The value of hippocampal sub-region imaging features for the diagnosis and severity grading of ASD in children.

Background: Hippocampal structural changes in Autism Spectrum Disorder (ASD) are inconsistent. This study investigates hippocampal subregion changes in ASD patients to reveal intrinsic hippocampal anomalies.

Methods: A retrospective study from Hainan Children's Hospital database (2020-2023) included ASD patients and matched controls. We classified ASD participants based on severity, dividing all subjects into four groups: normal, mild, moderate, and severe. High-resolution T1-weighted MRI images were analyzed for hippocampal subregion segmentation and volume calculations using Freesurfer. Texture features were extracted via the Gray-Level Co-occurrence Matrix. The Receiver Operating Characteristic curve was used to evaluate seven random forest predictive models constructed from volume, subregion, and texture features, as well as their combinations following feature selection.

Results: The study included 114 ASD patients (98 boys, 2-8 years; 16 girls, 2-6 years; 17 mild, 57 moderate, 40 severe) and 111 healthy controls (HCs). No significant differences in volumes were found between ASD patients and HCs (adjusted P-value >0.05). The seven random forest models showed that single volume and texture features performed poorly for ASD classification; however, integrating various feature types improved AUC values. Further selection of texture, subregion, and volume features enhanced AUC performance across normal and varying severity categories, demonstrating the potential value of specific subregions and integrated features in ASD diagnosis.

Conclusion: Random forest models revealed that hippocampal volume, texture features, and subregion characteristics are crucial for diagnosing and assessing the severity of ASD. Integrating selected texture and subregion features optimized diagnostic efficacy, while combining texture, subregion, and volume features further improved severity grading effectiveness.