Development and Validation of An Explainable Artificial Intelligence–Driven Structural Brain Health and Neurodegenerative Risk Index Using Automated Mri Volumetry, Cortical Thickness, Brain Age Prediction, Network-Based Biomarkers, And Hemispheric Asymmet

Abstract

Background: Artificial intelligence (AI)-based volumetric analysis has transformed structural brain MRI by enabling automated quantification of hundreds of cortical and subcortical anatomical parameters. Although numerous regional biomarkers have been proposed, current approaches rely predominantly on isolated volumetric measurements, limiting their clinical applicability. A comprehensive, explainable structural biomarker integrating volumetry, cortical thickness, hemispheric asymmetry, ventricular metrics, and network-based analysis has not yet been established.

Purpose: To develop and validate an Explainable Artificial Intelligence–Driven Structural Brain Health Index (SBHI) integrating automated MRI volumetry, cortical thickness, hemispheric asymmetry, brain age estimation, and network-specific biomarkers for objective assessment of structural brain health and early neurobiological vulnerability.

Materials and Methods: This prospective observational study included adult participants who underwent three-dimensional T1-weighted structural MRI. Automated image processing was performed using an AI-based volumetric pipeline to obtain normalized intracranial volumes, cortical thickness measurements, hemispheric asymmetry indices, ventricular metrics, and cortical and subcortical regional volumes. Composite indices—including the Brain Reserve Index, Executive Network Index, Reward Circuit Index, Limbic Vulnerability Index, Salience Network Index, Ventricular Expansion Index, Brain Age Gap, and Structural Network Efficiency Score—were generated through feature engineering. Machine-learning models including Random Forest, Extreme Gradient Boosting (XGBoost), Support Vector Machine, Logistic Regression, LightGBM, and CatBoost were developed. Explainable AI techniques, including SHAP analysis, permutation feature importance, and local interpretable model-agnostic explanations (LIME), were employed to identify the most influential neuroanatomical biomarkers.

Results: The proposed AI framework demonstrated that integrating multiple structural MRI biomarkers produced substantially greater predictive performance than isolated regional volumetric analysis. Composite structural indices consistently identified patterns involving frontal executive regions, reward circuitry, limbic structures, ventricular morphology, and cortical thickness. Explainable AI demonstrated that medial orbitofrontal cortex, anterior cingulate cortex, hippocampus, nucleus accumbens, ventricular volume, cortical thickness, and hemispheric asymmetry contributed most strongly to overall structural brain health prediction.

Conclusion: The proposed Structural Brain Health Index represents a comprehensive explainable AI framework that integrates automated MRI volumetry with advanced feature engineering to generate clinically interpretable biomarkers of brain health. This approach provides a scalable platform for individualized neuroimaging assessment and has potential applications in cognitive decline, neurodegenerative disorders, behavioral addictions, psychiatric disorders, and preventive neuroradiology.