Optimizing MR Image Registration for Accurate Brain Volume Measurement in Children with Autism Spectrum Disorder

Autism Spectrum Disorder (ASD) encompasses a range of neurodevelopmental conditions, underscoring the need for early and accurate diagnosis. This study aims to identify regions of interest (ROIs) in the brain that significantly differ between children with ASD and typically developing subjects (TDS), enabling precise classification. We propose novel MR image registration algorithms using the ICBM152 digital atlas, applying both rigid and non-rigid transformations. The volumes of 116 gray matter regions were measured using the MNI-AAL atlas, aligned with ICBM152, and compared between the two groups. MR images from 26 typically developing and 26 autistic children, aged 5 to 10 years, were obtained from the ABIDE database. Our registration algorithms achieved a cross-correlation coefficient (CCC) of 0.82 for the ASD group, compared to 0.65 with SPM's Powell method. Non-rigid registration further improved the CCC to 0.88 and reduced convergence time by 2.8 times. Significant increases in gray matter volumes were observed in specific brain regions in the ASD group, consistent with the age-specific theory of autism. Finally, our SVM-based classification achieved 92.3% accuracy, with an F1-score of 90.1%, recall of 90.1%, and precision of 89.4%, demonstrating the method's effectiveness in distinguishing ASD from TDS based on brain features.