README.md

---
title: ECG to ECHO Screening AI
emoji: 🫀
colorFrom: blue
colorTo: indigo
sdk: streamlit
sdk_version: 1.39.0
python_version: "3.10"
app_file: app.py
pinned: false
license: cc-by-nc-4.0
---

# 🫀 ECG → ECHO Clinical Screening AI

AI-powered screening tool that predicts **Left Ventricular Dysfunction** and **Regional Wall Motion Abnormalities** from a single 12-lead ECG image. Designed for rural cardiac screening in resource-limited settings.

## How it works

A 5-fold ensemble of EfficientNet-B3 (NoisyStudent pretrained) models analyzes the ECG image and outputs:

1. **EF Risk** — likelihood of reduced ejection fraction (< 55%)
2. **RWMA Risk** — likelihood of regional wall motion abnormality

## Training

- **Dataset:** 500 paired ECG-ECHO records from cardiology clinic in Karnataka, India
- **Validation:** 5-fold Multilabel Stratified K-Fold cross-validation
- **Backbone:** EfficientNet-B3 with NoisyStudent pretraining

## ⚠️ Disclaimer

This is a **research prototype** for screening purposes only. NOT approved for clinical diagnostic use. All predictions must be reviewed by a qualified cardiologist.