Cardioish: Lead-Based Feature Extraction for ECG Signals

Area of Science:

• Cardiology and Artificial Intelligence
• Biomedical Signal Processing
• Explainable Artificial Intelligence (XAI)

Background:

• Electrocardiography (ECG) is crucial for diagnosing cardiac disorders, with 12-lead ECGs being the standard.
• Existing methods often lack explainability, hindering clinical interpretation.
• A novel symbolic language, Cardioish, is introduced for enhanced ECG analysis.

Purpose of the Study:

• To develop a new feature engineering model for ECG signals.
• To achieve high classification accuracy and provide explainable results.
• To introduce a symbolic language (Cardioish) for interpretable ECG feature extraction.

Main Methods:

• A Cardioish-based Explainable Feature Engineering (XFE) model was developed.
• The model involves lead transformation, transition table feature extraction (144 features), Iterative Neighborhood Component Analysis (INCA) for feature selection, and k-nearest neighbors (kNN) classification.
• Explainable Artificial Intelligence (XAI) is achieved through Cardioish symbol generation and sentence analysis.

Main Results:

• The Cardioish-based XFE model achieved over 99% classification accuracy on two public datasets (mental disorder and myocardial infarction).
• The model successfully generated explainable results (XAI) for the classified cardiac disorders.
• The generated Cardioish sentences provide interpretable insights into the ECG signals.

Conclusions:

• The Cardioish-based XFE model demonstrates high performance in ECG classification accuracy.
• The model offers a significant advancement in providing explainable results for ECG interpretation.
• This approach offers a novel pathway for improving ECG classification and clinical understanding.