Automatic Detection for Multi-Labeled Cardiac Arrhythmia Based on Frame Blocking Preprocessing and Residual Networks.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Electrocardiograms (ECG) are crucial for diagnosing heart conditions like arrhythmias.
• Existing neural network algorithms for arrhythmia detection require performance enhancement.
• Accurate multi-type cardiac state classification from ECG remains a challenge.

Purpose of the Study:

• To develop an advanced auto-detection algorithm for classifying multiple cardiac states using 12-lead ECG.
• To improve the accuracy and reliability of automated cardiac arrhythmia detection.
• To extract and utilize valid features from ECG signals for enhanced diagnostic capabilities.

Main Methods:

• A preprocessing component using frame blocking to standardize ECG recording lengths.
• A binary classifier integrating ResNet with an attention-based bidirectional long-short term memory (BiLSTM) model.
• Multi-label classification training and testing on ECG data encompassing nine cardiac states.

Main Results:

• The algorithm achieved a high performance on multi-label classification of cardiac states.
• An averaged F1-score of 0.908 was obtained.
• An averaged area under the curve (AUC) of 0.974 was achieved, demonstrating strong predictive power.

Conclusions:

• The proposed frame blocking and BiLSTM-based algorithm shows significant improvement over existing methods.
• This approach enhances the auto-detection and classification of diverse cardiac abnormalities from ECG.
• The findings suggest a promising direction for automated cardiac diagnostics.