Supplemental Material for:A Single-Center Retrospective Analysis of Artificial Intelligence-Based Emotion Recognition Systems for Attention-Regulation Training in Children with Attention-Deficit/Hyperactivity Disorder

Existing interventions typically lack real-time responsiveness and ecological validity. This study retrospectively examined historical training records from an AI-assisted closed-loop system integrating emotion recognition and behavioral feedback for attention regulation in children with ADHD.</p><p dir="ltr"><b>Methods:</b> This single-center retrospective analysis included historical training records from 60 children with ADHD who had completed either AI-assisted attention training or conventional attention training between 2021 and 2024.

The AI-assisted training-record group comprised children whose archived records included 12 sessions of integrated cognitive training involving Go/No-Go, CPT, and Flanker tasks with real-time AI-generated feedback based on task performance and facial-emotion monitoring. The conventional training-record group completed the same attention tasks without AI-based feedback. Assessments included task performance, Conners Continuous Performance Test, 3rd Edition (CPT-3), teacher ratings, and emotion metrics, analyzed using linear mixed-effects models, linear growth modeling, and structural equation modeling (SEM).</p><p dir="ltr"><b>Results:</b> Compared with conventional training records, AI-assisted training records showed greater pre-post changes in reaction time (RT), RT variability, and task accuracy, together with fewer commission errors in CPT-3.