Habitat Analysis of Multi-parametric MRI for Predicting Postoperative Recurrence of Meningioma

Abstract

We used multi-parametric MRI to predict the postoperative recurrence of meningioma. Compared to the usual radiomics on analyzing the whole tumor, we used an unsupervised clustering method to explore the tumor habitats. Interpretable features were extracted from subregions of the lesion and used to build a habitat radiomic model. The habitat model achieved an AUC of 0.711 compared with 0.569 achieved by whole tumor analysis. The split subregions of the tumor also have clear biological meanings to the radiologists.