Heart risk, drawn on the heart.
Drop in a cardiology report. CardioScope reads it, estimates coronary artery disease and the stenosis risk of the LAD, LCX and RCA, and colours each artery on a 3D heart. Everything runs on your own computer.
Drag to rotate.
From PDF to a coloured artery in four steps
The whole pipeline runs locally: PDF text extraction, a local language model for reading fields, and ONNX models for the prediction.
Add a report
Drop in a cardiology report as a PDF, or type the values in by hand.
Fields extracted
A local model pulls age, blood pressure, labs, ECG and echo findings into a form you can check.
Four risk scores
Overall CAD plus LAD, LCX and RCA stenosis probabilities from calibrated models.
See why
Top drivers per vessel, a sensitivity range, a patient timeline and a chat for follow-up questions.
What a result looks like
Example output. Bars show the predicted probability of stenosis.
How well it works
ROC-AUC from nested 5-fold cross-validation on 303 patients (Z-Alizadeh Sani dataset). 0.5 is chance.
CAD and LAD are reliable. LCX and RCA are weaker, so the app marks them lower-confidence. Validation is internal, from a single dataset.
Your data stays on your machine
There is no account to create in the cloud and no upload to a server.
Fully offline
After setup, no internet connection or cloud API is needed. The server listens on this computer only.
Encrypted storage
Reports, predictions and chats are saved in an AES-encrypted local database in your home folder.
Open source
Read the code, the models and the training write-up on GitHub before you run anything.
Before you open it
CardioScope uses a local language model through Ollama to read reports and write summaries. Install Ollama once, then pull the model while you are online.
1. Install Ollama
Download it from ollama.com and leave it running.
2. Pull the model
One time, needs internet (several GB).
ollama pull llama3.1:8b3. Run the app
- Download the .exe from the button below and open it.
- Register a user, then upload a report PDF or enter values by hand.
- Click an artery on the heart to see its risk and the main drivers.
Set your own database passwords before the first run if you build from source. The steps are in the GitHub README.