Skip to content

README.md

Latest commit

 

History

History
88 lines (59 loc) · 2.36 KB

README.md

File metadata and controls

88 lines (59 loc) · 2.36 KB
Raw

ECG Arrhythmia Detection using Deep Learning

Project Overview

This project implements advanced deep learning techniques (LSTM, CNN and AlexNet) for detecting heart arrhythmias using the St. Petersburg INCART 12-lead Arrhythmia Database. The goal is to develop robust machine learning models that can accurately identify abnormal heart rhythms from ECG signals.

Dataset

  • Source: St. Petersburg INCART 12-lead Arrhythmia Database
  • Characteristics:
    • 75 subjects with 30-minute ECG recordings
    • Sampled at 257 Hz
    • Annotated by medical experts for arrhythmia detection

Key Features

Data Preprocessing

  • Custom data extraction function to segment ECG signals
  • Feature extraction using Short-Time Fourier Transform (STFT)
  • Handling class imbalance with custom loss function

Model Architectures

The project implements three different neural network architectures:

  1. Bidirectional LSTM

    • Captures temporal dependencies in ECG signals
    • Processes sequential data with bidirectional context

  2. 1D Convolutional Neural Network (CNN)

    • Extracts spatial patterns from ECG signals
    • Uses convolutional and pooling layers for feature detection

  3. AlexNet-Inspired Architecture

    • Adapted from the original AlexNet image classification model
    • Customized for 1D ECG signal processing

Key Techniques

  • Custom Binary Cross-Entropy (BCE) loss function
  • Handling class imbalance
  • Data visualization of ECG signals
  • Performance evaluation using accuracy and F1 score

Dependencies

  • Python 3.x
  • PyTorch
  • NumPy
  • Pandas
  • Matplotlib
  • SciPy
  • Scikit-learn
  • WFDB (Waveform Database Software Package)

Results

The project compares the performance of different neural network architectures:

  • Bidirectional LSTM
  • 1D Convolutional Neural Network
  • AlexNet-inspired Architecture

Each model is evaluated using:

  • Training accuracy
  • Test accuracy
  • F1 score

Key Insights

  • The dataset shows a significant class imbalance (only 13.3% abnormal samples)
  • Custom loss function helps address class imbalance
  • CNN outperformed LSTM in this specific classification task

Future Work

  • Experiment with more advanced architectures
  • Incorporate additional feature extraction techniques
  • Collect more diverse ECG data

Acknowledgments

  • St. Petersburg INCART for providing the ECG database
  • PyTorch and scientific computing community