نهمین کنگره نقشه برداری مغز

نمایش مقاله

Title:

Cross-Task Behavioral Phenotyping in Pediatric ADHD: Unsupervised Discovery of Engagement Styles across Clinical and Game-Based Environments

Subject:

Personalized Interventions and Diagnosis of Complex Neurological Disorders

Presenter Author:
Name:	Somayeh Malekshahi

Author:
Name 1:	Somayeh Malekshahi - Presenter
Affiliation 1:	School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran

Name 2:	Reza Rostami
Affiliation 2:	Department of Psychology, University of Tehran, Tehran, Iran

Name 3:	Hamid Soltanian-Zadeh - Corresponding
Affiliation 3:	School of Electrical and Computer Engineering, College of Engineering, University of Tehran, Tehran, Iran

Name 4:
Affiliation 4:

Name 5:
Affiliation 5:

Name 6:
Affiliation 6:

Name 7:
Affiliation 7:

Name 8:
Affiliation 8:

Name 9:
Affiliation 9:

Name 10:
Affiliation 10:

Abstract

Original Research

introduction :

Clinical assessment of ADHD often overlooks behavioral heterogeneity within the diagnosis. This study applies unsupervised machine learning to identify Digital Phenotypes—consistent behavioral signatures—expressed across two distinct environments: a standardized clinical task (Integrated Visual and Auditory Continuous Performance Test; IVA-2) and a naturalistic Serious Game designed to elicit sustained engagement.

Methods:

We analyzed multimodal data (video and behavioral telemetry) from 21 participants (ages 6-12) during the 12.5-minute IVA-2 and around a 30-minute engagement-focused Serious Game. Computer-vision features were extracted from IVA-2 video recordings using OpenFace, including head-pose variance, gaze variability, and facial action unit (AU) statistics during the clinical task (IVA-2). Clinical performance measures (omission rate, commission rate, and reaction time) were integrated with these features. These digital phenotypes were then correlated with game-based behavioral telemetry (impulsivity indexed by extra click percentage and inattention indexed by click-time dispersion) to evaluate cross-environment behavioral consistency. Unsupervised clustering analysis was performed on standardized clinical and movement features to identify natural groupings, followed by non-parametric Kruskal–Wallis tests to assess the discriminative power of physiological markers.

Results:

Unsupervised clustering revealed three distinct behavioral phenotypes differing in motor variability and attentional performance. Phenotype 1 (n=8) exhibited the most stable behavioral profile, low head-pose variance (mean = 0.054), low omission rate (5.8%), and the lowest gaze variability (0.102). During the Serious Game, this group demonstrated relatively controlled interaction patterns, with moderate extra clicks (51.9%) and sustained task engagement. Phenotype 2 (n=12) showed marked behavioral instability, characterized by the highest pose variance (0.151), higher omission rates (16.1%), and significantly elevated gaze variability (0.170, p=0.007). Notably, despite their clinical severity, this group showed the lowest extra click percentage (47.6%) in the game environment, suggesting that highly engaging, reward-based tasks may mitigate the impulsivity symptoms observed in standardized clinical settings. Phenotype 3 (n=1), a high-severity outlier, combined moderate motor variability (0.081), with the highest game impulsivity (55.4%) and the shortest interaction time. Principal component analysis showed that two components explained 75.4% of behavioral variance. Significant between-cluster differences were found across multiple physiological markers, most notably in lower-face muscle activity: AU26 (Jaw Drop; p < 0.001), AU15 (Lip Corner Depressor; p = 0.001), and AU17 (Chin Raiser; p = 0.001). These findings suggest that facial tension and mouth-opening behaviors are primary physiological indicators of ADHD severity and effortful control during clinical tasks.

Conclusion:

These findings suggest that ADHD manifestations are not strictly task-dependent but represent stable Digital Phenotypes that can be mapped non-intrusively using computer vision. The discovery that facial Action Units and gaze variability effectively distinguish clinical severity groups provides a framework for more objective, context-aware ADHD monitoring. In such a framework, the environment (e.g., a game) can be dynamically adapted to the specific physiological and behavioral style of the child to maintain optimal levels of engagement and performance.

Keywords:

ADHD, Phenotyping, Pediatric, Engagement, Clinical, Serious Game