README.md

# ECG MCTnet — 5-class ECG Arrhythmia Classifier

## Model
MCTnet: Dual-branch Conv + LTCAR Attention + VIB + Label Decoupling Head

## Classes
- Normal
- AFIB (Atrial Fibrillation)  
- AFL (Atrial Flutter)
- LBBB (Left Bundle Branch Block)
- RBBB (Right Bundle Branch Block)

## Performance
- Val Macro-F1: 0.8198
- Dataset: TEKNOFEST ECG 2026

## Input
- Shape: (batch, 12, 5000)
- 12-lead ECG @ 500Hz, 10 seconds

## Usage
```python
import torch
from huggingface_hub import hf_hub_download

# Model indir
weights = torch.load(hf_hub_download(repo_id="SarmAI/EKG-mctnet", filename="model.pt"))
model = MCTnet(MCTnetConfig())
model.load_state_dict(weights)
model.eval()
```