Autism Detection based on Structural MRI using YOLO

This study proposes a novel diagnostic framework for Autism Spectrum Disorder (ASD) using structural magnetic resonance imaging (sMRI) and advanced object detection models using deep learning (DL) from the You Only Look Once (YOLO) family. This framework aims to contribute to early diagnosis and intervention, which are crucial for post-diagnosis treatment and symptom relief. Unlike traditional approaches, this study focuses on neuroimaging biomarkers to understand the multidimensional nature of ASD. A preprocessing pathway was developed to generate 2D image sequences from raw 3D sMRI images, extracting 50 slices in the axial, sagittal, and coronal planes and explicitly modeling age-related brain structural changes to improve diagnostic generalization. The experimental evaluation demonstrated the significant superiority of the YOLO models over traditional deep learning architectures, such as ResNet50 and VGG16. YOLOv11 achieved accuracy rates between 0.9957 and 0.98, and recall rates between 0.998 and 0.98 across the dataset. These results highlight YOLO’s diagnostic capabilities, computational efficiency, and robustness in handling MRI images. Furthermore, the framework demonstrated its reliability across YOLO versions 8–11, with YOLOv9 and YOLOv11 achieving the best overall results. This outstanding performance is attributed to the integration of multiscale features and spatial attention processes that reveal subtle neu-roanatomical abnormalities in the early stages of autism spectrum disorder. Based on these findings, we can offer promising insights into the early diagnosis of ASD using YOLO technology , which could potentially be clinically applicable in the future, supporting early detection and personalized intervention strategies.