HeartScope

AI-powered angiogram analysis that spots dangerous coronary blockages doctors might miss. Like a powerful magnifying glass for your cath lab, helping save lives by catching subtle dangers in seconds.

Comment

Inspiration

In the field of cardiology, accurately detecting coronary artery disease through X-ray angiography is a critical yet challenging task. Even experienced physicians can miss subtle signs of blockages due to the complex nature of these images. Having witnessed firsthand how AI has transformed various medical fields, we recognized an opportunity to enhance diagnostic accuracy and potentially save lives by creating an AI assistant for cardiologists.

What it does

HeartScope serves as an AI diagnostic assistant that enhances cardiologists' capabilities in detecting coronary artery disease:

  • Automatically analyzes X-ray angiography images to detect areas of stenosis
  • Highlights regions with potential plaque buildup that might be difficult for human eyes to detect
  • Provides precise marking of affected areas to assist in diagnostic decision-making
  • Generates detailed reports of detected abnormalities by taking into consideration patient history
  • Offers real-time analysis during procedures to support immediate clinical decisions
  • Maintains a database of analyses for tracking disease progression over time

How we built it

Core Detection System:

  • Trained a YOLOv11 model for coronary artery detection, specifically utilizing its segmentation architecture to create precise masks around blocked regions rather than simple bounding boxes
  • Developed using a comprehensive dataset of X-ray angiography images
  • Fine-tuned the model to specifically identify stenosis from plaque build up and calcification

Image Processing Pipeline:

  • Created custom preprocessing algorithms to enhance image quality
  • Implemented automatic region-based segmentation for precise arterial analysis
  • Developed post-processing annotation scripts to highlight detected abnormalities

Clinical Integration:

  • Built a user-friendly interface designed for clinical settings
  • Securely store relevant data

Challenges we ran into

Training the YOLO model on medical imaging data presented unique challenges. Medical images often have subtle features that are crucial for diagnosis but challenging for AI to detect. We had to carefully balance model sensitivity to avoid false positives while ensuring no critical blockages went undetected.

Accomplishments that we're proud of

We successfully created a system that achieves:

  • High accuracy in detecting stenosis
  • Real-time analysis capabilities suitable for clinical settings
  • Seamless integration for ease of use

What we learned

Training computer visions models is not a one size fits all.

What's next for HeartScope

The future of HeartScope lies in expanding the training dataset to improve model performance, while implementing additional detection capabilities for other cardiovascular conditions

Built With

Share this project:

Updates