The most acute type of heart attack is the ST segment elevation myocardial infarction (STEMI), requiring emergency care to clear a blocked coronary artery. Takotsubo syndrome (TTS) often mimics the clinical and imaging characteristics of STEMI on presentation. Although highly specialized cardiovascular imaging can provide instant details of cardiovascular dysfunction, interpretation of the finer points of imaging, especially echocardiography, is still often a “judgement call.” The wrong “call” can not only cost precious minutes in an emergency setting but lead to potentially harmful treatment.

Kan Liu, MD, PhD, clinical professor in Cardiovascular Medicine, and his team drew on their experience with deep learning (DL) neural networks to determine whether an artificial intelligence (AI) program could help reduce errors in the immediate diagnosis of these two very different cardiomyopathies. Their findings were recently published in the Lancet’s EClinicalMedicine.

Liu and his team collaborated with the UI Department of Electrical and Electronic Engineering and researchers across the United States to create DL neural networks to distinguish TTS from STEMI. Using a database of more than 17,000 echocardiographic images and videos to train and validate DL models, they found that two spatio-temporal hybrid DL neural networks consistently outperformed human specialists in correctly diagnosing TTS. The team concluded that human diagnoses are biased due to limited personal experience with rare cardiomyopathies.”The eye sees what it expects to see,” Liu said. “DL neural networks are more objective and efficient at identifying delicate features, pixel by pixel, that might be overlooked by even highly experienced clinicians.”

Liu and his colleagues believe that real-time DL neural network models like these have “great potential to increase clinical relevance.” He said, “There are many possible applications of AI for urgently needed triage and management decisions in acute cardiovascular disorders.”