Sangwon Byun1, Ah Young Kim2, Eun Hye Jang2, Seunghwan Kim2, Kwan Woo Choi3, Han Young Yu4, Hong Jin Jeon5. 1. Department of Electronics Engineering, Incheon National University, 22012, Incheon, South Korea. Electronic address: swbyun@inu.ac.kr. 2. Bio-Medical IT Convergence Research Division, Electronics and Telecommunications Research Institute (ETRI), 34129, Daejeon, South Korea. 3. Department of Psychiatry, Depression Center, Samsung Medical Center, Sungkyunkwan University School of Medicine, 06351, Seoul, South Korea; Department of Psychiatry, Korea University Anam Hospital, Korea University College of Medicine, Seoul, 02841, South Korea. 4. Bio-Medical IT Convergence Research Division, Electronics and Telecommunications Research Institute (ETRI), 34129, Daejeon, South Korea. Electronic address: uhan0@etri.re.kr. 5. Department of Psychiatry, Depression Center, Samsung Medical Center, Sungkyunkwan University School of Medicine, 06351, Seoul, South Korea. Electronic address: jeonhj@skku.edu.
Abstract
BACKGROUND: Major depressive disorder (MDD) is one of the leading causes of disability; however, current MDD diagnosis methods lack an objective assessment of depressive symptoms. Here, a machine learning approach to separate MDD patients from healthy controls was developed based on linear and nonlinear heart rate variability (HRV), which reflects the autonomic cardiovascular regulation. METHODS: HRV data were collected from 37 MDD patients and 41 healthy controls during five 5-min experimental phases: the baseline, a mental stress task, stress recovery, a relaxation task, and relaxation task recovery. The experimental protocol was designed to assess the autonomic responses to stress and recovery. Twenty HRV indices were extracted from each phase, and a total of 100 features were used for classification using a support vector machine (SVM). SVM-recursive feature elimination (RFE) and statistical filter were employed to perform feature selection. RESULTS: We achieved 74.4% accuracy, 73% sensitivity, and 75.6% specificity with two optimal features selected by SVM-RFE, which were extracted from the stress task recovery and mental stress phases. Classification performance worsened when individual phases were used separately as input data, compared to when all phases were included. The SVM-RFE using nonlinear and Poincaré plot HRV features performed better than that using the linear indices and matched the best performance achieved by using all features. CONCLUSIONS: We demonstrated the machine learning-based diagnosis of MDD using HRV analysis. Monitoring the changes in linear and nonlinear HRV features for various autonomic nervous system states can facilitate the more objective identification of MDD patients.
BACKGROUND: Major depressive disorder (MDD) is one of the leading causes of disability; however, current MDD diagnosis methods lack an objective assessment of depressive symptoms. Here, a machine learning approach to separate MDDpatients from healthy controls was developed based on linear and nonlinear heart rate variability (HRV), which reflects the autonomic cardiovascular regulation. METHODS: HRV data were collected from 37 MDDpatients and 41 healthy controls during five 5-min experimental phases: the baseline, a mental stress task, stress recovery, a relaxation task, and relaxation task recovery. The experimental protocol was designed to assess the autonomic responses to stress and recovery. Twenty HRV indices were extracted from each phase, and a total of 100 features were used for classification using a support vector machine (SVM). SVM-recursive feature elimination (RFE) and statistical filter were employed to perform feature selection. RESULTS: We achieved 74.4% accuracy, 73% sensitivity, and 75.6% specificity with two optimal features selected by SVM-RFE, which were extracted from the stress task recovery and mental stress phases. Classification performance worsened when individual phases were used separately as input data, compared to when all phases were included. The SVM-RFE using nonlinear and Poincaré plot HRV features performed better than that using the linear indices and matched the best performance achieved by using all features. CONCLUSIONS: We demonstrated the machine learning-based diagnosis of MDD using HRV analysis. Monitoring the changes in linear and nonlinear HRV features for various autonomic nervous system states can facilitate the more objective identification of MDDpatients.
Authors: Carmen Schiweck; Ali Gholamrezaei; Maxim Hellyn; Thomas Vaessen; Elske Vrieze; Stephan Claes Journal: Front Psychiatry Date: 2022-04-18 Impact factor: 5.435