Advanced autism detection and visualization through XGBoost algorithm for fNIRS hemo-dynamic signals

Early detection of Autism Spectrum Disorder (ASD), a neuro-developmental condition characterized by impairments in social communication, is critical for prompt intervention and care. Recent advancements in digital medicine have significantly enhanced the precision and efficiency of ASD diagnosis, management, and care coordination. These technologies offer considerable potential for optimizing treatment pathways and improving patient outcomes. However, challenges remain in sustaining long-term user engagement and effectively integrating these innovations into routine clinical practice for ASD management. In this study, we report the findings from a clinic-based, prospective investigation aimed at evaluating the diagnostic validity of ASD employing XGBoost algorithm. Of the 51 cases assessed, 24 were diagnosed with ASD, while 27 were identified as having developmental delay without autism. Functional near-infrared spectroscopy (fNIRS) was employed to measure resting-state hemo-dynamic fluctuations in the bilateral temporal lobes, revealing connectivity differences through computer vision and machine learning analysis. Agglomerative hierarchical clustering (AgHC) was employed to evaluate the active coupling between fluctuations of deoxygenated hemoglobin (HbR) and oxygenated hemoglobin (HbO) across specific channels, elucidating inter-channel relationships in neurovascular coupling. The algorithm, using a range of significant features, achieved robust diagnostic performance with area under receiver operating characteristic curve of 0.99 ± 1, demonstrating sensitivity of 98%, specificity of 95%, negative predictive value (NPV) of 98%, and positive predictive value (PPV) of 95%, indicating high diagnostic accuracy in distinguishing relevant cases. These findings underscore the potential of digital medicine to offer objective and scalable framework for the diagnosis of autism in real-world clinical environments.