Computer Vision News - February 2022

31 Automated Echocardiographic Detection of ... variability among doctors, facilitate their job, and significantly impact the clinical workflow and patient care, providing an earlier and improved detection. The authors had previously worked on an AI pipeline that can automatically process and extract novel imaging features from stress echocardiograms. In this work, they incorporate the features into a machine learning model trained to identify patients with significant coronary artery disease. In the image on the left, we can have a peak through the vision of Ultromics: a cardiologist can provide a patient’s diagnosis based on both her expertise and a software powered by AI. This is truthfully a vision shared with most of the medical AI research out there. But for Ultromics and their team, the clinical translation of the developed AI models is not just a dream anymore: it is a successfully achieved goal. Data This study leverages on a dataset of stress echocardiography (SE) images which were provided by a multicenter study (EVAREST) and include images from different operators and US vendors. This modality is a valid non-invasive, non-ionizing and low-cost alternative imaging test to diagnose CAD, and it relies on the manual observation by clinicians who identify regional wall motion abnormalities on the echocardiogram “by eye”, comparing pre- and post-induced stress images. In fact, in areas affected by flow-limiting CAD, the contractility of the myocardium reduces as heart rate rises with stress (induced by exercise or medications). A set of images with apical 4 chamber (A4C), apical 2 chamber (A2C), parasternal short-axis mid- ventricular (SAX) views are included in the study, with sufficient frames between end-systole and end-diastole, from patients with no previous history of coronary artery bypass or other surgery. Ground truth answers for severity of disease were based on decisions made by a committee with at least 1 cardiologist, blinded to the SE, who inspected the results of invasive coronary angiography (ICA).