEGF to inhibit the effects VEGF has on increasing the permeability of the blood-retinal barrier.3 Because VEGF also increases the chance of leakage, we can also expect anti-VEGF treatments to decrease the amount of leakage DME patients suffer from. Anti-VEGF treatments also have a peripheral effect of modifying the disease burden and allowing for extended time in between treatments.4 Although current conventional treatment parameters exist to determine the efficacy of such VEGF treatments, many of these markers rely on clinicians to make a judgment call based on a minor qualitative difference of retinal scans or involve clinicians taking a fluid assessment, an option deemed too invasive to demand from all patients. In this work, we seek to find new imaging features that derive from a sub-visual feature analysis, and ideally provide a prognostic metric for clinicians to help streamline the diagnostic process. The rationale for these new biomarkers derives from leakage properties and their activity in the retina once edema develops. A decrease in leakage within certain structures in the eye would also lead to a change in the densities of leakage patterns, correlating with better clinical outcomes. In this work, we use morphological and graph-based attributes to model the global properties and spatial distribution of leakage areas on baseline FA scans of patients subsequently treated with intravitreal anti-VEGF therapy (i.e. aibercept). The features were then used in conjunction with a classifier to distinguish between eyes tolerating extended dosing intervals (N=15) and those eyes requiring more frequent dosing (N=12), based on initial response following treatment interval extension. The cross-validated area under the receiver operating characteristic curve (AUC) was found to be 0.74±0.11% using the computed imaging attributes. Edge length disorder of minimum spanning tree showed a statistically significant difference (p=0.007) between the two groups. Clinical parameters such as central subfield thickness and macular volume were not statistically significantly different. Our results indicate that there may be differences in spatial distribution of leakage areas between eyes that will exhibit favorable response to extended interval aibercept dosing and eyes that require more frequent dosing.

Diabetic macular edema is a leading cause of vision loss in diabetic patients. The underlying cause for the onset of DME is 1) the long term presence of hyperglycemia and the eventual degradation of the blood-retinal barrier (BRB) via an uptick in vascular endothelial growth factor (VEGF); VEGF increases the permeability of the blood retinal barrier and alters the length of capillaries, thereby inhibiting the ability of these vessels in performing their primary function of filtration. The lack of a proper filtration system in combination with the ongoing change in intra-retinal vasculature that stems from it, results in the eventual loss of visual acuity in DME patients. Due to the large role in which VEGF plays in acting as a catalyst for the onset of DME, current treatments now focus on utilizing anti-VEGF therapy as a first line treatment for DME. Anti-VEGF therapy improves clinical outcomes in the form of improved visual acuity and reduction in macular edema. Anti-VEGF treatments also have a peripheral effect of modifying the disease burden and allowing for extended time in between treatments. However, there is still a void in understanding how anti-VEGF affects the underlying pathophysiology. This study focuses on using quantification of the geometric properties of vasculature on Fluorescein Angiography(FA) to understand the impact anti-VEGF treatment has on retinal vascular dynamics. We hypothesize that vasculature disorder, due to VEGF action, differs across patients and can be modeled mathematically to identify candidates for anti-VEGF treatment. We use VaNgOGH, a Hough transform-based descriptor to model the disorder of the retinal vascular network on baseline FA of patients subsequently treated with intravitreal anti-VEGF therapy (aibercept). VaNgOGH computes local measures of vessel-curvature and identifies dominant peaks in the accumulator space. We explored the differences in such features on baseline FA between eyes tolerating extended dosing interval (N=15) and those eyes requiring more frequent dosing (N=12), based on initial response following treatment interval extension. The cross-validated AUC was found to be 0.73±0.1 using VaNgOGH. The variance of local orientations showed a statistically significant difference (p=0.008) between the two categories, unlike clinical parameters on baseline OCT. Our results suggest there may be fundamental differences in localized vessel orientations between eyes that will exhibit favorable response to extended interval aibercept dosing and eyes that require more frequent dosing.

Contrast agents are required to view and differentiate soft tissue structures in computed tomography (CT) yet in research, histology is still considered to be the gold standard. Preclinical contrast agents used for radiographic imaging are not visible when viewed histologically, nor are histological stains visible radiographically. By identifying a single agent that is visible in both x-ray and optical imaging, we can ensure that the target tissues can be easily identified and correlated in both images, without the need of additional staining techniques. Here we present an approach to allow for the correlation of imaging murine melanoma tumours using micro-computed tomography (micro-CT) and optical projection tomography (OPT), using fluorescently-labelled gold nanoparticles without additional tissue staining. B16F10 cells (murine melanoma cell line) were used to induce tumour growth in the right hind legs of twelve C57Bl6 mice. Tumor growth times varied between 2-3 weeks, with maximal tumor size of 1 cm³. We injected Cy3 fluorescently coated gold nanorods directly into the tumours. The mice were scanned with in vivo micro-CT (for pre- and post-contrast scans at a resolution of 50 microns) and once euthanized and hind leg dissected, further scanned with a specimen micro-CT at a higher resolution of 10-17.2 microns. Results showed that the distribution of the gold nanoparticles appeared to be contained and isolated to the murine melanoma tumour, allowing for contrast and visualization. Correlation of micro-CT specimen scans with optical projection tomography (OPT) imaging was possible, although there was also auto-fluorescence of the surrounding muscle tissue and melanoma cells. This study highlights the potential use of fluorescently-labelled gold nanoparticles as a dual-contrast agent for radiographic imaging of murine melanoma tumours using micro-CT and optical imaging using OPT.

Purpose: Investigate the use of neuro pressure gradients (NPG) to assess the impact of carotid artery disease on distal flow conditions using 3D printed patient-specific phantoms. Materials and Methods: Seven patients (five various degrees of carotid artery disease, two healthy) underwent 320- detector row CT angiography (Aquilion ONE, Canon Medical Systems). The internal carotid, vertebral, basilar arteries, as well as the Circle of Willis, middle cerebral arteries (MCA), anterior cerebral arteries (ACA), and posterior cerebral arteries (PCA) were segmented using a Vitrea workstation (Vital Images). The patient anatomy was manipulated in Autodesk Meshmixer and each phantom was 3D printed using material that simulates the compliance of vasculature, Tango+, using a Stratasys Eden260V printer. Phantoms were connected in a pulsatile flow loop with physiological flow rates. Distal resistance was manipulated to simulate physiological conditions. The pressure was measured in the proximal internal carotid arteries and the distal right and left MCA to calculate the neuro pressure gradient (NPG). Results: All seven phantoms were successfully tested in the simulated physiological flow loop. Neuro pressure gradients (NPG) were measured in each phantom and demonstrated a dependence on percent stenosis and Circle of Willis anatomy. NPG ranged -0.67 to 1.10 mmHg/cm and moderately correlated with the stenosis grade and location, and the Circle of Willis configuration. Conclusions: We have successfully assessed the feasibility of measuring NPG in 3D printed patient-specific neurovasculature phantoms with carotid artery disease.

