Thermoacoustic imaging (TAI) combines microwave energy's penetration depth with ultrasound's spatial resolution for medical imaging. Denoising is crucial in TAI to refine low energy thermoacoustic signals, overcoming depth limitations and improving imaging precision. We utilized the MI-TAT system to capture signals from different phantoms and gather data for training and validation. Our architectural approach harnesses both time and spatial signal features, enabling the design of an advanced deep-learning model.
KEYWORDS: Image restoration, Deep learning, Brain, Signal attenuation, Photoacoustic spectroscopy, Education and training, Data acquisition, Signal to noise ratio, Network architectures, Gallium nitride
We developed a deep learning algorithm, called enhancement Unet (E-Unet), to improve the signal-to-noise ratio (SNR) of signals acquired in a photoacoustic computed microscopy (PAM) system. We tried various combination of custom loss functions which included peak-amplitude, peak-position and mean-squared signal value with Adam optimizer for training purposes. For the testing purposes, we acquired PAM data with complicated phantoms in biological tissue. The performance of the improved signals is evaluated in terms of SNR, structural similarity index (SSIM), root mean square error (RMSE) and Pearson correlation.
During medical investigations of the head, ultrasound measurements can offer information with simple, non-invasive, and real-time procedure. However, for human adult applications, the clinical treatment of transcranial acoustic imaging remains a challenge by the presence of the skull, results in acoustic aberrations caused by two main phenomena, i.e., attenuation and distortion. These aberrations may affect the signal understanding because of the induced artifacts and the inaccuracy of the imaging target structural information. Variations of the physical properties of the skull, its thickness and porosity, will strongly affect the mechanical properties of the medium and thus the acoustic response. We propose a method to understand the influence of these characteristics on the signal degradation. In order to mimic the human adult skull, a large quantity of epoxy resin-based phantoms is created to explore all the possible physical characteristic variation in the bone. Additional components, titanium dioxide and seeds, will be added to the samples to recreate the acoustic scattering effects of a skull bone. Signal features from pulse-echo mode ultrasound, such as signal attenuation or broadening, will be extracted and studied in the time and frequency domain. In this paper, we are looking for relationship between these physical parameters and the signal features, with the objective to determine bone characteristics without any direct access in later experiments; and going a step further into aberration correction during transcranial imaging procedure.
Photoacoustic microscopy (PAM) is a high-resolution imaging modality capable of visualizing fine microvasculature in a biological tissue. Clinical translation of PAM system is still an issue due to the use of expensive and high energy laser sources for imaging. Although low energy laser sources can facilitate clinical transition of PAM systems as they are rugged, portable, affordable, and safe to use, the photoacoustic (PA) signal they generate is very weak, resulting in very low signal-to-noise ratio (SNR) PA signals and in turn low quality PA images. In this study, we have developed an enhancement autoencoder (EAE) utilizing fully convolutional neural network, that can improve the quality of the PA signals received, consequently improving the SNR of the reconstructed images. We acquired PAM data from rat brain tissue with both high energy (target data) and low energy (input data) of the laser for training purposes and tested our trained model on PAM data obtained from new rat brains. Our effort is to reconstruct the vascular structure as well as an accurate reading for the blood concentration. The latter has been neglected in the previous studies. The performance of our EAE is evaluated in terms of SNR, structural similarity index (SSIM), root mean square error (RMSE) and correlation.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.