Optical coherence tomography (OCT) is a high resolution and non-invasive imaging modality that has become one of the most prevalent techniques for ophthalmic diagnostic. However, manual segmentation is often a time-consuming and subjective process. In this work, we present a new method for retinal layer segmentation in retinal optical coherence tomography images, which uses a deep convolutional feature to train a structured random forest classifier. The experimental results show that our method achieves good results with the mean distance error of 1.45 pixels whereas that of the state-of-the-art was 1.68 pixels, and achieve a F-score of 0.86 which is also better than 0.83 that is obtained by the state-of-the-art method.
In the retinal image, characteristics of fluid have great significance for diagnosis in eye disease. In the clinical, the segmentation of fluid is usually conducted manually, but is time-consuming and the accuracy is highly depend on the expert’s experience. In this paper, we proposed a segmentation method based on convolution neural network (CNN) for segmenting the fluid from fundus image. The B-scans of OCT are segmented into layers, and patches from specific region with annotation are used for training. After the data set being divided into training set and test set, network training is performed and a good segmentation result is obtained, which has a significant advantage over traditional methods such as threshold method.
Age-related Macular Degeneration (AMD) is a kind of macular disease which mostly occurs in old people，and it may cause decreased vision or even lead to permanent blindness. Drusen is an important clinical indicator for AMD which can help doctor diagnose disease and decide the strategy of treatment. Optical Coherence Tomography (OCT) is widely used in the diagnosis of ophthalmic diseases, include AMD. In this paper, we propose a classification method based on Multiple Instance Learning (MIL) to detect AMD. Drusen can exist in a few slices of OCT images, and MIL is utilized in our method. We divided the method into two phases: training phase and testing phase. We train the initial features and clustered to create a codebook, and employ the trained classifier in the test set. Experiment results show that our method achieved high accuracy and effectiveness.
Image denoising is a crucial step before performing segmentation or feature extraction on an image, which affects the final result in image processing. In recent years, utilizing the self-similarity characteristics of the images, many patch-based image denoising methods have been proposed, but most of them, named the internal denoising methods, utilized the noisy image only where the performances are constrained by the limited information they used. We proposed a patch-based method, which uses a low-rank technique and targeted database, to denoise the optical coherence tomography (OCT) image. When selecting the similar patches for the noisy patch, our method combined internal and external denoising, utilizing the other images relevant to the noisy image, in which our targeted database is made up of these two kinds of images and is an improvement compared with the previous methods. Next, we leverage the low-rank technique to denoise the group matrix consisting of the noisy patch and the corresponding similar patches, for the fact that a clean image can be seen as a low-rank matrix and rank of the noisy image is much larger than the clean image. After the first-step denoising is accomplished, we take advantage of Gabor transform, which considered the layer characteristic of the OCT retinal images, to construct a noisy image before the second step. Experimental results demonstrate that our method compares favorably with the existing state-of-the-art methods.
Changes in the structure of the retina can reflect a variety of pathological physiological changes. To analyze the structural characteristics of the retina layers, an automated segmentation algorithm of retinal layers was developed. This algorithm, based on the Dijkstra's algorithm, limiting the search region with statistics information of layer thickness information, constructs a graph from every 2D OCT image and use a shortest path algorithm to iteratively segment multiple layers. The experiments showed that this segmentation algorithm has great repeatability, accuracy, and high efficiency.
Computer-aided diagnosis (CAD) system is helpful for lesion detection. In this study, we proposed a new mass detection method with analysis of bilateral mammograms. First of all, the mass candidates were detected in single view. To utilize the information in dual view, we match corresponding regions in mediolateral oblique (MLO) and craniocaudally (CC) views of the breast. In this paper, we introduced twin support vector machines (TWSVM) as classifier for mass detection, and proposed a new method for feature selection called multiple twin support vector machines (MTWSVM-RFE) to improve the accuracy of detection.
Image denoising is a very important step in image processing. In recent years, a lot of image denoising algorithms have been proposed, several of them are transform domain based methods, such as wavelet, contourlet, and shearlet. Shearlet is a new type of multiscale geometric analysis tool, which can obtain a sparse representation of the image and produce the optimal approximation. The transform generates shearlet functions with different features by scaling, shearing, and translation of the basic functions. In this paper, we introduced shearlet transformation into optical coherence tomography images to reduce noise, and proposed a multiscale, directional adapted speckle reduction method. Experiment results showed the effectiveness of the proposed method.
Architecture distortion is one of the most common signs of breast cancer in mammograms, and it is difficult to detect due to its subtlety. Computer-Aided Diagnosis (CAD) technology has been widely used for the detection and diagnosis of breast cancer. In this paper, Gabor filters and phase portrait analysis are used to locate suspicious regions based on the image characteristic of architectural distortion. Twin bounded Support Vector Machine (TWSVM), a kind of binary classifier, is employed reduce the large amounts of false positives. In this paper, we proposed a novel feature selection which is based on Multiple Twin Bound Support Vector Machines Recursive Feature Elimination (MTWSVM-RFE). The results showed that our proposed method detect the region of architecture distortion with high accuracy.
Mammogram is currently the best way for early detection of breast cancer. Mass is a typical sign of breast cancer,
and the classification of masses as malignant or benign may assist radiologists in reducing the biopsy rate without
increasing false negatives. Typically, different geometry and texture features are extracted and utilized to train a
classifier to classify a mass. However, not each feature is equally important for a classifier, and some features may
indeed decrease the performance of a classifier. In this paper, we investigated the usage of semi-supervised feature
selection method for classification. After a mass is extracted from a ROI (region of interest) with level set method.
Morphological and texture features are extracted from the segmented regions and surrounding regions. SSLFE (Semi-
Supervised Local Feature Extraction, proposed in our previous work) is utilized to select important features for KNN
classifier. Mammography images from DDSM were used for experiment. The experimental result shows that by
incorporating information embedded in unlabeled data, SSLFE can improve the performance compared to the method
without feature selection and traditional Relief method.