Artificial Intelligent Feature Point Identification for Multi-Channel Mechanocardiogram Spectrum

This proposal plans to design and develop an artificial intelligence-based feature point identification method for multi-channel mechanocardiogram spectrum. The process of feature point identification is the most basic and fundamental step to lead the Mechanocardiography measurement technology to its valuable applications on various vascular cardiac diseases. Even in the same human subject, the cardiac vibration period varies a lot and not to mention the periods among different subjects. Therefore, if the feature point identification only relies on manual identification, it would restrict its applications. Even to summarize the methods into rules for auto feature point identification, it lacks of a commonly known rule set which can provide high precision for the feature point identification results so far because of the huge variance on the signal. As for the artificial intelligence based methods, the performance was still not good enough. Even assuming the feature points can be identified, there is no way to evaluate their reliability so far. In this study, we propose using of mechanocardiogram signal classification to assist the normalization of MCG signals with dynamic time wrapping method to reduce the huge variance of the signals. We also propose to use 1D-CNN and transfer learning methods to increase the performance of artificial intelligence-based feature point identification. A feature point identification reliability formula and method are also proposed to evaluate the correctness of the identification results. With these ways, the goal of this research proposal is to develop a high performance and high precision artificial intelligence-based feature point identification for the multi-channel mechanocardiogram spectrum measurement technology. The achievements and developed algorithms from this study could be integrated into the vascular cardiac disease early prediction system based on the multi-channel mechanocardiogram spectrum measure technology which is going to be developed next. This would help the deployment of the system and bring it more close to the goals. Furthermore, it would reduce the difficulties and contribute a lot for the potential vascular cardiac disease patients to perform the self-management and self-monitoring of cardiac health at home.

Project ID:PB10907-2880
External Project ID:MOST109-2221-E182-016