Explainable MRI radiomics identifies a PTEN-associated imaging phenotype in adult gliomas.
Authors
Affiliations (8)
Affiliations (8)
- Division of Neuroimaging and Neurointervention, Department of Radiology, Stanford University, Stanford, CA, United States.
- Department of Electrical and Computer Engineering, National Technical University of Athens NTUA, Athens, Greece.
- Department of Radiology, Stanford University, Stanford, CA, United States.
- Department of Medicine, National and Kapodistrian University of Athens, Athens, Greece.
- Department of Biological Sciences, Kean University, Union, NJ, United States.
- Internal Medicine-Hematology, University of Patras Medical School, Rion, Greece.
- 1st Department of Neurology, Medical School, National and Kapodistrian University of Athens, Eginition Hospital, Athens, Greece.
- Department of Radiology, University of Miami, Miami, FL, United States.
Abstract
PTEN mutation is a recurrent molecular alteration in adult gliomas, associated with PI3K/AKT/mTOR pathway activation, aggressive tumor behavior, and resistance to therapy. PTEN status is currently determined through tissue biopsy, which is susceptible to sampling bias and intratumoral heterogeneity. We investigated whether conventional multiparametric MRI radiomics can identify a noninvasive, interpretable PTEN-associated imaging phenotype in adult gliomas. We retrospectively analyzed 195 adult glioma patients with known PTEN mutation status (184 PTEN-negative, 11 PTEN-positive) from the TCIA MU-Glioma-Post database. A total of 3,669 radiomic features were extracted from enhancing tumor core segmentations on T1CE, T2, and FLAIR MRI. A leakage-conscious nested cross-validation pipeline (5 outer/3 inner folds, 30 Optuna trials) was used for redundancy-aware feature selection, imbalance-adjusted model tuning, and performance evaluation across four classifier architectures. Feature-level explainability was assessed using linear SHAP-style additive contributions. <i>Post-hoc</i> Firth-penalized logistic regression was performed to assess whether the radiomic association was explained by age, IDH status, MRI timing, or treatment exposure. The primary elastic-net logistic regression classifier achieved an outer out-of-fold ROC-AUC of 0.874 (95% CI 0.808-0.936), sensitivity of 90.9%, specificity of 71.7%, balanced accuracy of 81.3%, and Brier score of 0.052. PR-AUC was 0.211 (95% CI, 0.112-0.400), reflecting a low PTEN-positive prevalence (5.6%) and indicating limited positive predictive value despite strong discrimination. The most influential features were wavelet- and LoG-filtered GLCM and NGTDM texture descriptors from T1CE, T2, and FLAIR, suggesting a multisequence phenotype characterized by heterogeneous enhancement architecture and disrupted local texture organization. The radiomic score remained independently associated with PTEN positivity after adjustment for age, MRI timing, and documented treatment exposure across multiple Firth-penalized models (OR per SD range: 1.57-1.62, all p ≤ 0.005). Conventional multiparametric MRI radiomics identifies a biologically plausible, explainable PTEN-associated imaging phenotype in adult gliomas that is robust to adjustment for clinical and treatment-related confounders. However, given the small PTEN-positive cohort, severe class imbalance, modest PR-AUC, and absence of external validation, the findings are exploratory and hypothesis-generating. Prospective multi-institutional validation in larger, molecularly balanced cohorts is required before clinical translation.