Heart disease is the leading cause of death in the developed world.¹ Cardiac pathologies include abnormal closure of the mitral valve,² which can be treated by surgical operations, but the repair outcome varies greatly based on the experience of the surgeon. Simulating the procedure with a computer-based tool can greatly improve valve repair. Various teams are working on biomechanical models to compute the valve behaviour during peak systole.^3–5 Although they use an accurate finite element method, they also use a tedious manual segmentation of the valve. Providing means to automatically segment the chordae and the leaflets would allow significant progress in the perspective of simulating the surgical gesture for the mitral valve repair. Valve chordae are generalized cylinders: Instead of being limited to a line, the central axis is a continuous curve. Instead of a constant radius, the radius varies along the axis. In most of the cases chordae sections are flattened ellipses and classical model-based methods commonly used for vessel enhancement⁶ or vessel segmentation⁷ fail. In this paper, we exploit the fact that there are no other generalized cylinders than the chordae in the micro CT scan and we propose a topology-based method for the chordae extraction. This approach is flexible and only requires the knowledge of an upper bound of the maximum chordae radius. Examples of segmentation are provided on three porcine datasets. The reliability of the segmentation is proved with a dataset where the ground truth is available.

Computed tomography (CT) images can potentially provide insights into bone structure for diagnosis of disorders and diseases. However, evaluation of trabecular bone structure and whole bone shape is often qualitative or semiquantitative. This limits inter-study comparisons and the ability to detect subtle bone quality variations during early disease onset or in response to new treatments. In this work, we enable quantitative characterization of bone diseases through bone morphometry, texture analysis, and shape analysis methods. The potential of our analysis methods to identify the impact of hemophilia is validated in a mouse femur wound model. In our results, shape localizes and characterizes the formation of spurious bone, and our texture and bone morphometry analysis results provide extra information about the composition of that bone. Some of our one-dimensional (1D) textural features were able to significantly differentiate our injured femurs from our healthy femurs, even with this small sample size demonstrating the potential of the proposed analysis framework. While trabecular bone morphometrics have been a pillar in 3D microCT bone research for decades, the proposed analysis framework augments how we define and understand phenotypical presentation of bone disease. The contributed open source software is exposed to the medical image analysis community through 3D Slicer extensions to ensure both robustness and reproducibility.

Purpose: A high-resolution cone-beam CT (CBCT) system for extremity imaging has been developed using a custom complementary metal–oxide–semiconductor (CMOS) x-ray detector. The system has spatial resolution capability beyond that of recently introduced clinical orthopedic CBCT. We evaluate performance of this new scanner in quantifying trabecular microstructure in subchondral bone of the knee. Methods: The high-resolution scanner uses the same mechanical platform as the commercially available Carestream OnSight 3D extremity CBCT, but replaces the conventional amorphous silicon flat-panel detector (a-Si:H FPD with 0.137 mm pixels and a ~0.7 mm thick scintillator) with a Dalsa Xineos3030 CMOS detector (0.1 mm pixels and a custom 0.4 mm scintillator). The CMOS system demonstrates ~40% improved spatial resolution (FWHM of a ~0.1 mm tungsten wire) and ~4x faster scan time than FPD-based extremity CBCT (FPD-CBCT). To investigate potential benefits of this enhanced spatial resolution in quantitative assessment of bone microstructure, 26 trabecular core samples were obtained from four cadaveric tibias and imaged using FPD-CBCT (75 μm voxels), CMOS-CBCT (75 μm voxels), and reference micro-CT (μCT, 15 μm voxels). CBCT bone segmentations were obtained using local Bernsen’s thresholding combined with global histogram-based pre-thresholding; μCT segmentation involved Otsu’s method. Measurements of trabecular thickness (Tb.Th), spacing (Tb.Sp), number (Tb.N) and bone volume (BV/TV) were performed in registered regions of interest in the segmented CBCT and μCT reconstructions. Results: CMOS-CBCT achieved noticeably improved delineation of trabecular detail compared to FPD-CBCT. Correlations with reference μCT for metrics of bone microstructure were better for CMOS-CBCT than FPD-CBCT, in particular for Tb.Th (increase in Pearson correlation from 0.84 with FPD-CBCT to 0.96 with CMOS-CBCT) and Tb.Sp (increase from 0.80 to 0.85). This improved quantitative performance of CMOS-CBCT is accompanied by a reduction in scan time, from ~60 sec for a clinical high resolution protocol on FPD-CBCT to ~17 sec for CMOS-CBCT. Conclusion: The CMOS-based extremity CBCT prototype achieves improved performance in quantification of bone microstructure, while retaining other diagnostic capabilities of its FPD-based precursor, including weight-bearing imaging. The new system offers a promising platform for quantitative imaging of skeletal health in osteoporosis and osteoarthritis.

Temporomandibular Joint (TMJ) Osteoarthritis (OA) is associated with significant pain and disability. It is really hard to diagnose TMJ OA during early stages of the disease. Subchondral bone texture has been observed to change in the TMJ early during TMJ OA progression. We believe that raw probability-distribution matrices describing image texture encode important information that might aid diagnosing TMJ OA. In this paper we present novel statistical methods for High Dimensionality Low Sample Size Data (HDLSSD) to test the discriminatory power of probability-distribution matrices in computed from TMJ OA medical scans. Our results, and comparison with previous results obtained from the summary features obtained from them indicate that probability-distribution matrices are an important piece of information provided by texture analysis methods and should not be down sampled for analysis.

