OCD Severity Based on EEG Signals

Obsessive Compulsive Disorder (OCD) is a mental condition that causes constant thoughts to perform repetitive activities. It may occur in different aspects and different severities. Determination of OCD severity can help to choose more effective treatments. Rule-based Representation Learner (RRL) is a recently introduced method, for solving rule-based models issue, i.e. difficulty in optimization, especially in the case of large scale datasets. RRL solves the problem by learning interpretable non-fuzzy rules. Additionally, “gradient grafting” was proposed to improve the RRL performance, and was used as a new training approach that can directly optimize the discrete model using gradient descent. Due to the astonishing features of RRL, in this paper we propose applying this method to OCD data. An open-source dataset is used for this study which contains EEG, eye-tracking, and vegetatics from ۳۲ OCD patients. Three severity classes including low, intermediate, and high are defined based on Y-BOCS scores. Applying RRL to EEG data results in an accuracy of ۹۳.۸۳% which outperforms previous work.

Keywords:

RRL , OCD , OCD severity , classification , EEG