CardioScope

• 

• 

• 

• 

  after given values result shown with an iterative 3D model

• 

• 

  no values given

• 

Inspiration

I found this hackathon while browsing Unstop and was drawn to its topics and real-world focus. I chose to explore how machine learning could make clinical data easier to understand through CardioScope.

How I built it

I built a web dashboard that accepts clinical measurements and displays model estimates for CAD and three coronary vessels: LAD, LCX, and RCA. It also includes an interactive 3D heart visualization and a list of input features that contributed to the CAD estimate. GitHub Copilot helped prepare the code, and Claude helped edit the schematic artery overlays.

What I learned

I learned how to connect a machine-learning workflow to a web application, preprocess clinical data, evaluate classification models, and present model outputs in an interactive interface. I also learned to communicate limitations clearly: CardioScope is an educational prototype, and its estimates are not a diagnosis.

Challenges

One challenge was connecting the prediction service, dashboard, and 3D viewer into one working experience. I also had to prevent target leakage during model training and make sure the 3D arteries were clearly presented as schematic overlays rather than patient-specific anatomy.

Built With

• css
• fastapi
• github-copilot
• html
• javascript
• pandas
• python
• scikit-learn
• three.js-/-react-three-fiber