Apppublic

Dumpers/ecg-image-classifier

sourceHugging Faceupdated 4mo agoView on Hugging Face

0likes

App README

ECG Image-Based Arrhythmia Classifier

Classifies ECG images into three categories using MobileNetV2, and shows which parts of the image influenced the prediction using Grad-CAM.

Classes

—Abnormal Heartbeat
—History of Myocardial Infarction
—Normal Person

Results

Four pretrained CNN models were compared using 5-fold stratified cross-validation on a dataset of 707 ECG images.

Model	Test Accuracy
ResNet18	98.13%
MobileNetV2	99.07%
EfficientNetB0	94.39%
VGG19-BN	97.20%

MobileNetV2 performed best and is used in the demo.

Running locally

Install dependencies:

pip install -r requirements.txt

Train the model and save weights:

python train_and_save.py

Launch the app:

python app.py

Dataset

The dataset is not included in this repository due to size. Place the ECG Dataset 2 folder in the project root before training. The folder should contain three subfolders matching the class names above.