AI System PanEcho Shows High Accuracy in Comprehensive Echocardiogram Analysis, Potentially Reducing Wait Times for Results

An artificial intelligence program called PanEcho may reduce wait times for echocardiogram results and facilitate more timely medical care, according to research presented at the American Heart Association's Scientific Sessions 2024. PanEcho represents the first AI system capable of automatically assessing all key areas of heart health from echocardiograms with images from multiple views, identifying which views are most relevant for each imaging task.

The system builds on previous AI applications in cardiology that were limited to single heart views and disease-specific criteria. Researchers developed this novel AI system for comprehensive reporting of all major findings from any set of echocardiography videos. Gregory Holste, M.S.E., a researcher with the Cardiovascular Data Science (CarDS) Lab at the Yale School of Medicine, stated that PanEcho has potential for use in simplified, AI-assisted screening echocardiograms, particularly in settings where expert readers may not be readily accessible.

PanEcho's diagnostic performance was evaluated using the area under the receiver operating characteristic curve (AUC), with a perfect test scoring 1.0 and random guessing scoring 0.5. When evaluated across 18 different diagnostic classification tasks, PanEcho achieved an average AUC score of 0.91. Specific accuracy scores included 0.95 AUC for detecting increased left ventricle size, 0.98 AUC for identifying left ventricle systolic dysfunction, and 0.91 AUC for detecting left ventricle hypertrophy.

For valvular disease diagnosis, PanEcho demonstrated exceptional accuracy with 0.99 AUC for identifying severe aortic stenosis, 0.96 AUC for mitral stenosis, 0.93 AUC for moderate or greater aortic regurgitation, and 0.96 AUC for moderate or greater mitral regurgitation. The system also showed precision in estimating continuous echocardiographic parameters, achieving a median normalized mean absolute error of 0.13 across 21 tasks.

PanEcho was developed using 1.23 million echocardiogram videos from nearly 34,000 transthoracic echocardiography tests conducted at Yale-New Haven Health System facilities between 2016 and 2022. The echocardiograms came from 26,067 unique individuals with an average age of 67, and approximately 52% of the imaging data was from adults who self-identified as men. The study population was predominantly white (80%), with 14.2% Black participants, 1.8% Asian participants, and 4.3% from people who self-identified as other races.

While the findings are promising, the research team notes that PanEcho is limited by its retrospective validation using previously acquired data. The next steps involve prospective validation in real-world patient care environments and evaluation for use with portable echocardiogram machines in emergency rooms and smaller medical clinics, where AI tools could have the most significant positive impact. The American Heart Association's scientific statement on the Use of Artificial Intelligence in Improving Outcomes in Heart Disease provides additional context for these developments. Additional resources and the study abstract are available through the Association's scientific sessions materials.