HeartLens is a collaborative project aiming to develop an AI-based tool to support scalable cardiovascular screening in routine chest imaging, assisting radiologists in reviewing CT scans, detecting coronary artery calcium (CAC), and informing diagnostic and treatment decisions.
CAC scoring helps assess cardiovascular disease severity but relies on electrocardiogram-gated CT scans, restricting its use to specialized settings. Early detection of calcification enables more timely, less invasive interventions.
The CARD-ViT model highlights calcified areas in coronary arteries, color-coding them by severity to aid clinical review. Beyond CAC scoring, the tool has potential applications for detecting other heart conditions including stenosis.
Long-term goal: Integration into electronic health records for point-of-care clinical decision-making.
The approach utilizes a self-supervised learning framework based on DINOv2 with a register-based approach. PCA analysis on four register maps produces feature maps that highlight and precisely locate calcified areas.
Training data: Public Stanford CT datasets + internal CT data. Uses proprietary Label-Guided data augmentation technique. Designed for single-channel medical images.
HeartLens is still in development. Next steps include robust clinical evaluation.
Funded by the EXCEL Research Initiative. Uses the Vision Foundry platform and NVIDIA DGX infrastructure.