Explainable incremental-value analysis of apparent diffusion coefficient and arterial spin labeling radiomics for ATRX status prediction in glioblastoma.
Authors
Affiliations (8)
Affiliations (8)
- Division of Neuroimaging and Neurointervention, Department of Radiology, Stanford University, Stanford, CA, United States.
- Department of Information Technology, K. J. Somaiya School of Engineering, Somaiya Vidyavihar University, Mumbai, India.
- Department of Electrical and Computer Engineering, National Technical University of Athens NTUA, Athens, Greece.
- 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
Alpha-thalassemia/mental retardation syndrome X-linked (ATRX) mutation is an uncommon but biologically relevant molecular feature in glioblastoma (GBM), linked to tumor heterogeneity, DNA damage response pathways, and treatment-relevant biology. Noninvasive prediction of ATRX status remains challenging, and the incremental value of physiologic MRI beyond structural imaging is unclear. We analyzed 106 patients with GBM with available ATRX status and complete multiparametric MRI. Four radiomics models were compared. Model 1 used structural MRI features from contrast-enhanced T1-weighted, T2-weighted, and FLAIR images, along with age and sex. Model 1A additionally incorporated ADC and ASL-CBF radiomic features. Models 1B and 1C served as ablation models isolating the individual contribution of ADC and ASL-CBF, respectively. Six machine-learning classifiers were evaluated using stratified cross-validation, class-imbalance-aware metrics, bootstrapped confidence intervals, paired DeLong testing, and SHAP explainability. The best structural model achieved an ROC-AUC of 0.721, a PR-AUC of 0.322, and a sensitivity of 0.737. Model 1A demonstrated statistically significant improvements, achieving an ROC-AUC of 0.753, a PR-AUC of 0.364, and a sensitivity of 0.947. Across classifiers, ADC and ASL improved discrimination in five of six classifiers (DeLong, p<0.05). SHAP analysis showed that age remained the dominant predictor, while ASL- and ADC-derived texture features contributed meaningful physiologic information. ADC and ASL-CBF radiomics provide a modest but statistically significant incremental value for ATRX prediction in GBM. These findings support further validation of functional MRI sequences as a complementary radiogenomic marker.