Detection of High-Risk Depression Groups Based on Eye-Tracking Data

摘要

Depression is the most common psychiatric disorder in the general population. An effective treatment of depression requires early detection. In reschedule this paper, a novel algorithm is presented based on eye-tracking and a self-rating high-risk depression screening scale (S-hr-DS) for early depression screening. In this algorithm, a subject scan path is encoded by semantic areas of interest (AOIs). AOIs are dynamically generated by the POS (part-of-speech) tagging of Chinese words in the S-hr-DS items. The proposed method considers both temporal and spatial information of the eye-tracking data and encodes the subject scan path with semantic features of items. The support vector machine recursive feature elimination (SVM-RFE) algorithm is employed for feature selection and model training. Experimental results on a data set including 69 subjects show that our proposed algorithm can achieve an accuracy of 81% with 76% in sensitivity and 79% in F1-score, demonstrating a potential application in high-risk depression detection.