C Lin¹, J Qiu², K Hou³, W Lu³, W Lu³, X Liu³, J Qiu³, L Shi³*, (1) Taian Disabled Soldiers' Hospital Of Shandong Province, Taian, CN, (2) Taian Municipal Center For disease control and prevention, Taian, CN, (3) Shandong First Medical University & Shandong Academy Of Medical Sciences, Taian, CN

• 
  L Shi

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: study aimed to investigate whether the structural MRI-based radiomics and machine learning can separate the subtypes of ADHD.


Methods: study collected 88 ADHD patients’ structural MRI data of New York University (NYU) Medical Center from the ADHD-200 Global Competition, including 56 with ADHD-combined type (ADHD-1) and 32 with ADHD-inattentive type (ADHD-3). We performed image preprocessing, including image reconstruction, correction, registration and segmentation, using cat12 software based on SPM12 in Matlab 2013. The standardized gray matter (GM) images and white matter (WM) images were exported after the preprocessing was finished. A total of 1057 radiomics features were extracted from the whole GM and the whole WM using IBEX source, respectively, including first-order statistical features and high-order texture features calculated based on gray-level co-occurrence matrix (GLCM), gray-level run length matrix (GLRLM) and neighborhood intensity difference matrix (NIDM). The two-sided Wilcoxon rank sum test was used to calculate the differences between ADHD-1 group and ADHD-3 group; the features with P < 0.05 were selected for further analysis. We further selected the predictive features and classified the two groups using a sequential backward elimination support vector machine (SBE-SVM) algorithm. In this procedure, the 88 patients were divided into the training set (n = 56), the validation set (n = 7) and the test set (n = 25).


Results: structural MRI-based radiomics model was able to classify ADHD-1 and ADHD-3 (total accuracy: 84%; area under curve (AUC): 0.81; accuracy in ADHD-1 group: 93%; accuracy in ADHD-3 group: 70%). This model finally selected one GM-based NIDM-feature and 18 WM-based features, including 9 GLCM-features, 8 first-order features and one NIDM-feature (P = 0.003-0.049).


Conclusion: demonstrate that the structural MRI-based radiomics model can classify ADHD-1 and ADHD-3. The structural MRI-based radiomics may be the potential biomarker for the clinical subtyping of ADHD.

Funding Support, Disclosures, and Conflict of Interest: This study was supported by the Shandong Province Key Research and Development Program (2017GSF218075) and Taishan Scholars Program of Shandong Province.

Keywords

Brain, Feature Extraction, MRI

Taxonomy

IM- MRI : Radiomics

Contact Email