Attention-Enhanced Temporal and Spatial Feature Extraction Network for ADHD Diagnosis based on fMRI
IEEE Journal of Biomedical and Health Informatics
|February 25, 2026
Summary
A new deep learning model, AE-STEN, improves Attention Deficit Hyperactivity Disorder (ADHD) diagnosis by analyzing brain activity patterns from fMRI data, achieving high accuracy and clinical relevance.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Medical Imaging
Background:
- Attention Deficit Hyperactivity Disorder (ADHD) is a common neurodevelopmental disorder requiring accurate diagnosis for effective intervention.
- Functional magnetic resonance imaging (fMRI) combined with deep learning offers potential for identifying ADHD-related spatiotemporal brain features.
- Current deep learning models struggle to fully capture the complex temporal and spatial characteristics of fMRI data in ADHD diagnosis.
Purpose of the Study:
- To develop an advanced deep learning framework, AE-STEN, for comprehensive spatiotemporal feature extraction from fMRI signals in ADHD.
- To enhance the modeling of temporal dependencies and the integration of static and dynamic brain network interactions for improved ADHD detection.
Main Methods:
- Proposed the Attention-Enhanced Spatiotemporal Feature Extraction Network (AE-STEN), incorporating a Temporal Cross-scale Convolutional Attention Module (TCAM) and a Spatial Collaborative Attention-Guided Graph Representation Module (SCGRM).
- TCAM captures multi-scale temporal dependencies in fMRI time series.
- SCGRM models collaborative interactions between static and dynamic fMRI data for consistent spatial feature extraction.
Main Results:
- AE-STEN achieved a classification accuracy of 76.06% ± 0.65% on the ADHD-200 dataset (747 subjects).
- The model successfully identified brain regions implicated in ADHD, aligning with established clinical findings.
- AE-STEN demonstrated strong interpretability, suggesting clinical utility.
Conclusions:
- AE-STEN effectively extracts crucial spatiotemporal features from fMRI data for ADHD diagnosis.
- The proposed model overcomes limitations of existing methods in capturing temporal dynamics and network interactions.
- AE-STEN shows significant promise for clinical application in ADHD diagnosis due to its accuracy and interpretability.
