MLFusion: Multilevel Data Fusion using CNNs for atrial fibrillation detection.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Data fusion enhances accuracy in instrumentation by integrating multi-sensor signals.
• Traditional methods require manual selection of fusion levels, limiting adaptability.
• Accurate atrial fibrillation detection is crucial for cardiovascular health management.

Purpose of the Study:

• To develop a novel, automated data fusion methodology for multi-sensor multimodal data.
• To integrate feature extraction and fusion into a unified framework using CNNs.
• To incorporate signal quality indicators (SQIs) for quality-aware fusion.

Main Methods:

• A novel fusion methodology using convolutional neural networks (CNNs) for automated optimal fusion level selection.
• Integration of feature extraction and fusion into a unified framework.
• Inclusion of signal quality indicators (SQIs) as input streams for quality-aware fusion.

Main Results:

• The proposed fusion network achieved 99.33% accuracy and 99.74% sensitivity for atrial fibrillation detection using ECG and PPG signals.
• The model demonstrated robustness against noisy inputs, highlighting the effectiveness of SQI-based multi-level fusion.
• The data-driven approach automatically determined the optimal level of information abstraction for fusion.

Conclusions:

• The novel methodology offers a fully automated, quality-aware approach to multi-sensor multimodal signal fusion.
• This self-learning fusion process eliminates the need for manual intervention by designers.
• The approach significantly advances data fusion techniques for improved instrumentation and diagnostic applications.