pratyaksh10 / ecg_cnn

An Optimal Deep Learning Framework for Detecting Abnormal Heart Beats Using ECG Signals

An Optimal Deep Learning Framework for Detecting Abnormal Heartbeats Using ECG Signals

The following project detects abnormal heart beats using Electrocardiogram (ECG) signals. A Convolutional Neural Network is used to predict if the given heartbeat has arrhythmia using a 6 second window.

This project was inspired from Andrew Ng’s team’s work on heart arrhythmia detector. For detailed information please refer to https://stanfordmlgroup.github.io/projects/ecg/

FILES

• load_data : This folder contains a python file that extracts ECG signals, labels, and annotations from the dataset and processes it in order to feed it into the CNN model.

• model_cnn.py : Code that trains a CNN model that is used to predict if a given heartbeat has arrhythmia.

Dependensies

• Tensorflow

• Keras

DATASET

We will use the MIH-BIH Arrythmia dataset from https://physionet.org/content/mitdb/1.0.0/ which is made available under the ODC Attribution License. The dataset consists of 48 half-hour two-channel ECG recordings which is measured at a frequency of 360Hz.

MODEL

We have used a 1-D Convolutional Neural Network (CNN). A CNN is a deep learning model that uses kernals (or filters) and convolutional operators to reduce the number of parameters. Our model uses Dropout to reduce overfitting of the dataset.

RESULTS

Our model achieved a training accuracy of 98.2% and a testing accuracy of 87%.