Computerised electrocardiology employing bi-group neural networks

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Signal Processing

Background:

• 12-lead electrocardiograms (ECGs) are crucial for diagnosing cardiac conditions.
• Accurate interpretation of ECGs remains a challenge, necessitating advanced computational methods.
• Existing classification techniques may not fully capture the complexities of ECG data.

Purpose of the Study:

• To propose and evaluate a novel framework combining bi-group neural networks (BGNNs) with evidential reasoning for ECG interpretation.
• To investigate the impact of pre-processing feature selection techniques on BGNN performance.
• To compare the proposed framework's classification accuracy against conventional methods.

Main Methods:

• A configuration of bi-group neural networks (BGNNs) was developed.
• Evidential reasoning was integrated with BGNN outputs for classification.
• Various pre-processing feature selection techniques were applied to the input data.
• Network outputs were discounted within a belief interval based on test data performance.
• The framework was compared with multi-output neural networks and linear multiple regression.

Main Results:

• Feature selection techniques significantly enhanced individual BGNN performance.
• The proposed BGNN framework with evidential reasoning achieved higher classification accuracy.
• The framework attained 70.4% classification accuracy, outperforming multi-output neural networks (66.7%) and linear multiple regression (66.7%).

Conclusions:

• The integration of feature selection and evidential reasoning with BGNNs offers a superior approach for 12-lead ECG interpretation.
• The proposed framework demonstrates enhanced diagnostic capability compared to conventional classification techniques.
• This approach holds promise for improving automated cardiac condition diagnosis through ECG analysis.