TIC-XNet: a structured evidence translation framework for interpretable multimodal pediatric tic event detection with...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Tic disorders affect children, necessitating accurate detection methods.
• Current diagnostic tools may lack objectivity and interpretability.
• Multimodal data integration offers potential for improved tic event detection.

Purpose of the Study:

• To develop an interpretable multimodal framework (TIC-XNet) for detecting tic events in children.
• To translate model decisions into structured, time-aligned evidence from synchronized video and physiological signals.
• To enhance the understanding and diagnosis of tic disorders through explainable AI.

Main Methods:

• TIC-XNet jointly analyzes synchronized video, heart rate, and electrodermal activity signals.
• Data collected from 417 children with diagnosed tic disorders during clinical assessments and home observations.
• TIC-XNet compared against a black-box model and a post-hoc explainable model.

Main Results:

• TIC-XNet achieved superior performance with a window-level AUC of 0.915 ± 0.019.
• Demonstrated higher event-level recall, precision, fewer missed events, and lower prediction latency.
• Translated outputs showed greater decision fidelity, stability, and temporal alignment with expert annotations.
• Subject-level signals correlated with tic severity.

Conclusions:

• Evidence translation supports interpretable multimodal detection of tic events.
• TIC-XNet offers strong predictive performance with enhanced interpretability for tic disorders.
• This framework can improve diagnostic accuracy and understanding in pediatric tic disorders.