A cardiologist-like computer-aided interpretation framework to improve arrhythmia diagnosis from imbalanced training...

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Signal Processing

Background:

• Arrhythmias pose significant risks, including stroke and sudden death.
• Computer-aided electrocardiogram (ECG) systems may misdiagnose vulnerable arrhythmias due to aggressive types.
• Accurate arrhythmia detection is crucial for timely intervention and patient outcomes.

Purpose of the Study:

• To develop an improved method for arrhythmia diagnosis in computer-aided ECG interpretation.
• To address the underdiagnosis of vulnerable arrhythmias caused by aggressive types.
• To enhance the reliability and interpretability of automated arrhythmia detection systems.

Main Methods:

• A novel method combining morphological-characteristic-based waveform clustering with Bayesian theory.
• Inspiration drawn from the diagnostic reasoning processes of experienced cardiologists.
• Hyperparameter optimization, including spliced heartbeats and clusters, for optimal performance.

Main Results:

• Achieved optimal performance in macro-recall and macro-precision through hyperparameter optimization.
• Demonstrated the highest average recall and lowest average drop in recall for nine vulnerable arrhythmias, even with increasing aggressive arrhythmia presence.
• Identified maximum cluster characteristics consistent with established clinical arrhythmia diagnostic criteria, ensuring interpretability.

Conclusions:

• The proposed method effectively enhances the diagnosis of vulnerable arrhythmias in ECG interpretation.
• The approach offers improved accuracy and robustness against confounding aggressive arrhythmias.
• The method's interpretability aligns with clinical diagnostic standards, facilitating trust and adoption.