Glycosaminoglycans (GAG) in the extracellular matrix of the articular cartilage are biomarkers of cartilage health. Loss of GAG has been associated with early stage osteoarthritis, with zonal depletion of intra-articular GAG levels occurring prior to cartilage degeneration. Detecting this biochemical change in articular cartilage may facilitate early diagnosis of osteoarthritis. GAG is negatively-charged and repels anionic contrast media. Increased uptake of anionic contrast agents could be correlated with depleted GAG levels in the cartilage. Photon-counting detector (PCD) based computed tomography (CT) offers high-resolution imaging and x-ray energy discrimination capabilities. This allows delineation of finer anatomical structures, and the generation of quantitative material maps using energy-resolved CT data. In this study, we demonstrate quantitative GAG imaging in porcine cartilage using a research whole-body PCD-CT system and an anionic contrast agent. Hind knee joints were harvested from euthanized pigs. GAG depletion mimicking early-OA was induced using trypsin treatment. Both the control group and the trypsin-treated group were incubated in an anionic gadolinium contrast prior to PCD-CT scanning. The specimens were scanned at ultra-high resolution using the PCD-CT system at 120kV, 330mAs, and [25, 51] keV energy thresholds. An image-domain material decomposition was employed to generate the mass density map for gadolinium in cartilage using energy-resolved PCD-CT data. The results showed significantly higher gadolinium uptake (p < 0.0001) in the trypsin-treated specimens, compared to the control specimens. We demonstrated high-resolution ex vivo cartilage imaging using PCD-CT to quantify gadolinium uptake in articular cartilage as an inverse marker of GAG.

Determining the EE (End of Expiration) and EI (End of Inspiration) time points in the respiratory cycle is one key step during the 4D image construction from free-breathing dynamic thoracic computed tomography (CT) or magnetic resonance imaging (MRI) acquisitions. However, the cost of manually labeling EE and EI time points is extensive. An automatic image-based EE and EI labeling method makes image annotation independent of the image acquisition process, avoiding use of internal or external markers for the patient during image acquisition. The purpose of this paper is to introduce a novel optical-flow-based technique for finding EE and EI time points from dynamic thoracic MRI acquired during natural tidal-breathing. The diaphragm is tracked as a marker to determine the state of breathing. A region of interest (ROI) containing the diaphragm is selected to calculate the pixel optical flow values between two adjacent time slices. The average optical flow values of all pixels including diaphragm motion speed is used as a reference for labeling EE and EI. When the direction of movement of the diaphragm changes, EE or EI is found depending on the direction of the change. Quantitative evaluation was carried out to evaluate the effectiveness of our method in different locations in the lungs as compared to manual labeling. When tested on 28 patient dynamic thoracic MRI data sets, the average error was found to be less than 1 time point. Automatic labeling greatly shortened the labeling time, requiring less than 8 minutes compared to 4 hours for manual labeling per study.

T1-mapping cardiac magnetic resonance (CMR) is a rapidly expanding non-invasive tool for quantitative assessment of myocardial fibrosis. To achieve both efficiency and reproducibility in quantification of T1 measures, automated myocardial boundary tracing is desirable. Accordingly, the application of robust segmentation algorithms for this modality are of significant interest. However, conventional algorithms may fail in myocardial segmentation of T1-mapping images due to low signal gradients at the endocardial-blood pool boundary. In this work, we propose using prior information from cinematic (CINE) CMR images toward accurate myocardial segmentation of native T1-mapping images, acquired using the shortened modified Look-Locker imaging (shMOLLI) technique. We use a three-step framework, which begins with pre-processing and resizing of both CINE and shMOLLI images. Next, we implement semi-automated segmentation of the myocardium on resized CINE images using a deformable model-based technique, via the freely available software Segment v2.2. The final step of our framework is registration and propagation of the CINE contours to corresponding (slice-matched) native shMOLLI images using a non-rigid registration technique based on a modality independent neighborhood descriptor (MIND). We validate our technique on 20 image sets obtained from 20 patients with confirmed myocardial fibrosis related to ischemic injury (myocardial infarction). Our method achieved an average Dice similarity coefficient (DSC) of 84.36% ± 4.03%, precision of 91.68% ± 7.89%, recall of 91.33% ± 8.41% and relative area error of 16.29% ± 8.58%.

In this study, we investigate if differences in interaction between different brain regions for subjects with autism spectrum disorder (ASD) and healthy controls can be captured using resting-state fMRI. To this end, we investigate the use of mutual connectivity analysis with Local Models (MCA-LM), which estimates nonlinear measures of interaction between pairs of time-series in terms of cross-predictability. These pairwise measures provide a high-dimensional representation of connectivity profiles for subjects and are used as features for classification. Subsequently, we perform feature selection, reducing the dimension of the input space with the Kendall’s τ coefficient method. The Random Forests (RF) and AdaBoost classifiers are used. Performing machine learning on functional connectivity measures is commonly known as multi-voxel pattern analysis (MVPA). Traditionally, measures of functional connectivity are obtained with cross-correlation. Hence, as a metric to evaluate MCA-LM against, we also investigate classification performance with cross-correlation. The high area under receiver operating curve (AUC) and accuracy values for 100 different train/test separations across both classifiers using MCA-LM (mean AUC ranges between 0.78 - 0.85 and mean accuracy between 0.7 - 0.81) compared with standard MVPA analysis using cross-correlation between fMRI time-series (mean AUC ranges between 0.54 - 0.6 and mean accuracy between 0.50 - 0.57), across all the number of features selected demonstrates that such a nonlinear measure may be better suited at extracting information from the time-series data and has potential for the development of novel neuro-imaging biomarkers for ASD.

Temporal signal drift is one of the significant artifacts in functional Magnetic Resonance Imaging (fMRI) data that is not given as much attention as motion or physiological artifacts. However, signal drift if not accounted for, can introduce spurious correlation between different regions in resting state fMRI data. Hence detection and removal of signal drift is an important preprocessing step in fMRI data analysis. Here we propose an automated data driven approach that makes use of Principal Component Analysis (PCA) to eliminate not only low frequency signal drift but also spontaneous high frequency global signal fluctuations. This approach is also able to identify the most dominant component for each voxel separately. For task fMRI, this can help us identify regions that respond in a time locked manner to the experiment paradigm. Such regions can be thought of as activation regions. The dominant principal components corresponding to such regions can also be used to investigate intra-region Hemodynamic Response (HR) variability within subjects and across subjects.

