ECG Arrythmia Detection

ECG is a biosignal that provides insight on heart's electrical activity and it's rythm. Detecting Arrythmia from ECG signals require medical expertise which is not accessible to all. A machine learning model paired with a ECG Holter can help address this issue.

This project utilizes MIT-BIH dataset to train different machine learning models (LR, SVM and CNN) and find the optimum model for arrythmia detection. Based on the R-peak of ECG signal, the models classify the ECG signals into seven categories:

• Normal
• Arrhythmias
  • Left Bundle Branch Block
  • Right Bundle Branch Block
  • Premature Ventricular Contraction
  • Ventricular Escape Rhythm
  • Paced Rhythm
  • Atrial Premature Contraction

Dataset

The MIT-BIH Arrhythmia Dataset was the first set of standard test material that was generally available to evaluate arrhythmia detection. ECG signals were described by a text header file (.hea), a binary file (.dat) and a binary annotation file (.atr). The source of the ECGs included in the MIT-BIH Arrhythmia Database is a set of over 4000 long-term Holter recordings. In most records, the upper signal is a modified limb lead II (MLII), obtained by placing the electrodes on the chest.

Progress

• Data Extraction
• R-Peak Detection
• ECG Signal Segmentation
• Resample
• Standardize
• Train Model (LR, SVM, 1D-CNN)
• Test Model (Precision, Recall, Accuracy, Loss, Confusion Matrix)
• Web Deployment

CNN Architecture

Model performance of CNN

	Paced	APC	VEB	LBB	Normal	RBB	PVC	Total
Training data	16200	16200	16200	16200	16200	16200	16200	113400
Validation data	1800	1800	1800	1800	1800	1800	1800	12600
Correct Prediction	16194	16170	16200	16189	16140	16191	16165	113249
Sensitivity	99.9%	99.8%	100%	99.9%	99.6%	99.9%	99.7%	99.8%

Training and validation gain/loss of the CNN model

The training performance gradually improved after each iteration and remained stable. After 25 iterations or epochs, an averaged accuracy of 99.8% was achieved in the validation set for the arrhythmia classification. The loss also reduced after each iterations and an average loss of 0.5% was achieved.

Tech Stack

• DSP (bioSPPy)
• sklearn
• Matplotlib
• Pandas
• Numpy
• wfdb
• Keras/ tensorflow