Application of Explainable Artificial Intelligence in Autism Spectrum Disorder Detection

Autism spectrum disorder (ASD) is a developmental disorder typically diagnosed in early childhood. With the advent of machine learning (ML) and deep learning (DL) models, accurate diagnosis of ASD has been enhanced. However, the widespread adoption of these AI models in real-life scenarios has been limited due to their "black box" nature, which lacks transparency and interpretability. To address this, eXplainable Artificial Intelligence (XAI) models have gained popularity, offering more transparent and interpretable detection methods. This review systematically explores XAI frameworks and underlying AI models by addressing four critical research questions (RQs). Relevant research outputs were selected using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) approach from five major databases: IEEE, PubMed, Springer, ScienceDirect and ACM. From an initial pool of 1551 articles, 38 studies were identified that focused on learning models and XAI in ASD prediction. These studies were critically analysed across six modalities, twenty classifiers, and five XAI frameworks. The selected studies demonstrate the application of various XAI frameworks in enhancing the transparency and interpretability of AI models used for ASD prediction. The review highlights the benefits of XAI in improving model trustworthiness and adoption, while identifying challenges, such as the trade-off between interpretability and model performance. This review provides a comprehensive overview of the current state of the art of XAI in ASD prediction, identifying key benefits, challenges, and future research avenues. The insights gained from this review could guide researchers in further developing XAI frameworks that balance interpretability and predictive accuracy, thereby facilitating broader adoption in clinical practice.