An explainable machine unlearning algorithm framework for early detection of autism spectrum disorder and brain disabilities in newborns

Autism Spectrum Disorder (ASD) is a rising disorder in the brain among newborns aged 0 to 36 months, emphasizing the urgent need for early and accurate detection to mitigate potential developmental risks. Deep learning and machine learning techniques have shown promise in diagnosing ASD, particularly in children aged 36 to 72 months. However, the effectiveness of these diagnostic models heavily depends on the quality and accuracy of the training data. Inaccurate or mismatched data can lead to unreliable diagnoses and inappropriate clinical interventions. In this paper, we propose a machine unlearning-integrated explainable AI (XAI) framework for the early detection of autism spectrum disorder and brain disabilities in newborns. Central to our framework is the Machine Unlearning-Integrated Explainable AI Partitioning Autism Detection and Brain Disorder (MUXAI-ASDBD) algorithm, a novel partitioning-based XAI approach built on Convolutional Neural Network (CNN). This algorithm provides layer-wise explainability during training and incorporates a machine unlearning mechanism at the summation layer, along with probability distribution analysis, to enhance diagnostic accuracy and computational efficiency. The MUXAI-ASDBD framework was evaluated using multiple benchmark datasets, including Ages and Stages Questionnaires (ASQ), CSBS, PEDS, the Modified Checklist for Autism in Toddlers (M-CHAT), and the Screening Tool for Autism in Toddlers and Children (STAT). We also integrated neuroimaging data from brain disorder MRI and EEG datasets. This work extends our previously published implementation, in which we detected autism but did not correct errors, as we do here. Simulation results showed that the MUXAI-ASDBD framework reduced error rates by 23%, reduced false positives, improved accuracy to 98%, and reduced processing time for autism detection using the given multimodal data.