ECGPredict Application

A web application for AI-powered 12-lead ECG diagnosis using a trained ResNet1D model on the PTB-XL dataset.

https://imhnor-cardiopulse-ai.hf.space/

---

Quick Start

1. Start the backend server

# From repo root — uses the ecg_env virtualenv automatically
.\ECG - application\start.ps1

Or manually:

cd "ECG - application\backend"
..\..ecg_env\Scripts\python.exe -m uvicorn app:app --reload --port 8000

2. Open the frontend

Open in browser: http://localhost:8000/app

Or open directly: ECG - application/frontend/index.html

If opening the HTML file directly (without the server), edit app.js line 5: const API_BASE = 'http://localhost:8000'; — this already works for local use.

---

Project Structure

ECG - application/
├── start.ps1               ← One-click startup script
│
├── backend/
│   ├── app.py              ← FastAPI server (endpoints: /upload, /predict, /health)
│   ├── upload_handler.py   ← Session-based temp file storage
│   ├── format_detector.py  ← Auto-detect WFDB / DICOM / XML / SCP-ECG
│   ├── ecg_extractor.py    ← Per-format signal extraction
│   ├── preprocessing.py    ← Validate → resample → normalize → tensor
│   ├── inference.py        ← ResNet1D definition + model load + predict
│   └── requirements.txt
│
└── frontend/
    ├── index.html          ← Single-page medical dashboard
    ├── style.css           ← Premium medical UI theme
    └── app.js              ← Upload + API calls + result rendering

---

API Endpoints

Method	Path	Description
GET	/api/health	Server health + model status
GET	/api/model-info	Model architecture metadata
POST	/api/upload	Upload ECG file(s) → returns session_id + signal info
POST	/api/predict	Run inference on session → returns probabilities
POST	/api/upload-and-predict	Combined single-step endpoint

Upload Request

POST /api/upload
Content-Type: multipart/form-data

files: <ecg_file>            # For WFDB: both .hea and .dat files

Upload Response

{
  "success": true,
  "session_id": "uuid-...",
  "format": "WFDB",
  "format_display": "WFDB (PhysioNet)",
  "lead_names": ["I","II","III","aVR","aVL","aVF","V1","V2","V3","V4","V5","V6"],
  "n_leads": 12,
  "fs": 100.0,
  "n_samples": 1000,
  "duration_s": 10.0,
  "patient_info": {}
}

Predict Request

POST /api/predict
Content-Type: multipart/form-data

session_id: <uuid from upload>

Predict Response

{
  "success": true,
  "probabilities": { "NORM": 0.92, "MI": 0.05, "STTC": 0.03, "CD": 0.02, "HYP": 0.01 },
  "predictions":   { "NORM": true, "MI": false, "STTC": false, "CD": false, "HYP": false },
  "top_diagnosis": "Normal ECG",
  "top_code": "NORM",
  "top_confidence": 0.92,
  "is_normal": true,
  "labels_info": [ ... ],
  "preprocessing": {
    "original_fs": 100,
    "target_fs": 100,
    "resampled": false,
    "padded": false,
    "trimmed": false,
    "lead_order": ["I","II","III","aVR","aVL","aVF","V1","V2","V3","V4","V5","V6"]
  }
}

---

Supported ECG Formats

Format	Extension(s)	Notes
WFDB	.hea + .dat	Upload both files together
DICOM	.dcm	Standard DICOM waveform
XML	.xml	HL7 aECG, Philips, GE, Schiller
SCP-ECG	.scp	EN 1064 binary format

---

Model Details

Property	Value
Architecture	ResNet1D (4-stage, ~2M params)
Training Data	PTB-XL (17,221 records)
Input Shape	(1, 12, 1000)
Sampling Rate	100 Hz
Lead Order	I, II, III, aVR, aVL, aVF, V1–V6
Normalisation	Per-lead z-score
Output	5-class sigmoid (multi-label)
Classes	NORM, MI, STTC, CD, HYP
AUROC	0.92

---

Error Handling

The application detects and displays clear error messages for:

• Unsupported format — file type not recognised
• Missing leads — fewer than 12 canonical leads found
• Invalid sampling rate — below 50 Hz or above 10,000 Hz
• Corrupted file — NaN/Inf values or unreadable binary
• Model not found — models/resnet1d.pth missing

---

Requirements

Python packages (auto-installed by start.ps1):

• fastapi, uvicorn
• torch, numpy, scipy
• wfdb, pydicom
• pandas