Multimodal MRI lesion habitat-based radiomics analysis for preoperative prediction of spatial pattern in locally recurrent high-grade gliomas

Abstract

Abstract Purpose This study aims to preoperatively predict spatial patterns in locally recurrent high-grade gliomas (HGGs) based on lesion habitat radiomics analysis of multimodal MRI and to evaluate the predictive performance of this approach. Methods Our study included 121 patients with locally recurrent HGGs after maximum safe surgical resections and radiotherapy combined with temozolomide (training set, n = 84; validation set, n = 37). Local recurrence was divided into intra-resection cavity recurrence (ICR) and extra-resection cavity recurrence (ECR), according to the distance between the recurrent tumor and the surgical area or resection cavity. Radiomic features were extracted from the lesion habitat (T1WI enhancing lesion and T2WI/FLAIR hyperintense perilesional region) on contrast-enhanced T1WI and FLAIR, respectively. The LASSO was used to select radiomic features and calculate radiomics score. Logistic regression analysis was used to construct a predictive radiomics model, which was evaluated using calibration curves and the area under the receiver operating characteristic curve (AUC). Results Seven features with nonzero coefficients related to spatial recurrence patterns were selected. The radiomics score of patients with ECR was higher than that of patients with ICR in the training set [0.424 (0.278–0.573) vs. -0.030 (-0.226-0.248), p < 0.001] and in the validation set [0.369 (0.258–0.487) vs. 0.277 (0.103–0.322), p = 0.033]. The radiomics model demonstrated good calibration and performed well in predicting ECR, with AUC values of 0.844 in the training set and 0.706 in the validation set. Conclusion Radiomics analysis of lesion habitat can preoperatively predict spatial patterns in locally recurrent HGGs, providing a basis for determining personalized treatment strategies for HGGs.