Background: Hemodynamics is a driving factor behind remodeling of the cerebral vasculature, yet mechanisms of flowinduced remodeling remain incompletely understood. Studies employing serial imaging could help characterize hemodynamic-induced pathologic and physiologic remodeling of cerebral arteries. Methods: This preliminary study was performed us ing 4 mice. In 3, we induced flow-driven vascular remodeling in the Circle of Willis (CoW). This was done by ligation of the left common carotid artery (CCA), and the right external carot id and pterygopalatine arteries, which resulted in an increase of blood flow through the basilar artery and the right internal carotid artery. The remaining mouse was used as a wild-type control. In the 3 experimental mice, we performed 9.4 Tesla Magnetic Resonance Imaging (MRI) over a span of 3 months. 3D images were reconstructed for serial computational evaluation of gross morphological changes . These measurements were verified by the terminal vascular corrosion casting and scanning electron microscope imaging. Results: This study demonstrated the feasibility to distinguish and serially measure pathologic cerebral vascular changes in the mouse CoW, specifically in the anterior vasculature. We showed that these changes were characterized by compensatory arterial dilation and increased tortuosity on the anterior cerebral artery. From scanning electron microscope images, we also found that there was microscopic damage, akin to aneurysmal remodeling, at the right olfactory artery origin. Conclusions: MRI-based serial imaging has the potential to serially characterize gross morphological changes in the CoW in response to flow manipulation. In the future, combining this analysis with computational fluid dynamics simulations will help to define the hemodynamic environments corresponding to these and other pathologic remodeling changes in the mouse CoW.

The motivation of this study derives from the need for tumour margin estimation after surgical excision. Conventional beta autoradiography of beta emitters can be used to image tissue sections providing high spatial resolution compared to in-vivo molecular imaging. However, it requires sectioning of the specimen and it provides a 2D image of the tissue. Imaging of the 3D tissue sample can be achieved either by imaging sequential 2D sections, which is time-consuming and laborious, or by using a specialised detector for imaging that records the particles’ direction, in addition to their position, when they hit the detector. In this work we investigate whether a novel beta-tomosynthesis approach can be used for depth resolution of beta emitters. The technique involves acquiring multiple 2D images of the intact tissue sample while the detector rotates around the sample. The images are then combined and used to reconstruct the 3D position of the sources from a limited angle of conventional 2D autoradiography images. We present the results from Geant4 forward simulations and the reconstructed images from a breast tissue sample containing a Fluorine-18 positron emission source. The experiments show that the proposed method can provide depth resolution under certain conditions, indicating that there is potential for its use as a 3D molecular imaging technique of surgical samples in the future.

Small animal imaging has become essential in evaluating new cancer therapies as they are translated from the preclinical to clinical domain. However, preclinical imaging is faced with unique challenges that emphasize the gap between mouse and man. One example is the difference in breathing patterns and breath-holding ability, which can dramatically affect tumor burden assessment in lung tissue. Our group is developing quantitative imaging methods for the preclinical arm of a co-clinical trial studying synergy between immunotherapy (anti-PD-1) and radiotherapy in a soft tissue sarcoma model. To mimic imaging performed in patients, primary sarcomas lesions are imaged with micro-MRI, while detection of lung metastases is performed with micro-CT. This study addresses whether respiratory gating during micro-CT acquisition improves lung tumor volume quantitation. Accuracy and precision of lung tumor measurements was determined by performing experiments involving simulations, a pocket phantom and in vivo scans with and without prospective respiratory gating. Sensitivity and precision of segmentation with and without gating was studied using simulated lung tumors. A clinically-inspired “pocket phantom” was used during in vivo mouse scanning to aid in refining and assessing the gating protocols. Finally, we performed a series of in vivo scans on tumor-bearing mice while varying the animal’s position (test-retest), and performing the analyses in triplicate to assess the effects of gating. Application of respiratory gating techniques reduced variance of repeated volume measurements and significantly improved the accuracy of tumor volume quantitation in vivo.

High-resolution cardiac imaging and fiber analysis methods are desired for deeper understanding cardiac anatomy. Although refraction-contrast X-ray CT (RCT) has high contrast for soft tissues, its scanning cost is very high. On the other hand, micro-focus X-ray CT (μCT) is a modality that is commercially available with lower cost, but its contrast for soft tissue is not as high as RCT. To investigate the efficacy of μCT for fiber analysis, we scanned a common rabbit heart with both modalities with our original protocol of preparing materials, and compared their image-based analysis results. Their results were very similar, with correlation coefficient of 0.95. We confirmed that µCT volumes prepared by our protocol are useful for fiber analysis as well as RCT.

Image-guided neuro-endovascular interventional studies often require a high resolution image quality which requires the use of small focal-spot sizes. There is always a tradeoff while choosing focal-spot sizes. The use of small- focal-spots gives better spatial resolution but limits x-ray tube output and tube loading. The use of larger- focal-spots provides better heat dissipation capacity and higher x-ray output but gives rise to geometric blurring hence loss of spatial resolution. A method has been proposed which incorporates the use of the line-focus principle that can achieve a smaller projected focal-spot while keeping the actual focal-spot size as the medium focal-spot. Here the gantry is tilted to reduce the effective anode angle, hence the central axis is tilted producing a smaller projected focal-spot. This was tested by acquiring images of a Pipeline stent in the right jugular vein and a guidewire of a rabbit using a high-resolution CMOSbased fluoroscopic detector with 75um pixels. Also the gantry was tilted in the anode-cathode direction shifting the central axis to the anode side of the beam and the rabbit was aligned with the new perpendicular ray. Acquired images, for small focal-spot, medium focal-spot, and 7-degree tilted medium focal-spot were compared. While small focal-spot images demonstrated superior resolution, and both tilted and untilted medium focal spot images demonstrated lower noise due to the increased tube output the tilted medium focal spot images exhibited improved resolution. Line profiles and quantitative measures of generalized relative object detectability confirmed the advantage of the tilted medium focalspot method above all.

