15-B AI tools to characterize and identify unstable atherosclerotic plaques (ATLANTIS)



Strokes and myocardial infarctions (MIs), caused by the rupture of unstable atherosclerotic plaques in the carotid and coronary arteries, are the leading causes of death world-wide. Current clinical guidelines recommend surgical interventions based solely on the degree of artery stenosis. However, stenosis alone is an incomplete determinant of stroke/MI risk, leading to suboptimal medical decisions or inappropriate treatment allocation. Although plaque morphology and composition is a more accurate indicator of plaque instability and a better predictor of clinical outcomes, no method currently exists for its accurate, quantitative, and noninvasive characterization.


We aim to develop quantitative, artificial intelligence-based platforms that utilize machine-learning algorithms to characterize features of plaque morphology and composition from 1) histological (HistoAI) and 2) ultrasound plaque images (UltrasoundAI).


Data are available through our existing biobank (~600 patients). Preoperative 2D- and 3D-ultrasound plaque images have been obtained from patients who underwent surgical intervention for plaque removal. Excised plaque specimens from the same patient will be stained for histological analysis of various features of plaque stability/instability, and annotated. Histology and ultrasound plaque images will be used to develop machine-learning algorithms to i) quantify plaque features of stability/instability and ii) determine the overall score of plaque instability. HistoAI will be based solely on histological images, while UltrasoundAI, during the development period, will be based on corresponding paired histological and ultrasonic images from the same patient. HistoAI will help train UltrasoundAI to allow for accurate annotation and artificial intelligencebased analysis that reflect the gold-standard histological features of the plaque stability/instability continuum. Once developed, UltrasoundAI will be used in clinical practice without the need for histology images. Impact: While UltrasoundAI will be informed by HistoAI, ultimately the two platforms will function independently, and will target different markets. These platforms will provide a comprehensive and accurate analysis of plaque features and instability, while eliminating bias and inter-individual variability. These innovations will be used to improve the prediction, treatment, and prevention of strokes and MIs.