Heart disease\u00a0is\u00a0the leading cause of adult death worldwide, making cardiovascular disease diagnosis and management\u00a0a global health priority.\u00a0An echocardiogram, or cardiac ultrasound, is one of the\u00a0most commonly used\u00a0imaging\u00a0tools\u00a0employed by physicians to diagnose\u00a0a variety of heart diseases and conditions.\u202f<\/p>\n
Most standard echocardiograms provide two-dimensional\u00a0visual\u00a0images (2D) of the three-dimensional (3D) cardiac\u00a0anatomy. These echocardiograms\u00a0often\u00a0capture\u00a0hundreds of\u00a02D slices or views\u00a0of a beating heart that can\u00a0enable\u00a0physicians to make\u00a0clinical\u00a0assessments about the function and structure of the heart.\u00a0<\/p>\n
To improve diagnostic\u00a0accuracy of cardiac conditions, researchers from UC San Francisco\u00a0set out to\u00a0determine\u00a0whether deep neural networks\u00a0(DNNs), a\u00a0type\u00a0of AI algorithm, could be\u00a0re-designed\u00a0to better capture complex\u00a03D\u00a0anatomy and physiology from\u00a0multiple imaging views simultaneously.\u00a0They developed\u00a0a new\u00a0“multiview”\u00a0DNN\u00a0structure-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\u00a0detect disease states for three cardiovascular conditions:\u00a0left and right\u00a0ventricular abnormalities,\u00a0diastolic dysfunction, and valvular regurgitation.\u00a0<\/p>\n
In a\u202fstudy\u202fpublished March 17 in\u202fNature Cardiovascular Research,\u00a0the researchers compared the performance of DNNs that analyzed data from\u00a0either single view or multiple views of the echocardiograms from UCSF and the Montreal Heart Institute.\u00a0They found that\u00a0DNNs\u00a0trained on multiple views\u00a0improved\u00a0diagnostic\u00a0accuracy\u00a0compared to DNNs trained on any single view,\u00a0demonstrating\u00a0that AI models combining\u00a0information from multiple imaging views simultaneously better captured\u00a0the disease state of these\u00a0heart conditions.\u00a0<\/p>\n
<\/p>\n
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\u00a0represent\u00a0a significant step toward maximizing AI performance in medical imaging. In the case of echocardiography, most diagnoses\u00a0necessitate considering information from more than one view because the information from any single view tells only part of the story.”\u00a0<\/p>\n
\n
<\/p>\n
Geoffrey Tison, MD, MPH, senior study author, cardiologist and\u00a0co-director of the UCSF Center for\u00a0Biosignal\u00a0Research<\/p>\n
<\/p>\n
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\u00a0and anterolateral walls),\u00a0whereas\u00a0another 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\u00a0they examined, such as\u00a0identifying\u00a0left and right ventricular abnormalities and diastolic dysfunction, the researchers’ results suggest that the\u00a0multiview\u00a0DNNs\u00a0likely learn\u00a0interrelated information between features from each view to achieve higher overall performance.\u00a0<\/p>\n
<\/p>\n
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.”\u00a0<\/p>\n
\n
<\/p>\n
Joshua Barrios, PhD,\u00a0study first author, assistant\u00a0professor, UCSF Division of Cardiology<\/p>\n
<\/p>\n
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\u00a0providing\u00a0a viable\u00a0alternative to training a\u00a0multiview\u00a0DNN. Comparatively, however, the\u00a0multiview\u00a0DNN provided the strongest performance. \u202fThey suggest that future research should examine how\u00a0multiview\u00a0DNN architectures may\u00a0assist\u00a0other medical tasks or imaging modalities.\u00a0<\/p>\n
Source:<\/p>\n
University of California San Francisco Medical Center<\/a><\/p>\n Journal reference:<\/p>\n