Objective Assessment of Depression Using Multiple Physiological Signals

At present, the diagnosis of depression in clinical practice mainly relies on the subjective judgment of physicians and patients, and lacks a more objective diagnostic method. Previous studies have proposed that depressed patients have abnormal autonomic nervous system activities and inadequate response under cognitive tasks, which can be objectively assessed using physiological signal features. In this study, Electrocardiographic (ECG) and photoplethysmogram (PPG) signals were collected from 17 depressed patients and 19 healthy controls in resting and task states. Statistical, time-domain, frequency-domain, and non-linear features were extracted. Unlike previous studies, linear fusion and merge fusion were performed on the features of both resting state and task state. LightGBM feature importance was adopted for feature selection, And the LightGBM classification algorithm was used to distinguish depressed patients from healthy controls. The accuracy of the fusion modality for both resting state and task state higher than that of the single modality, such as only rest state, only task state, which can obtain 85.32% accuracy. Conclusions: It is shown that the fusion of resting and task states obtains higher accuracy rate for depression recognition compared with individual resting state and individual task state, that the method of multimodal features based on LightGBM Classifier is effective for depression assessment, and that this study may provide some help in the objective assessment of depression.