Large trials have demonstrated the prognostic value of quantifying left ventricular (LV) twist because of its crucial role in the coupling of systolic and diastolic cardiac function. Current methods for measuring LV twist evaluate rotation in a 2D plane, chosen prospectively, and the data is acquired over multiple heartbeats. In this paper, a new method for assessing 3D endocardial LV twist from single-heartbeat, ECG-gated, 4DCT volumes is proposed. In this study, the ability of the novel LV twist algorithm to accurately measure rotation in a mathematical phantom with known deformation is evaluated. The mathematical phantom was then 3D-printed to determine the accuracy of the rotation measurement from CT images in the presence of varying levels of noise. Lastly, as a proof-of-concept, LV twist was measured in human hearts across the cardiac cycle to determine whether reasonable estimates of endocardial rotation could be obtained from 4DCT studies of standard clinical quality. In both the mathematical and 3D-printed phantoms (for CNR≥9.3), the measured LV twist was highly correlated (r²≥ 0.98, p<0.001) with the known ground truth rotation function. In the healthy controls, the mean endocardial LV twist was found to be 25.3° ± 6.5° and occurred within 30-36% of the R-R interval. From these results, it is clear that 3D rotational information and LV twist can be obtained from ECG-gated 4DCT volumes. The accuracy of LV twist in clinical data requires validation via a gold standard, such as MRI-tagging

To permit optimization of 4D flow protocols in imaging of thoracic aorta, a flow phantom was designed and constructed from clear acrylic plastic. The phantom was precision machined out of clear acrylic plastic for continuous flow, ability to see unwanted air bubbles, and MR compatibility. The solid model of the phantom was designed in SolidWorks and fed to a computer numeric control (CNC) machine for precision machining. The design permits the operator to switch aortic valves constructed from a silicone mold with various degrees of calcifications (different percentage openings), modeling an aortic valve at various stages of disease. The valve opens and closes during the cardiac cycle as in the in-vivo case. The inner diameter of the tube throughout the phantom was 1”, which corresponds to human anatomical measurements in the average person. The phantom was placed in an MR compatible flow circuit, with a 60:40 distilled water/glycerol fluid mixture resulting in a viscosity of 0043 Pa*s and density of 1,060 kg/m3 similar to those of blood. The pump driving the working fluid in the phantom is programmable, capable of delivering physiologic flow rates up to peak flow of 400 ml/s The phantom was placed inside a Philips Achieva 1.5 T scanner and imaged with a 16 element XL Torso Coil. 4D flow imaging was performed at a Venc of 250 cm/s. The field of view was 120 mm x 120 mm x 150 mm, with a voxel size of 1.5 mm x 1.5 mm x 5 mm, and 14 phases. Other scan parameters were as follows: TR=11 ms, TE=4 ms and TFE factor=2.

Gated myocardial perfusion SPECT (MPS) is widely used to assess the left ventricular (LV) function. Its performance relies on the accuracy of segmentation on LV cavity. We propose a novel machine-learningbased method to automatically segment LV cavity and measure its volume in gated MPS imaging. To perform end-to-end segmentation, a multi-label V-Net is used to build the network architecture. The network segments a probability map for each heart contour (epicardium, endocardium and myocardium). To evaluate the accuracy of segmentation, we retrospectively investigated gated MPS images from 32 patients. The LV cavity was automatically segmented by the proposed method, and compared to manually outlined contours, which were taken as the ground truth. The derived LV cavity volumes were extracted from both ground truth and results of proposed method for comparison and evaluation. The mean DSC, sensitivity and specificity of the contours delineated by our method are all above 0.9 among all 32 patients and 8 phases. The correlation coefficient of the LV cavity volume between ground truth and results produced by the proposed method is 0.910±0.061, and the mean relative error of LV cavity volume among all patients and all phases is - 1.09±3.66 %. These results indicate that the proposed method accurately quantifies the changes in LV cavity volume during the cardiac cycle. It also demonstrates the potential of our learning-based segmentation methods in gated MPS imaging for clinical use.

Cardiovascular disease (CVD) has been the leading global cause of death for the last 15 years. In 2016, CVD, and resulting sudden and severe medical emergencies, accounted for 15.2 million deaths globally (1). Early CVD pathophysiology is characterized by both inflammation and microcalcification of vasculature. Currently, detection and observation of the disease at this stage are difficult. Additionally, the cause-effect relationships among symptoms remain unknown (2, 3, 4, 5). Nevertheless, inflammation and calcification have independently been connected to CVD risk (1, 6). Moreover, studies have shown that calcification specifically in the aorta is associated with an increased risk of death from CVD (7). For these reasons, this study aims to establish a method to examine and quantify the relationship between inflammation and plaque microcalcification in the descending thoracic aorta, and develop a strategy to better detect these CVD risk factors. PET/CT imaging with 2-deoxy-2-[18F]fluoro-D-glucose (FDG) and 18-Sodium Fluoride (NaF) radiotracers were used to detect plaque inflammation and vascular microcalcification, respectively. The thoracic aorta was then manually segmented on PMOD, and inflammation/microcalcification in each participant’s aorta were quantified by calculating the mean standard uptake value (SUV) for both radiotracers. The relationship between inflammation and microcalcification, as well as how both contribute to CVD, were analyzed by comparing SUVs for control participants and patients. It was found that participants with CVD have significantly more inflammation and microcalcification in this area than that among controls and that aortic inflammation and microcalcification are positively correlated with each other and with age.

There are multiple quantitative methods for assessing coronary calcifications with CT including Agatston, mass score, and volume score. Several studies have shown mass score in mg-calcium to be the most reproducible. Since we are interested in tracking changes in individual calcifications over time as a new biomarker of vascular disease, we have analyzed ways to further improve reproducibility. The conventional way to calculate calcium mass score is to sum all voxels above 130- HU and convert to mass score using a calibration constant. However, this does not account for CT system blurring or partial volumes in voxels containing both calcification and soft tissue. To improve coronary calcification measurements, we used Richardson-Lucy deconvolution with a measured impulse response (Philips IQon) and/or partial volume correction processing. At 120 kVp, we imaged a phantom with calcium inserts and calcified cadaver hearts at three rotational orientations at high (0.4883-mm, 0.67-mm-thick) and normal clinical (0.4883-mm, 2.5-mm-thick) resolution. Processing improves accuracy as the absolute difference in conventional and processed results is (12.67 mg, 8.29 mg) for conventional resolution image and (7.09 mg, 5.26 mg) for high resolution image respectively across phantom calcium inserts with known values. Deconvolution also increased contrast (and HU) of small calcifications in cadaver hearts. For low resolution images, across rotation angle, average absolute difference, as compared to high resolution images, was improved with processing by 30.8%. Processing also improves reproducibility across rotation angle. Results were similar in virtual 70 keV images.

