An Optimized Update Method for Atrial Fibrillation Detection for Wearable Devices

Abstract

Atrial fibrillation (AF) is a disease of the elderly with high rates of disability and mortality. In order to solve the problems of missed early AF diagnosis and wearable device AF data analysis not fast and accurate enough, this paper uses deep incremental learning to train AF signals as a capture model based on AF data and normal ECG data in public databases and so on. Capturing atrial fibrillation signals from early stage clinical atrial fibrillation patients is considered as a new task, and the established capture model for the old task is updated and learned online, including the online update algorithm of multi-task atrial fibrillation signal capture model based on knowledge distillation and knowledge verification. Finally, the model parameters are adaptively optimised to solve the problems of time-consuming online updating and poor diagnostic performance of the model. The experimental results show that the diagnostic result of AF based on knowledge review is 0.94, and the diagnostic result of AF based on multi-task incremental learning is 0.88 after adding new samples from the clinic.In summary, the results of this research can improve the ability of early detection of AF, which can help promote the practical process of AF diagnostic technology in the clinic.