Radiomic Features of Atherosclerotic Plaques in the Anterior and Posterior Intracranial Circulation: Multicenter Prospective Study

  Objective   The study aimed to explore the risk factors of atherosclerosis in the intracranial anterior circulation (AC) and posterior circulation (PC), and to explore their features of conventional high-resolutionmagnetic resonance imaging (HRMRI) and radiomics approach.

  Methods   Consecutive patients with suspected symptomatic intracranial arterial stenosis were enrolled from Beijing Tian Tan Hospital and Chinese People's Liberation Army General Hospital between September 2014 and January 2017. All patients underwent HRMRI. Related risk factors (age, hypertension, diabetes mellitus, hyperlipidemia, smoking, obesity, sex), conventional HRMRI features (plaque distribution, calcification, hemorrhage, remodeling pattern), and radiomic features were compared between the AC and PC groups. All plaques were segmented manually.

  Results   A total of 141 patients were enrolled in this study. There were 60 patients in the AC group (including 52 with infarction and 8 with transient ischemic stroke) and 81 patients in the PC group (including 62 with infarction and 19 with transient ischemic stroke); ischemic type showed no statistical significance between the two groups (χ²=2.282, P=0.131). Among the risk factors, hypertension, diabetes mellitus, and hyperlipidemia showed the statistically different between the AC and PC groups (χ²=7.047, 5.979, 11.176, and P=0.008, 0.014, 0.001, respectively). Sixty-three culprit plaques of AC and 105 of PC were identified and the plaque distribution between AC and PC groups showed a significant difference (χ²=34.363, P < 0.001). Among 178 radiomic features, 21.3%(37/174) showed significantly different between AC and PC plaques (t=2.0052-7.7029, P < 0.05). The five selected features with the highest discriminant accuracy were GLCM Cluster Shade, Shape maximum 2D Diameter Colum, Shape maximum 2D Diameter Row, Firstoder Skewness, and Shape Least Axis, and their area under the receiver operating curve were 0.807, 0.760, 0.786, 0.791, and 0.746, respectively.

  Conclusions   There are significant differences in risk factors, conventional HRMRI features, and radiomic features between intracranial AC and PC atherosclerosis. More differences can be detected by radiomic features rather than the conventional HRMRI.