Patient-specific Outcome Prediction from Cardiovascular Multimodality Imaging by Artificial Intelligence

点击翻译按钮获取中文摘要

PROJECT SUMMARY Coronary artery disease (CAD) remains a major public health concern with a high prevalence in the US population. Functional, molecular, and structural imaging offer a unique opportunity to understand the pathophysiology of CAD, especially in high-risk groups such as patients with obesity, diabetes, and chronic kidney disease (cardiometabolic disease). CAD evaluation by imaging is based on modalities that assess (1) myocardial ischemia and myocardial blood flow (2) anatomic burden of atherosclerosis, and (3) disease activity using novel techniques. However, physicians are not yet able to use these data optimally to identify patients at highest risk of adverse events—due to technical complexity of advanced multivariable data, and lack of automation and integrative tools. While positron emission tomography (PET) can measure myocardial blood flow, and depict high-risk plaque in the arteries and CT can reliably detect coronary artery calcium —an unequivocal marker for atherosclerotic disease– physicians are not able to combine these data effectively to identify patients at highest risk of adverse events, due to complexity and lack of automation. Critically, there is an unmet need for efficient integration of diverse imaging and clinical data by a robust, automated clinical tool after non-invasive imaging. Highly efficient artificial intelligence (AI) methods are revolutionizing image analysis and could improve CAD detection and management. The overall vision for the research program is to further the clinical utility of PET/CT in detecting high-risk CAD and guiding subsequent management by automation and integrating all image and clinical data with state-of-the-art AI. We will establish a large multicenter PET and CT imaging registry and with image-based AI, automate analysis and quality control for robust analysis even at less experienced centers, and develop decision support tools utilizing collectively all available PET/CT images and clinical information (beyond what is possible by subjective visual analysis and mental integration). We will develop direct interpretation of images by AI, and patient-specific explanation of the AI findings to the physician. Precise quantitative results will be presented to clinicians (and patients) in easy to understand terms (e.g., % risk per year or as the relative risk of one therapy compared to the alternative) for a specific patient. This work will allow accurate identification of patients with high-risk disease who can benefit treatment from advanced therapies and enable precise patient-specific risk estimates and treatment recommendations in challenging clinical scenarios—in CAD with cardiometabolic disease and advanced high- risk disease.