Researchers use multiview deep learning to enhance echocardiogram analysis

Heart disease is the leading cause of adult death worldwide, making cardiovascular disease diagnosis and management a global health priority. An echocardiogram, or cardiac ultrasound, is one of the most commonly used imaging tools employed by physicians to diagnose a variety of heart diseases and conditions. 

Most standard echocardiograms provide two-dimensional visual images (2D) of the three-dimensional (3D) cardiac anatomy. These echocardiograms often capture hundreds of 2D slices or views of a beating heart that can enable physicians to make clinical assessments about the function and structure of the heart. 

To improve diagnostic accuracy of cardiac conditions, researchers from UC San Francisco set out to determine whether deep neural networks (DNNs), a type of AI algorithm, could be re-designed to better capture complex 3D anatomy and physiology from multiple imaging views simultaneously. They developed a new "multiview" DNN structure-or architecture-to enable it to draw information from multiple imaging views at once, rather than the current approach of using only a single view. They then trained demonstration DNNs using this architecture to detect disease states for three cardiovascular conditions: left and right ventricular abnormalities, diastolic dysfunction, and valvular regurgitation. 

In a study published March 17 in Nature Cardiovascular Research, the researchers compared the performance of DNNs that analyzed data from either single view or multiple views of the echocardiograms from UCSF and the Montreal Heart Institute. They found that DNNs trained on multiple views improved diagnostic accuracy compared to DNNs trained on any single view, demonstrating that AI models combining information from multiple imaging views simultaneously better captured the disease state of these heart conditions. 

Until now, AI has primarily been used to analyze one 2D view at a time-from either images or videos-which limits an AI algorithm's ability to learn disease-relevant information between views. DNN architectures that can integrate information across multiple high-resolution views represent a significant step toward maximizing AI performance in medical imaging. In the case of echocardiography, most diagnoses necessitate considering information from more than one view because the information from any single view tells only part of the story." 

Geoffrey Tison, MD, MPH, senior study author, cardiologist and co-director of the UCSF Center for Biosignal Research

For example, for the assessment of left ventricle (LV) size or function, the echocardiogram view showing all the chambers of the heart at once (A4c) best captures certain left ventricular walls (inferoseptal and anterolateral walls), whereas another perpendicular echo view (A2c) captures other important walls (anterior and inferior walls). Often the function of LV walls may appear completely normal in one view but have significant dysfunction in another view. For the echocardiogram tasks they examined, such as identifying left and right ventricular abnormalities and diastolic dysfunction, the researchers' results suggest that the multiview DNNs likely learn interrelated information between features from each view to achieve higher overall performance. 

Our multi-view neural network architecture is explicitly designed to enable the model to learn complex relationships between information in multiple imaging views. We find that this approach improves performance for diagnostic tasks in echocardiography, but this new AI architecture can also be applied to other medical imaging modalities where multiple views contain complimentary information." 

Joshua Barrios, PhD, study first author, assistant professor, UCSF Division of Cardiology

The researchers also found that averaging the predictions of three single-view DNNs improves performance beyond a single-view DNN while also being less computationally expensive, thus providing a viable alternative to training a multiview DNN. Comparatively, however, the multiview DNN provided the strongest performance.  They suggest that future research should examine how multiview DNN architectures may assist other medical tasks or imaging modalities.