Digital pathology of the living brain: a voxel-level spatio-temporal network for explainable ADHD diagnosis from raw rs-fMRI across multiple scanner sites - Report - MDSpire
Advertisement
Digital pathology of the living brain: a voxel-level spatio-temporal network for explainable ADHD diagnosis from raw rs-fMRI across multiple scanner sites
Voxel-Level Spatio-Temporal Network Analysis for Objective ADHD Diagnosis
Overview
This study introduces VoxSTNet, a framework for diagnosing ADHD using raw resting-state fMRI data. The model operates directly on BOLD volumes, addressing limitations of traditional diagnostic methods.
Background
ADHD is a prevalent neurodevelopmental disorder affecting children and adolescents. Current diagnostic practices rely on behavioral assessments, which can lead to variability in diagnosis. The application of resting-state fMRI presents an opportunity for objective identification of ADHD.
Data Highlights
Metric
Value
Accuracy (5-fold CV)
98.7 ± 0.4%
Sensitivity
98.2%
Specificity
99.1%
AUC
99.4% (95% CI: 97.9–99.5%)
Accuracy (LOSO)
78.4% (95% CI: 75.1–81.7%)
Key Findings
VoxSTNet operates directly on raw BOLD volumes, preserving voxel-level information.
The model achieved an accuracy of 98.7% in five-fold cross-validation.
Under the LOSO protocol, the model maintained a mean accuracy of 78.4%.
HiResCAM saliency maps highlighted the right caudate nucleus as significant in ADHD diagnosis.
The findings indicate that voxel-level analysis of rs-fMRI data may enhance the objectivity of ADHD diagnosis.
Conclusion
VoxSTNet demonstrates strong performance and interpretability in the objective diagnosis of ADHD. Future research should focus on enhancing cross-site generalizability.
These 10 factors were identified in national surveys and peer-reviewed studies examining physician burnout, workload, administrative burden, staffing challenges, and practice conditions.
