Supervised SVM Transfer Learning for Modality-Specific Artefact Detection in ECG.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Electrocardiograms (ECG) are crucial for diagnosing heart conditions.
• Capacitively coupled ECG (ccECG) offers remote monitoring but introduces more artefacts.
• Artefact detection is essential for reliable ccECG interpretation.

Purpose of the Study:

• To apply transfer learning to adapt artefact detection models from contact ECG to ccECG.
• To reduce the need for costly ground truth data in ccECG artefact detection.
• To improve the performance of ccECG artefact detection using pre-existing knowledge.

Main Methods:

• Utilized transfer learning to optimize an artefact detection model.
• Trained the model on existing contact ECG datasets.
• Tested the adapted model on a ccECG dataset with varying recording devices.
• Evaluated the impact of limited ccECG data on model performance.

Main Results:

• Transfer learning improved the accuracy of contact-ECG classifiers on ccECG data by 5-8%.
• A small subset of 20 ccECG segments was sufficient to significantly boost classifier accuracy.
• Demonstrated the effectiveness of knowledge transfer between different ECG recording modalities.

Conclusions:

• Transfer learning is a viable and efficient strategy for ccECG artefact detection.
• Minimizing labelled data requirements for ccECG analysis is achievable.
• This approach facilitates the development of robust remote cardiac monitoring systems.