Cascade Classification with Adaptive Feature Extraction for Arrhythmia Detection.

Area of Science:

• Biomedical Engineering
• Computer Science
• Cardiology

Background:

• Arrhythmia detection from electrocardiogram (ECG) data is increasingly important for mobile health applications.
• Resource constraints on mobile devices necessitate a trade-off between computational efficiency and diagnostic accuracy.

Purpose of the Study:

• To develop an adaptive feature extraction strategy for arrhythmia detection on mobile devices.
• To enhance classification accuracy by optimizing feature selection based on available computational resources.

Main Methods:

• An adaptive feature extraction strategy was proposed, utilizing normalized beat morphology features in resource-constrained environments.
• A wider range of ECG features were incorporated in high-performance environments.
• A cascaded random forest classifier was employed to augment the feature extraction process.

Main Results:

• Experiments were conducted using data from the MIT-BIH Arrhythmia Database.
• Classification accuracies ranged from 96.59% to 98.51%.
• The achieved accuracies are comparable to state-of-the-art arrhythmia detection methods.

Conclusions:

• The proposed adaptive strategy effectively balances computational efficiency and accuracy for mobile arrhythmia detection.
• The cascaded random forest classifier enhances the performance of the adaptive feature extraction.
• This approach offers a viable solution for accurate arrhythmia detection in mobile health settings.