Abstract

Info

Title:: Intelligent depression detection based on multi-physiological signals acquired by wearable devices

Author(s):: CAO Keming; ZHAO Lulu; ZHAO Minghui; WANG Zining; LI Jianqing; LIU Chengyu; State Key Laboratory of Digital Medical Engineering/School of Instrument Science and Engineering, Southeast University, Nanjing 210096, China

Keywords:: wearable device multimodal artificial intelligence depression recognition

Abstract:: Depression, as a severe psychological and psychiatric disorder, significantly impairs the long-term physical and mental health of patients. Current depression detection methods are plagued by strong subjectivity, limited techniques, and inadequate intelligence. Previous studies have mostly relied on single-modal signal analysis, making it difficult to comprehensively reflect the multidimensional characteristics of depression. Based on the independently developed intelligent depression detection system, wearable devices are used to collect prefrontal dual-lead EEG signals, PPG signals, and single-lead ECG signals. Data from 30 patients with depression and 40 healthy controls are collected and analyzed. A multimodal depression recognition model named RBLF-Net is proposed, which integrates spatiotemporal features, weighted attention, and random forests to utilize the multi-signal features for depression recognition. The model exhibits superior performance in the five-fold cross-validation, achieving a classification accuracy of 81.43%, a precision of 81.02%, and a recall rate of 81.25%, outperforming other comparative models, and thus providing an intelligent analysis approach for depression recognition from the perspective of multi-modal fusion.