Automatic Aortic Aneurysm Screening using Deep-Learning Models

Y. Li, H. Nguyen
Rowan University,
United States

Keywords: CT scan, deep learning, aortic aneurysm, size measurement


An aortic aneurysm is a bulge that occurs in the aorta wall, which carries blood from the heart to the rest of the body. Both ascending and descending thoracic aortic aneurysms (TAAs) is a life-threatening disease, specifically due to the high risk of mortality if rupture occurs and repair is performed in an emergency setting. According to the Centers for Disease Control and Prevention (CDC), aortic aneurysms were the primary cause of more than 9,000 deaths in 2019. The difficulty with identifying this pathology is that aneurysms are asymptomatic until adverse effects occur, either dissection or rupture, both of which may result in death. Given the high rate of morbidity and mortality associated with aortic aneurysms, accurate diagnosis and preoperative evaluation are essential for improved patient outcomes. Our team at Rowan University are developing a new deep learning pipeline to evaluate and analyze CT scans of patients to detect and measure aortic aneurysms without the presence of a radiologist. Our method utilizes FasterRCNN, YOLO, and U-Net models to read through DICOM files from patients' chest CT scans and provide fast and accurate screening of both the ascending and descending TAAs, and provide gradient adjusted aorta size measurement. Further, benefiting from patients' electronic medical record (EMR), the pipeline will estimate the natural history risk of adverse events. The method will reduce time in reading and evaluating CT scans by radiologists, and help optimize the triage process for treatment to improve patient health outcomes.