Click here to

Session: Multi-Disciplinary General ePoster Viewing [Return to Session]

Classification of Attention-Deficit/hyperactivity Disorder Using Multisite Resting-State Functional MRI: A Radiomics Features Analysis

G Liu*, L Shi, J Qiu, Shandong First Medical University & Shandong Academy of Medical Sciences, Taian, CN


PO-GePV-M-29 (Sunday, 7/25/2021)   [Eastern Time (GMT-4)]

Purpose: The precise pathology of ADHD is still unknown and apart from behavioral symptoms investigated by psychiatric methods, there is no standard biological test to diagnose ADHD. In this study, we aimed to utilize resting-state functional MRI (rs-fMRI) to extract radiomics features. We hypothesized that radiomics features based on rs-fMRI could be used as a potential biomarker to distinguish ADHD patients from healthy controls.

Methods: In this study, rs-fMRI and clinical measures from 84 subjects with ADHD and 84 healthy controls were selected from 3 sites. Using preprocessed rs-fMRI, we extracted radiomics features by the 116 regions of interest (ROIs) of automated anatomical labeling (AAL) atlas. There are 93 features of 6 radiomics feature types extracted in each ROI. Finally, a total of 32364 radiomics features were obtained including 10788 features of mean regional homogeneity (mReHo), 10788 features of mean amplitude of low-frequency fluctuation (mALFF) and 10788 features of z-transformed voxel-mirrored homotopic connectivity (zVMHC). Then the Mann-Whitney U test and Spearman correlation were used to dimension reduction. The sequential backward elimination support vector machine (SBE-SVM) algorithm was used to features selection. Last, the selected features were used to train the SVM model to estimate the classification accuracy.

Results: After dimension reduction and features selection, there were 47 radiomics features retained. By training SVM within the training dataset based on radiomics features, we got the accuracy of 98.5% (AUC = 0.998) and the accuracy in the testing dataset is 88.2% (AUC = 0.955). The classification results were further verified by permutation testing and leave-one-site-out strategy.

Conclusion: This study provides a novel strategy to make more fully utilize of rs-fMRI base on radiomics features, which not only could effectively predict ADHD, but also have the potential to identify biomarker in it.

Funding Support, Disclosures, and Conflict of Interest: Funding Support: Taishan Scholars Program of Shandong Province(TS201712065), Academic promotion programme of Shandong First Medical University No. 2019QL009. Disclosures and Conflict of Interest: The authors declare no competing financial interests.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email