Exploring deep learning for carotid artery plaque segmentation: atherosclerosis to cardiovascular risk biomarkers

Atherosclerosis, caused by a variety of extrinsic risk factors, is the major cause of the cardiovascular and cerebrovascular diseases that bring high mortality and morbidity globally, with substantial socio-economic impact. The development and progression of atherosclerosis is insidious, and early detection and intervention is vital to prevent debilitating cardiovascular events like myocardial infarction and stroke. Imaging biomarkers like carotid intima-media thickness (cIMT) and plaque area/burden that can identify subclinical disease and stratify risk are therefore crucial. This comprehensive review focuses on these biomarkers, their imaging using ultrasound, assessment using machine learning and deep learning techniques, and their association with cardiovascular risk factors. Measurement of cIMT and plaque area/burden in the common carotid artery (CCA) using deep learning models built on convolutional neural networks like UNet have shown good accuracy and reliability compared to manual methods. Expanding the scope to include the internal carotid artery (ICA) and carotid bulb presents greater technical challenges due to image acquisition difficulties, but deep learning methods utilizing architectures like attention-UNet show promise. However, several biases exist in current deep learning systems stemming from limited multi-center datasets, as models trained on specific patient cohorts underperform when assessed on diverse unseen test data. Eliminating these biases through techniques like transfer learning, aggregating multi-ethnic data, comparing multiple models, and combining deep learning with optimization algorithms can make these AI systems more generalizable and robust. Moving forward, automated measurement of carotid ultrasound imaging biomarkers of subclinical atherosclerosis using bias-free deep learning approaches can enable large-scale screening to identify individuals at risk of cardiovascular events, allowing early intervention to modify risk and prevent disease progression.