Uncovering Health Insights: Elastic-based Motif Clustering and Digital Biomarkers Development from Wearable Device Data
時間
星期四 下午2:00~4:00
地點
數學系1樓21113演講室
摘要
Numerous studies have investigated the association between physical activity (PA) and various health outcomes. While wearable devices have made collecting free-living PA data increasingly convenient, analyzing this data remains challenging due to substantial variability in daily routines, measurement errors, and a lack of activity labels. Traditional approaches often rely on summary statistics (e.g., total activity counts), which fail to capture the nuanced, specific activity patterns essential for deep health insights. In this talk, I will introduce a novel framework for developing meaningful digital biomarkers from wearable device data. We proposed an elastic distance-based motif clustering algorithm to identify recurring specific activity patterns (motifs) in free-living PA data. Then, functional principal component analysis (FPCA) is applied to extract nuanced digital biomarkers from these identified motifs. The efficacy of our method was demonstrated through real-world applications. The results indicate that the motif-derived digital biomarkers are significantly associated with disease outcomes. Furthermore, these biomarkers serve as reliable features for predictive modeling, significantly enhancing the sensitivity and accuracy of patient classification. In conclusion, our method effectively identifies motifs in free-living physical activity data. The subsequent application of digital biomarkers derived from these motifs can advance personalized health assessment and disease detection, offering a promising future for healthcare.
彰化市進德路1號 
04-7232105 ext 3207 
04-7211192