ECG Analysis with MIT-BIH

Level: Hard
Goal: Work with ECG waveforms (e.g., MIT-BIH Arrhythmia).

Library & Data

• WFDB Python: https://wfdb.io/software/python.html
• MIT-BIH Database: Available via WFDB

Installation

pip install wfdb

Data Loading

import wfdb
import pandas as pd
import numpy as np

record = wfdb.rdrecord("100", pn_dir="mitdb")
signal = record.p_signal[:, 0]  # first channel
fs = record.fs

times = pd.to_timedelta(np.arange(len(signal)) / fs, unit="s")
ecg_series = pd.Series(signal, index=times)

Implementation Steps

1. Library Setup and Data Access

• Install WFDB library
• Access MIT-BIH Arrhythmia database
• Explore available records
• Select record(s) for analysis
• Understand WFDB data structure

2. Data Exploration

• Load ECG signals from selected records
• Inspect sampling rate (typically 360 Hz for MIT-BIH)
• Plot raw ECG waveforms
• Identify QRS complexes and other features
• Examine signal quality

3. Data Preprocessing

• Filtering:
  • Apply band-pass filter (e.g., 0.5-40 Hz)
  • Remove baseline wander
  • Remove power line interference (50/60 Hz)
• Artifact Removal:
  • Identify and handle artifacts
  • Use appropriate filtering techniques
• Normalization:
  • Normalize amplitude if needed

4. Feature Extraction

• QRS Detection:
  • Detect R-peaks (heartbeats)
  • Calculate heart rate (HR) time series
  • Extract RR intervals
• Waveform Features:
  • Extract P, Q, R, S, T wave features
  • Calculate morphological features
• Create Time Series:
  • Heart rate over time
  • RR interval series
  • Other derived features

5. Exploratory Data Analysis (EDA)

• Plot ECG waveforms
• Visualize heart rate over time
• Analyze RR interval distributions
• Identify arrhythmias or abnormal patterns
• Perform time series decomposition of HR

6. Stationarity Analysis

• Test extracted features (e.g., HR) for stationarity
• Apply transformations if needed
• Consider segmenting by rhythm type

7. Model Building

• Heart Rate Time Series:
  • Apply ARIMA/SARIMA to HR series
  • Model RR intervals
• Event Detection (advanced):
  • Detect arrhythmias
  • Model event occurrences
• Morphological Analysis (advanced):
  • Analyze waveform changes over time

8. Model Evaluation

• Split data temporally
• Generate forecasts (e.g., future HR)
• Calculate accuracy metrics
• Visualize results
• Compare with clinical expectations

9. Clinical Interpretation

• Interpret results in cardiology context
• Identify normal vs abnormal patterns
• Discuss implications for monitoring
• Compare with ECG literature

Expected Deliverables

1. EDA Report:

   • ECG waveform plots
   • Heart rate analysis
   • RR interval analysis
   • Pattern identification

2. Model Results:

   • Selected models for HR/features
   • Forecast accuracy
   • Visualization of results
   • Clinical interpretation

3. Code:

   • Complete Python notebook
   • WFDB data loading functions
   • QRS detection and feature extraction
   • Visualization utilities

Tips

• ECG data requires specialized preprocessing and analysis
• QRS detection is crucial for extracting meaningful features
• Heart rate time series is more suitable for standard time series methods than raw ECG
• Use appropriate filtering (band-pass, notch) for clean signals
• Understand ECG morphology (P, Q, R, S, T waves) for proper analysis
• Consider different rhythm types (normal sinus, arrhythmias) in analysis
• RR intervals can be modeled as time series
• Document all preprocessing steps clearly
• Use WFDB's built-in functions for reading and processing
• Consult cardiology literature for appropriate analysis methods
• Consider event-based analysis (arrhythmia detection) in addition to continuous forecasting
• High-frequency raw ECG may need downsampling or feature extraction for standard methods