A Multimodel Fusion Method for Cardiovascular Disease Detection Using ECG.

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence in Medicine

Background:

• Electrocardiogram (ECG) is a crucial diagnostic tool, but research is limited by a lack of well-labeled ECG databases.
• Current research often focuses on heartbeat arrhythmia detection, emphasizing ECG signal quality.

Purpose of the Study:

• To develop and evaluate a novel multimodel fusion method for enhanced ECG record classification.
• To improve the accuracy of detecting abnormal cardiovascular conditions using ECG data.

Main Methods:

• A record quality filter was designed to assess ECG signal quality.
• Implemented baseline models: random forest, multilayer perceptron, and a RESNET-based convolutional neural network.
• Constructed a new multimodel method by fusing random forest and RESNET approaches.

Main Results:

• The proposed multimodel fusion method achieved over 88% classification accuracy.
• This new method outperformed alternative approaches by integrating human-crafted features with RESNET deep features.
• Separable and multiscale convolutions were identified as vital for 1D ECG sequence classification.

Conclusions:

• A novel multimodel fusion method offers a significant advancement for abnormal cardiovascular detection using ECG data.
• The integration of diverse feature types and deep learning architectures is effective for ECG analysis.
• The findings highlight the importance of specific convolutional techniques for processing 1D ECG sequences.