Severity identification for internet gaming disorder using heart rate variability reactivity for gaming cues: a deep learning approach

• Published in: Frontiers in psychiatry 2023-11, Vol.14, p.1231045-1231045
• Main Authors: Hong, Sung Jun, Lee, Deokjong, Park, Jinsick, Kim, Taekyung, Jung, Young-Chul, Shon, Young-Min, Kim, In Young
• Format: Article
• Language: English
• Subjects: addiction; behavioral addiction; deep learning model; heart rate variability; internet gaming disorder
• ISSN: 1664-0640; 1664-0640
• DOI: 10.3389/fpsyt.2023.1231045

BackgroundThe diminished executive control along with cue-reactivity has been suggested to play an important role in addiction. Hear rate variability (HRV), which is related to the autonomic nervous system, is a useful biomarker that can reflect cognitive-emotional responses to stimuli. In this study, Internet gaming disorder (IGD) subjects' autonomic response to gaming-related cues was evaluated by measuring HRV changes in exposure to gaming situation. We investigated whether this HRV reactivity can significantly classify the categorical classification according to the severity of IGD.MethodsThe present study included 70 subjects and classified them into 4 classes (normal, mild, moderate and severe) according to their IGD severity. We measured HRV for 5 min after the start of their preferred Internet game to reflect the autonomic response upon exposure to gaming. The neural parameters of deep learning model were trained using time-frequency parameters of HRV. Using the Class Activation Mapping (CAM) algorithm, we analyzed whether the deep learning model could predict the severity classification of IGD and which areas of the time-frequency series were mainly involved.ResultsThe trained deep learning model showed an accuracy of 95.10% and F-1 scores of 0.995 (normal), 0.994 (mild), 0.995 (moderate), and 0.999 (severe) for the four classes of IGD severity classification. As a result of checking the input of the deep learning model using the CAM algorithm, the high frequency (HF)-HRV was related to the severity classification of IGD. In the case of severe IGD, low frequency (LF)-HRV as well as HF-HRV were identified as regions of interest in the deep learning model.ConclusionIn a deep learning model using the time-frequency HRV data, a significant predictor of IGD severity classification was parasympathetic tone reactivity when exposed to gaming situations. The reactivity of the sympathetic tone for the gaming situation could predict only the severe group of IGD. This study suggests that the autonomic response to the game-related cues can reflect the addiction status to the game.