Abstract
The increasing prevalence of Autism Spectrum Disorder (ASD) has intensified research efforts aimed at clarifying its neurobiological underpinnings. Electroencephalography (EEG) has enabled the identification of functional alterations in neuronal networks, contributing to the characterization of ASD-related brain dynamics and supporting the investigation of links between neural processes and behavioral impairments. In recent years, Artificial Intelligence (AI) methods have been increasingly applied to EEG analysis, allowing the extraction of complex, high-dimensional features. However, the limited interpretability of many AI-based models represents a major barrier to their clinical translation. To address this issue, Explainable Artificial Intelligence (XAI) approaches have emerged as promising tools to enhance model transparency and neurobiological interpretability. This systematic review examined studies explicitly applying XAI techniques to EEG or event-related potential data from individuals with ASD. A comprehensive literature search was conducted across multiple electronic databases up to November 2025. Studies were included if they involved ASD populations, electrophysiological data, and AI-based analytical approaches with explicit explainability components. Due to substantial methodological heterogeneity, a qualitative narrative synthesis was performed. Eleven studies met the inclusion criteria. Overall, included articles highlighted partially overlapping electrophysiological patterns involving spectral alterations, functional connectivity, and network organization; however, some studies also revealed marked heterogeneity in study design and limited clinical characterization. Consequently, they should be interpreted with caution, as the field remains at a preliminary stage. This review outlines current trends, methodological limitations, and key gaps in XAI-driven EEG research in ASD, and discusses future directions toward clinically meaningful and interpretable neurophysiological biomarkers. The review protocol was registered in PROSPERO (CRD420251231630).