Recognition value of diffusion basis spectrum imaging brain network analysis for preschool children with autism spectrum disorder.

Objective

To explore the value of binary brain network model and weighted brain network model based on diffusion basis spectrum imaging (DBSI) for identifying preschool children with autism spectrum disorder (ASD).

Methods

Three-dimensional T1 structural images and DBSI images were acquired from 31 ASD patients (ASD group) and 33 normally developing children (HC group). Using DSI Studio, binary and weighted brain network construction were performed on each subject's DBSI brain network matrix, with calculation of clustering_coeff_average, transitivity, network_characteristic_path_length, small-worldness, global_efficiency, assortativity_coefficient, rich_club. All data were randomly divided into a training dataset (45 cases, 22/23 = ASD/HC) and test dataset (19 cases, 9/10 = ASD/HC) in a 7:3 ratio. Receiver operating characteristic curve analysis was employed to evaluate model performance.

Results

The ASD group shows lower values in global_efficiency and assortativity_coefficient, as well as rich_club_k10, rich_club_k15, rich_club_k20, and rich_club_k25 compared with the HC group. The binary model based on 3 features demonstrated ASD identification accuracy/AUC values/sensitivity and specificity of 0.8947/0.8889/0.7778/1.0000, while the weighted brain network model based on one feature showed corresponding values of 0.9474/0.9333/0.8889/1.0000.

Conclusions

Different brain network construction approaches can affect the magnitude of between-group differences in indicators including clustering_coeff_average, network_characteristic_path_length, and rich_club_k5. Brain structural network characteristics derived from DBSI may possess potential discriminative value for identifying preschool-aged autism.