AI and Machine Learning Approaches to Oral Parafunctional Syndromes: Etiology, Disease Progression, and Neuropsychological Consequences of Thumb Sucking, Onychophagia, and Body-Focused Repetitive Behaviors

Abstract

Oral parafunctional behaviors like thumb sucking, onychophagia, and body-focused repetitive behaviors (BFRBs) can have clinically important oral, behavioral, and neuropsychological consequences when frequent or excessive. The current study aims to establish an integrated artificial intelligence (AI) and machine-learning (ML) framework to investigate the correlation between the behavioral characteristics, oral disease severity and neuropsychological factors. The authors suggest a quantitative, observational, cross-sectional design with 150 between thumb sucking (PS), onychophagia (OS), other BFRB (OBS), and healthy control (HC) groups. Behavioral characteristics, oral clinical conditions and neuropsychological variables are described and analysed with ANOVA, correlation analysis, multiple regression and 6 machine learning algorithms. Results from the illustrative findings suggest higher frequency of behaviors, more intense urges, greater problem stopping behaviors and higher oral severity ratings in participants with parafunctional behaviors. There are positive relationships between behavioral duration and severity and between behavioral frequency and severity in the oral domain, as well as high levels of stress, anxiety, and impulsivity among BFRB groups. Random Forest is the model with the highest illustrative performance among the evaluated models, with an accuracy of 90.7%, an F1 of 0.901 and an ROC AUC of 0.948. Behavioral duration, behavioral frequency, oral severity, stress and anxiety are found to be important predictive variables. The research underscores the promise of combining clinical, behavioral, and neuropsychological data with AI/ML to aid in early identification of risk, individual assessment, and multi-disciplinary treatment for oral parafunctional syndromes. The numerical results should, however, be seen as illustrative and should be validated with the actual participant-level data before being published empirically.