Purpose: Various CT-FFR methods are being proposed as a non-invasive method to estimate cardiac disease severity. 3D printed patient specific cardiovascular models with high geometric accuracy can be used to simulate blood flow conditions and perform precise and repeatable benchtop flow experiments for validation of such methods. Materials and Methods: Twelve patient-specific 3D printed cardiac phantoms were created from CT Angiography (CTA) scans using a compliant 3D printing material. Pressure sensors were connected to the aortic root and distal ends of the three main coronary arteries to measure benchtop pressure gradients for each stenosed vessel. The patient geometries were used in Canon Medical Systems 1D fluid dynamics algorithm to calculate the CT- FFR. In addition, a 3D computational fluid dynamics simulation was done using ANSYS to estimate pressure gradients across the coronary arteries. Experimental data and 1D and 3D flow simulations were compared to the standard catheter lab FFR measurement (Invasive-FFR). Results: The average percent difference in Benchtop FFR/Invasive FFR, CT-FFR/Invasive FFR, and CFDFFR/Invasive FFR was 0.05, 0.06, and 0.07 respectively. The average time it took for the CT-FFR simulation was ~35 minutes and it took ~15 hours for the CFD-FFR simulation but can vary based on the number of iterations the user defines the software to run. Conclusions: Benchtop FFR proved to be highly accurate when compared to both 1D and 3D CFD software and therefore, 3D printing of patient specific coronary phantoms is a quality tool for CT-FFR software validation.

[Objective] The objective was to examine the effect of immersion of the polymeric-bonded specimens in silver-nitrate medium (SiNi) to detect miro-gaps under cross-polarization optical coherence tomography (CP-OCT). [Methodology] The twenty prepared specimens were imaged under CP-OCT (Cont group). Later, they were immersed in SiNi solution and re-imaged under CP-OCT (A-Si group) followed by cross-sectioning and imaging under a stereomicroscope. CP-OCT quantified data analyses were performed using macro-file plugged into an image analysis software. [Results] The obtained results were analyzed using Mann-Whitney test, A-Si group significantly different from Cont group (p<0.05). [Conclusion] SiNi has a great influence on micro-gap detection when using CP-OCT.

Accurate and timely segmentation of coronary vessels in quantitative coronary angiography (QCA) may be important to ensure accurate patient diagnosis. This paper compares three variations of graph search algorithms for use in segmenting coronary arteries in X-ray angiographic images. For comparing these algorithms, we propose a semi-automatic vessel segmentation technique that combines Hessian-based filtering, Gabor filtering, and graph-based search routines for tracing the boundaries1,2. This allows for a more automated procedure by incorporating automatic centerline detection while the use of Gabor filtering promotes a more natural and geometrically continuous border segmentation1. The method requires minimal effort by the user; the only manual input required is a start and end-point along the vessel of interest. Three graph search methods were compared by analyzing the accuracy and computational speed of the segmentations while using each search technique: Dijkstra’s algorithm, a restricted Dijkstra’s algorithm, and the A* search algorithm were compared. The restricted Dijkstra’s and A* approaches reduced the computational time but resulted in low accuracies or outright segmentation failures. As outlined in the paper, Dijkstra’s algorithm results in a superior segmentation with only a marginal increase in computational time.

Worldwide, back pain is the single leading cause of disability, preventing many people from engaging in work as well as other everyday activities. Back pain affects nearly 80% of the population at some point and is the most common cause for missed work (1). Most cases of back pain are mechanical or non-organic—meaning they are not caused by serious conditions, such as inflammatory arthritis, infection, fracture or cancer (2). However, there is no single procedure to identify the mechanical components related to spine pain. This study reports on the development and application of a computer aided drafting (CAD) program that identifies structural and functional deficits of the spinal pelvic system from two dimensional radiographs. When injury to the spinal system disturbs balance or visual orientation, the righting reflex activates a compensatory response using muscle contraction and mass displacement. This mechanical displacement process proceeds along the pathway of the coupled motions of gait including pelvic and spine rotation. The gait cycle is the motion pathway of the body and the gait cycle is the primary pathway of mechanical compensation. As a result of injury, the spine adapts into a compensatory non-neutral position of gait with all its associated coupled motions. This compensatory, non-neutral position of gait becomes inherent to the spinal system and is recorded on a patient’s weight bearing x-rays. Therefore, a static biomechanical model of the spine and pelvis in a non-neutral position of gait is needed for patient comparison to assess structural integrity including mechanical and functional efficiency.

Magnetic particle imaging (MPI) is a relatively young, radiation-free imaging modality that measures the interaction between superparamagnetic nanoparticles and magnetic fields. Compared to standard imaging modalities, a key feature of MPI is its ability to measure 3D volumes of relatively high spatial resolution in real-time, while still maintaining high sensitivity. Therefore, MPI is considered promising especially for vascular imaging and interventions. Yet, to fully take advantage of the unique MPI properties, real-time 4D imaging has to be combined with appropriate real-time 4D visualization and image analysis techniques. The current work aims at identification of respective clinical use cases and scenarios to illustrate the potential of MPI in the context of vascular imaging and interventions; the implementation and exploration of suitable visualization and image analysis techniques; and evaluation and comparison of the resulting image data to standard clinical imaging approaches. The study is based on three clinical use cases and associated anatomical sites: mechanical thrombectomy (anatomical structure: middle cerebral artery, segments M1 and M2); endovascular coiling (internal carotid artery aneurysm); and chemoembolization (proper hepatic artery). Implemented visualization and image analysis options are based on direct volume rendering and cover aspects like optimal view point and view angle selection and application of cut-away views. We illustrate that combining MPI imaging and 4D visualization helps to improve vascular image interpretation.

We propose to integrate patch-based anatomical signatures and an auto-context model into a machine learning framework to iteratively segment MRI into air, soft tissue and bone. The proposed segmentation of MRIs consists of a training stage and a segmentation stage. During the training stage, patch-based anatomical features were extracted from the aligned MRI-CT training images, and the most informative features were identified to train a serious of classification forests with auto-context model. During the segmentation stage, we extracted the selected features from the MRI and fed them into the well-trained forests for MRI segmentation. Our classified results were compared with reference CTs to quantitatively evaluate segmentation accuracy using Dice similarity coefficients (DSC). This segmentation technique was validated with a clinical study of 11 patients with both MR and CT images of the brain. The DSC for air, bone and soft-tissue were 97.79±0.76%, 93.32±2.35% and 84.49±5.50%. The corresponding CT Hounsfield units (HU) can be assigned to three segmented masks (air, soft tissue and bone) for generating the synthetic CT (SCT), which demonstrates the proposed method has promising potential in generating synthetic CT from MRI for MRI-only photon or proton radiotherapy treatment planning.

