Click here to


Are you sure ?

Yes, do it No, cancel

Quantification of White Matter Hyperintensities Based On Diffusion Tensor Imaging and Support Vector Machine

L Zheng1, R Gao2, W Lu1, L Shi1, W Lu1*, J Qiu1, (1) Shandong First Medical University & Shandong Academy of Medical Sciences, Taian, Shandong, CN, (2) Shandong University Of Science And Technology, Qingdao, Shandong, CN


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

Room: AAPM ePoster Library

Purpose: Clinical quantification of white matter hyperintensities (WMHs) is based on the Fazekas scale on T2-weighted MRI images. Fractional anisotropy (FA) calculated from diffusion tensor imaging (DTI) is widely used to assess white matter lesions. Therefore, it is hypothesized that FA associated with machine learning algorithm such as support vector machine (SVM) could be used for WMHs quantification.

Methods: Subjects with WMHs were collected which included 11 Fazekas scale-0 subjects, 35 Fazekas scale-1 subjects, 17 Fazekas scale-2 subjects and 10 Fazekas scale-3 subjects. DTI images were acquired and processed by FSL. FA map were calculated from each DTI image, and segmented by the Johns Hopkins University (JHU) white matter atlas. Mean FA value of each white matter region was extracted and taken as feature for machine learning. The FA features were normalized to a range from 0 to 1. A quaternary label for subjects with different Fazekas scales was defined. Sequential backward elimination was used for feature selection. A linear SVM was configured to quantify WMHs and leave-one-out-cross-validation was applied for performance evaluation. Accuracy, sensitivity, specificity and receiver operating characteristics curve were used as evaluating metrics.

Results: Twenty-four FA features were selected by sequential feature selection approach. The linear SVM achieved a total accuracy of 84.2105% via the selected features. Quantification accuracies for Fazekas scale 0-3 subjects were 84.2105%, 65.7895%, 77.6316% and 93.4211%, respectively. The white matter regions contributed most to the quantification were the left inferior cerebellar peduncle, right cerebral peduncle, right anterior limb of internal capsule, bilateral posterior limb of internal capsule, right retrolenticular part of internal capsule, right anterior corona radiata, left posterior thalamic radiation and left cingulum.

Conclusion: DTI images and machine learning could be used to accurately quantify WMHs levels. Several white matter regions could be used as biomarkers for clinical quantification of WMHs.

Funding Support, Disclosures, and Conflict of Interest: This work was supported by the Key Research and Development Program of Shandong Province (2017GGX201010), Academic Promotion Programme of Shandong First Medical University (2019QL009), Traditional Chinese Medicine Science and Technology Development Plan of Shandong Province (2019-0359), and Taishan Scholars Program of Shandong Province (TS201712065).


Brain, CAD, Diffusion


IM- MRI : Diffusion MRI

Contact Email