Explainable Deep Learning for Personalized Age Prediction With Brain Morphology.

Area of Science:

• Computational neuroscience
• Artificial Intelligence
• Neuroimaging

Background:

• Predicting brain age using magnetic resonance imaging (MRI) is crucial for identifying brain diseases.
• Deep learning (DL) models enhance prediction accuracy but often lack interpretability.
• Explainable Artificial Intelligence (XAI) methods aim to clarify DL model decisions.

Purpose of the Study:

• To develop an explainable DL framework for predicting brain age from MRI scans.
• To utilize morphological features from the ABIDE I database for age prediction.
• To compare the reliability of SHAP and LIME XAI methods in explaining DL model outcomes.

Main Methods:

• A deep learning framework was implemented for brain age prediction using MRI morphological features.
• The SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) methods were integrated for local explanations.
• The framework was applied to a healthy cohort from the ABIDE I database.

Main Results:

• The explainable DL framework successfully predicted brain age from MRI scans.
• SHAP provided more reliable explanations of morphological aging mechanisms compared to LIME.
• The study identified the contribution of specific brain morphological descriptors to predicted age.

Conclusions:

• The developed XAI framework enhances the interpretability of DL-based brain age prediction.
• SHAP is a reliable tool for understanding morphological aging and identifying personalized imaging biomarkers.
• This approach holds potential for advancing neuroimaging biomarker discovery.