This paper proposes a novel method to learn a 3D non-rigid deformation for automatic image registration between Positron Emission Tomography (PET) and Computed Tomography (CT) scans obtained from the same patient. There are two modules in the proposed scheme including (1) low-resolution displacement vector field (LR-DVF) estimator, which uses a 3D deep convolutional network (ConvNet) to directly estimate the voxel-wise displacement (a 3D vector field) between PET/CT images, and (2) 3D spatial transformer and re-sampler, which warps the PET images to match the anatomical structures in the CT images using the estimated 3D vector field. The parameters of the ConvNet are learned from a number of PET/CT image pairs via an unsupervised learning method. The Normalized Cross Correlation (NCC) between PET/CT images is used as the similarity metric to guide an end-to-end learning process with a constraint (regular term) to preserve the smoothness of the 3D deformations. A dataset with 170 PET/CT scans is used in experiments based on 10-fold cross-validation, where a total of 22,338 3D patches are sampled from the dataset. In each fold, 3D patches from 153 patients (90%) are used for training the parameters, while the remaining whole-body voxels from 17 patients (10%) are used for testing the performance of the image registration. The experimental results demonstrate that the image registration accuracy (the mean value of NCCs) is increased from 0.402 (the initial situation) to 0.567 on PET/CT scans using the proposed scheme. We also compare the performance of our scheme with previous work (DIRNet) and the advantage of our scheme is confirmed via the promising results.

In clinical practice, cardiac magnetic resonance imaging (CMR) is considered the gold-standard imaging modality for the evaluation of function and structure of the left ventricle (LV). However, the quantification of LV parameters in all frames, even when performed by experienced radiologists, is very time consuming mainly due to the inhomogeneity of cardiac structures within each image, the variability of the cardiac structures across subjects and the complicated global/regional temporal deformation of the myocardium during the cardiac cycle. In this work, we employed a combination of two convolutional neural networks (CNN) to develop a fully automatic LV segmentation method for Short Axis CMR datasets. The first CNN defines the region of interest (ROI) of the cardiac chambers based on You Only Look Once (YOLO) network. The output of YOLO net is used to filter the image and feed the second CNN, based on UNet network, which segments the myocardium and the blood pool. The method was validated in CMR exams of 59 individuals from an institutional clinical protocol. Segmentation results, evaluated by metrics Percentage of Good Contours, Dice Index and Average Perpendicular distance, were 98,59% ± 4,28%, 0,93 ± 0,06 and 0,72 mm ± 0,62 mm, respectively, for the LV epicardium, and 94,98% ± 14,04%, 0,86 ± 0,13 and 1,19 mm ± 1,29 mm, respectively, for the LV endocardium. The combination of two CNNs demonstrated good performance in terms of the evaluated metrics when compared to literature results.

X-Ray image enhancement, along with many other medical image processing applications, requires the segmentation of images into bone, soft tissue, and open beam regions. We apply a machine learning approach to this problem, presenting an end-to-end solution which results in robust and efficient inference. Since medical institutions often do not have the resources to process and label the large quantity of X-Ray images usually needed for neural network training, we design an end-to-end solution for small datasets, while achieving state-of-the-art results. Our implementation produces an overall accuracy of 92%, F1 score of 0.92, and an AUC of 0.98, surpassing classical image processing techniques, such as clustering and entropy based methods, while improving upon the output of existing neural networks used for segmentation in non-medical contexts. The code used for this project is available online.¹

Accurate pressure ulcer measurement is critical in assessing the effectiveness of treatment. However, the traditional measuring process is subjective. Each health care provider may measure the same wound differently, especially related to the depth of the wound. Even the same health care provider may obtain inconsistent measurements when measuring the same wound at multiple times. Also, the measuring process requires frequent contact with the wound, which increases risk of contamination or infection and can be uncomfortable for the patient. This manuscript describes a new automatic pressure ulcer monitoring system (PrUMS), which uses a tablet connected to a 3D scanner, to provide an objective, consistent, noncontact measurement method. We combine color segmentation on 2D images and 3D surface gradients to automatically segment the wound region for advanced wound measurements. To demonstrate the system, two pressure ulcers on a mannequin are measured with PrUMS; ground-truth is provided by a clinically trained wound care nurse. The results of PrUMS 2D measurement (length and width) are within 1 mm average error and 2 mm standard deviation; the average error for the depth measurement is 2 mm and the standard deviation is 2 mm. PrUMS is tested on a small pilot dataset of 8 patients: the average errors are 3 mm, 3 mm, and 4 mm in length, width, and depth, respectively.

Quantitative evaluation of bones and ligaments around knee joint from magnetic resonance imaging (MRI) often requires the boundaries of selected structures to be manually traced using computer software. It may take several hours to delineate all structures of interest in a three-dimensional (3D) dataset used for the evaluation. Thus, providing automated tools, which can delineate knee anatomical structures can improve productivity and efficiency in radiology departments. In recent years, 3D deep convolutional neural networks (3D CNN) have been successfully used for segmentation of knee bones and cartilage. However, the key challenge is segmentation of the anterior cruciate ligament (ACL) and the posterior cruciate ligament (PCL), due to high variability of intensities in the areas of pathologies such as ligament tear. In this approach, an open source 3D CNN is adapted for segmentation of knee bones and ligaments in the knee MRI. The segmentation accuracy of ACL and PCL is improved further by atlas based segmentation technique. The atlas mask is non-rigidly aligned with the patient image based on composite of rigid and deformable vector field derived between the bone masks in the atlas and corresponding segmented bone masks in the patient image. The level set functions corresponding to particular objects of interest of the deformed atlas are used to refine segmentation of the corresponding objects in the patient image. The accuracy of the proposed method is assessed using Dice coefficient score for 50 manual segmentations of bone, cartilage and ligaments comprising of both normal and knee injury cases. Our results show that the proposed approach offers a viable alternative to manual contouring of knee MRI volume by a human reader with improved accuracy compared to the 3D CNN.

In this study, the deep learning technology was used to grade the severity of glaucoma depicted on color fundus images. We retrospectively collected a dataset of 5,978 fundus images acquired on different subjects and their glaucoma severities were annotated as none, mild, moderate, or severe, respectively, by the consensus of two experienced ophthalmologists. These images were preprocessed to generate global and local regions of interest (ROIs), namely the global field-of-view images and the local disc region images. These ROIs were separately fed into eight classical convolutional neural networks (CNNs) (i.e., VGG16, VGG19, ResNet, DenseNet, InceptionV3, InceptionResNet, Xception, and NASNetMobile) for classification purposes. Experimental results demonstrated that the available CNNs, except VGG16 and VGG19, achieved average quadratic kappa scores of 80.36% and 78.22% when trained from scratch on global and local ROIs, and 85.29% and 82.72% when fine-tuned using the imagenet weights, respectively. VGG16 and VGG19 achieved reasonable accuracy when trained from scratch, but they failed when using imagenet weights for both global and local ROIs. Among these CNNs, DenseNet had the highest classification accuracy (i.e., 75.50%) based on pre-trained weights when using global images, as compared to 65.50% when using local optic disc images.

