Detecting depression tendency based on deep learning and multi-sources data
Weijun Ma,Song Qiu,Jue Miao,Mingshuai Li,Ziqing Tian,Boyuan Zhang,Wanzhu Li,Rui Feng,Chunhui Wang,Yong Cui,Chen Li,Kyoko Yamashita,Wentao Dong
DOI: https://doi.org/10.1016/j.bspc.2023.105226
IF: 5.1
2023-07-28
Biomedical Signal Processing and Control
Abstract:In the past decade, cases of depression have been increasingly reported. While early detection offers excellent advantages in reducing these cases, it still faces several challenges. Firstly, the traditional diagnosis of depression is time-consuming and expensive. Secondly, detecting depression by deep learning that solely relies on single source data leads to low prediction accuracy. Thirdly, there is a lack of established linkage datasets based on individuals' multi-source information for detecting depression tendencies. To address these limitations, the study collected multi-source linkage datasets, including social network text data, human gait image data, and human gait key point data to establish three deep learning models: the social network model based on BERT, the gait image model based on 3D convolutional, and the gait key point model based on LSTM, for predicting depression tendencies. An integrated model was then created by adopting the Majority Voting method for the ensemble learning of these three models. The experimental results demonstrated that the accuracy of the integrated model (91.51 %) was significantly improved compared to the three models mentioned before and related models based on single source data, highlighting the necessity and effectiveness of the proposed method based on multi-sources data. Finally, this study offers practical implications for the development of self-testing Apps as well as the improvement of work efficiency for psychological clinicians.
engineering, biomedical