From Predictions to Explanations: Explainable AI for Autism Diagnosis and Identification of Critical Brain Regions

ORCID

Abstract

Autism spectrum disorder (ASD) is a neurodevelopmentalcondition characterized by atypical brain maturation. However, the adaptationof transfer learning paradigms in machine learning for ASD researchremains notably limited. In this study, we propose a computeraideddiagnostic framework with two modules. This chapter presents atwo-module framework combining deep learning and explainable AI forASD diagnosis. The first module leverages a deep learning model finetunedthrough cross-domain transfer learning for ASD classification. Thesecond module focuses on interpreting the model’s decisions and identifyingcritical brain regions. To achieve this, we employed three explainableAI (XAI) techniques: saliency mapping, Gradient-weighted Class ActivationMapping, and SHapley Additive exPlanations (SHAP) analysis.This framework demonstrates that cross-domain transfer learning caneffectively address data scarcity in ASD research. In addition, by applyingthree established explainability techniques, the approach reveals howthe model makes diagnostic decisions and identifies brain regions mostassociated with ASD. These findings were compared against establishedneurobiological evidence, highlighting strong alignment and reinforcingthe clinical relevance of the proposed approach.

Publication Date

2025-01-01

Publication Title

Communications in Computer and Information Science

ISSN

1865-0937

Acceptance Date

2025-01-01

Deposit Date

2025-09-17

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

This document is currently not available here.

Share

COinS