Prediction of white matter hyperintensities progression based on radiomics of whole-brain MRI: a study of risk factors

Abstract

Objective \nTo explore the risk factors of predicting white matter hyperintensities progression based on radiomics of MRI of whole-brain white matter. \n \n \nMethods \nThe imaging and clinical data of 152 patients with white matter hyperintensities admitted to Zhejiang People′s Hospital from March 2014 to October 2018 were retrospectively analyzed. The whole brain white matter on baseline T1WI images of each patient were segmented by SPM12 software package, and images of white matter were imported into AK software for texture feature extraction and dimensionality reduction. At last, least absolute shrinkage and selection operator(LASSO) was used to calculate the score of radiomics signature of each patient. According to the improved Fazekas scale, patients with WMH progression were divided into three groups: any white matter hyperintensities (AWMH), periventricular white matter hyperintensities (PWMH) and deep white matter hyperintensities (DWMH). Statistical differences of clinical factors and radiomics signature between WMH progression subgroups and non-progression subgroups were compared with independent sample t test or Mann-Whitney U test, and Univariate logistic regression was used to select independent clinical risk factors and multivariate logistic regression along with imaging omics tags were used to construct predictive models, which was evaluated by ROC curve. And the correlation between the clinical indicators of independent risk factors and the texture feature of radiomics signature was analyzed. \n \n \nResults \nThe efficacy of the model for the detection for AWMH progression, PWMH progression and DWMH progression was 0.818,0.778 and 0.824, respectively. Age is an independent risk factor for AWMH progression and DWMH progression[OR(95%CI), 4.776(2.152-10.601) vs. 3.851(1.101-8.245); P<0.05]. BMI is an independent risk factor for DWMH[OR(95%CI), 3.004(1.204-7.370); P<0.05], and hyperlipidemia is an independent risk factor for AWMH and PWMH[OR(95%CI), 4.062(1.834-8.998) vs. 3.549(1.666-7.563); P<0.05]. In AWMH subgroup, Surface Area was negatively correlated with age and low density lipoprotein(LDL) (r=-0.401, -0.312), and Inverse Difference Moment_ALLDirection_offset 1_SD was negatively correlated with age(r=-0.412). In PWMH subgroup, Compactness 1 was negatively correlated with LDL(r=-0.198), and Inverse Difference Moment_angle 0_offset 7 was positively correlated with LDL(r=0.252). In DWMH subgroup, LongRunEmphasis_ALLDirection_offset 7 was negatively correlated with age(r=-0.322), and GLCMEntropy_angle0_offset 4 was negatively correlated with age(r=-0.278). GLCMEntropy_AllDirection_offset4 was negatively correlated with body mass index(r=-0.514). \n \n \nConclusion \nRadiomics based on whole-brain white matter MR imaging can predict WMH progression and identify the risk factors in high-risk populations, thus providing early additional information to conventional magnetic resonance imaging to predict WMH progression. \n \n \nKey words: \nRadiomics;\xa0White matter hyperintensities;\xa0Risk factors;\xa0Magnetic resonance imaging