Ultrasound is a commonly used modality for medical imaging. While this modality has great advantages in terms of safety and cost relative to other imaging modalities, it also has several limitations. Signal-to-noise ratio varies greatly depending on the acoustic properties of the tissue being imaged and the depth of the target structures. In this work, we evaluate the use of deep learning based methods to reconstruct 3D surfaces of general objects imaged with ultrasound. We evaluate three variants of the 3D U-Net with different training scenarios. We were able to train networks to reconstruct three distinct categories of objects relatively well when trained on limited data from each category. However, the performance of the networks did not generalize well when testing on categories of objects not included in the training. We also investigated the effects of employing dual-task autoencoding on generalizability. These results provide a baseline for exploring modifications to the U-Net framework to improve generalizability. A generalizable method could improve visualization for a number of ultrasound imaging tasks.
We previously showed that domain adaptive deep neural networks (DNNs) can outperform delay-and-sum (DAS) beamforming in the context of abdominal imaging. We hypothesize the ability of our domain adaptive DNN framework to be applied to transthoracic echocardiography (TTE). We also propose architectural improvements, such as leveraging an encoder-decoder structure and skip connections, to further improve ultrasound image quality for echocardiography tasks such as the detection of thrombi in the left atrial appendage (LAA). DNN training data utilized simulated and in vivo cardiac data. Simulated anechoic and hypoechoic cysts with various amounts of clutter were generated through Field II and in vivo data was collected by scanning patients at Vanderbilt University Medical Center. Fundamental frequency TTE data from five separate cases were processed with DAS, ADMIRE, the baseline model, and multiple models with modified architectures. We found that even when varying the amount of training data, the DNNs consistently achieved higher generalized contrast-to-noise (gCNR) and contrast ratio (CR) but lower contrast-to-noise ratio when compared to DAS. The best-performing beamformer was one DNN with our architectural improvements, achieving higher average gCNR and CR values of .907 and 48.30 dB compared to the baseline DNN values of .788 and 39.45dB, and DAS values of .717 and 14.08dB. Our results demonstrate that our domain adaptive DNN can effectively be applied in the context of transthoracic cardiology, and an encoder-decoder architecture with skip connections can result in even more improvements. Further advancements may improve image quality even more.
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.