Chordoma is a rare type of tumor that usually appears in the bone near the spinal cord and skull base. Due to their location in the skull base and diverse appearance in size and shape, automatic segmentation of chordoma tumors from magnetic resonance images (MRI) is a challenging task. In addition, similar MR intensity distributions of different anatomical regions, specifically sinuses, make the segmentation task from MRI more challenging. In comparison, most of the state-of-the-art lesion segmentation methods are designed to segment pathologies inside the brain. In this work, we propose an automatic chordoma segmentation framework using two cascaded 3D convolutional neural networks (CNN) via an auto-context model. While the first network learns to detect all potential tumor voxels, the second network fine-tunes the classifier to distinguish true tumor voxels from the false positives detected by the first network. The proposed method is evaluated using multi-contrast MR images of 22 longitudinal scans from 8 patients. Preliminary results showed a linear correlation of 0.71 between the detected and manually outlined tumor volumes, compared to 0.40 for a random forest (RF) based method. Furthermore, the response of tumor growth over time, i.e. increasing, decreasing, or stable, is evaluated according to the response evaluation criteria in solid tumors with an outcome of 0.26 kappa coefficient, compared to 0.13 for the RF based method.

Shape analysis is an important method used in neuroimaging research community due to its potential to precisely locate morphological changes between healthy and pathological structures. A popular shape analysis framework in the neuroimaging community is based on the encoding surface locations as spherical harmonics for a representation called SPHARM-PDM. The SPHARM-PDM pipeline takes a set of brain segmentation of a single brain structure (for example, hippocampus) as input and converts them into a corresponding spherical harmonic description (SPHARM), which is then sampled into triangulated surface (SPHARM-PDM). At present, the SPHARM-PDM pipeline utilizes an area-preserving optimization of the spherical mapping based on an initial heat-equation based mapping of the surface mesh to the unit sphere. In the case of objects with complex shape, this initial spherical mapping suffers from a high degree of mapping distortion that cannot always be corrected by the following optimization procedure. Here we proposed the use of an alternative initialization based on a conformal flattening. This method adopts a bijective angle preserving conformal flattening scheme to replace the heat equation mapping scheme as initialization for use in the SPHARM-PDM pipeline. After quantitative measures of shape calculated from various complex structures, we concluded that in most cases, the new pipeline produced dramatically better results than the old pipeline. The main contribution of this paper is a command line tool based on the Slicer Execution Model, which merges the conformal flattening into the SPHARM-PDM pipeline for use in the SALT shape analysis toolbox.

A classification method that integrates delta-radiomic features, DSC MRI, and random forest approach on the glioblastoma classification task is proposed. 25 patients, 13 high and 12 low-grade gliomas, who underwent the standard brain tumor MRI protocol, including DSC MRI, were included. Tumor regions on all DSC MRI images were registered to and contoured in T2-weighted fluid-attenuated-inversion-recovery (FLAIR) images. These contours and its contralateral regions of the normal tissue were used to extract delta-radiomic features before applying feature selection. The most informative and non-redundant features were selected to train a random forest to differentiate high-grade (HG) and low-grade (LG) gliomas. These were then fed into a leaveone- out cross-validation random forest to classify these tumors for grading. Finally, a majority-voting method was applied to reduce binarization bias and to combine the results of various feature lists. Analysis of the predictions showed that the reported method consistently predicted the tumor grade of 24 out of 25 patients correctly (0.96). Finally, the mean prediction accuracy was 0.95±0.10 for HG and 0.85±0.25 for LG. The area under the receiver operating characteristic curve (AUC) was 0.94. This study shows that delta-radiomic features derived from DSC MRI data can be used to characterize and determine the tumor grades. The radiomic features from DSC MRI may be used to elucidate the underlying tumor biology and response to therapy.

Recent progress in artificial intelligence provides researchers with a powerful set of machine learning tools for analyzing brain imaging data. In this work, we explore a variety of classification algorithms and functional network features derived from resting-state fMRI data collected from clinical high-risk (prodromal schizophrenia) patients and controls, trying to identify features predictive of conversion to psychosis among a subset of CHR patients. While there are many existing studies suggesting that functional network features can be highly discriminative of schizophrenia when analyzing fMRI of patients suffering from the disease vs controls, few studies attempt to explore a similar approach to actual prediction of future psychosis development ahead of time, in the prodromal stage. Our preliminary results demonstrate the potential of fMRI functional network features to predict the conversion to psychosis in CHR patients. However, given the high variance of our results across different classifiers and subsets of data, a more extensive empirical investigation is required to reach more robust conclusions.

Electrical Impedance Tomography (EIT) is a method used to record the impedance distribution within a target. The best-known application of EIT is lung diagnostics using imaging algorithms. However, apart from this, there are individual research projects dealing with imaging analysis at the cellular level. For example, cell analysis using EIT can help to distinguish diseased cells from healthy cells. To achieve this goal, an existing EIT system was combined with a new EIT chip in a first step. This chip allows analysis in very small dimensions. Various parameters such as the diameter of the measuring environment, the necessary conductive solutions or the measuring methods used were varied and evaluated. In a next step, various image reconstructions were carried out using data acquired with C. elegans.

Magnetic Resonance Electrical Properties Tomography (MREPT) is an imaging modality that uses MR data to directly calculate the conductivity of the imaged object. This study evaluates if MREPT can be used to image differences between cancerous and benign prostate tissue. A total of 39 freshly excised prostates were imaged. MR data and four MREPT approaches were analyzed. Including a new MREPT approach that overlaps tiles (subdomains) resulting in an efficient approach that minimizes artifacts. No direct threshold value was found to differentiate the malignant from benign tissues. However, significance differences were found when comparing malignant and benign differences (differenced on a per slice basis), which reveals there are measurable differences between the two tissues. Ongoing work aims to develop a calibration technique that can exploit these differences so that malignant tissue can be robustly identified.