Radiomic analysis of the peritumoral zone identifies imaging signatures of glioma invasion associated with HSP70 expression.
AI interpretation is pending for this paper.
Open original publication →What the AI sees
Not AI summarized yet.
Research significance
Pending deeper interpretation.
Source abstract
OBJECTIVE: Glioma invasion into surrounding brain tissue drives disease progression but is challenging to evaluate with conventional MRI. This study aimed to determine whether radiomic features extracted from the peritumoral zone capture biologically relevant invasion-associated imaging phenotypes, using membrane-bound Hsp70 (mHsp70) as a molecular correlate. METHODS: A retrospective study was performed using preoperative MRI data from 80 patients with glioma, including adult glioblastoma (GBM), adult lower-grade glioma (LGG), and pediatric GBM/ATRT cases. Voxel-wise radiomic features were extracted from the peritumoral region and used to train a stochastic gradient descent (SGD) classifier to distinguish GBM-associated invasion-enriched imaging phenotypes from LGG-associated comparison phenotypes. Patient-level cohort separation was used throughout model development to prevent data leakage. Feature importance was assessed using SHAP analysis. In a biologically validated subgroup of 23 patients, membrane-bound Hsp70 (mHSP70) expression was quantified in peritumoral tissue samples using confocal microscopy. Patient-level radiomic invasion metrics were correlated with mHsp70 fluorescence intensity. RESULTS: The SVM classifier demonstrated good discriminative performance, achieving an AUC of 0.875 (95% CI: 0.736-0.896) in the independent test cohort. SHAP analysis identified higher-order texture features, including GLCM-, GLRLM-, and LoG-derived radiomic descriptors, as major contributors to model predictions. Reproducibility analysis demonstrated good-to-excellent agreement for all 20 SHAP-selected features, with mean ICC values ranging from 0.859 to 0.986. In the biological validation subgroup, the radiomics-derived Invasion Burden Index (IBI) showed a significant positive correlation with mHsp70 expression (Spearman ρ = 0.67, p = 0.0004, 95% CI: 0.232-0.805, n = 23). Leave-one-out sensitivity analysis demonstrated stable correlations, indicating that the observed association was not driven by individual cases. Regions exhibiting elevated invasion-associated radiomic signatures frequently extended beyond the conventional MRI-defined tumor margin and showed spatial correspondence with areas of subsequent tumor progression on follow-up imaging. CONCLUSIONS: Voxel-wise radiomic analysis of the peritumoral zone identifies invasion-associated imaging phenotypes that correlate with membrane Hsp70 expression, a biological marker associated with aggressive tumor behavior. These findings support the potential utility of radiomic invasion mapping as a non-invasive tool for characterizing infiltrative glioma biology and generating hypotheses regarding patterns of tumor progression. Further prospective studies with spatially matched biological validation are needed.