Abstract: We demonstrated that female and male glioma patients have significantly different radiomic profiles in MET PET imaging data. Future IDH prediction models shall not be built on mixed female-male cohorts, but shall rely on sex-specific cohorts and radiomic imaging biomarkers.
Introduction
Gliomas represent the most prevalent primary brain tumors, with patient prognosis heavily dictated by the mutation status of the isocitrate dehydrogenase (IDH) gene. Specifically, patients harboring IDH-mutant tumors typically experience significantly better clinical outcomes compared to those with IDH-wildtype tumors. Currently, the gold standard for determining IDH status involves invasive surgical procedures or biopsies.
Radiomics—the high-throughput extraction of quantitative features from medical imaging—presents a promising non-invasive alternative for predicting these genetic markers. However, a significant gap exists in current research: the oversight of "sex dysmorphism." Given that biological sex influences baseline brain metabolism and cancer incidence, this study investigates whether sex-blind radiomic models overlook critical nuances, potentially compromising diagnostic and prognostic accuracy.
Materials and Methods
The research team conducted a retrospective analysis of L-[S-methyl-$^{11}$C] methionine (MET)-PET scans from 35 patients (13 females, 22 males) diagnosed with astrocytic gliomas at the Medical University of Vienna. All patients had confirmed IDH1-R132H mutation status and underwent imaging prior to therapeutic intervention.
The workflow adhered to the Imaging Biomarker Standardization Initiative (IBSI) guidelines to ensure high reproducibility. Key technical methodologies included:
- Normalization Strategies: A comparison between Standardized Uptake Value (SUV) and Tumor-to-Background Ratio (TBR) was performed to determine which metric better accounted for sex-specific metabolic variations.
- Cross-Validation: A harmonized 100-fold Monte Carlo split was employed, ensuring each data subset maintained a balanced representation of sex and IDH status.
- Data Refinement: The team utilized Tomek links to identify borderline cases and Isolation Forests for outlier detection.
- Statistical Rigor: Features were ranked using R-squared and Mann-Whitney U-tests, applying Bonferroni correction to mitigate the risk of false discoveries within the small cohort.
Key Results and Insights
The study revealed that while individual radiomic features may not immediately distinguish sex in a vacuum, significant differences emerge when these features are correlated with IDH mutation status.
The Role of Normalization
Normalizing SUV to TBR (comparing tumor activity against healthy brain background) markedly improved the separation of male and female cases within the feature space. This finding suggests that sex-specific differences in healthy brain metabolism can effectively "mask" tumor-related signals if not properly normalized.
Predictive Biomarkers
Two specific features were identified as significant predictors for IDH status ($p < 0.0016$):
- Intensity Histogram Coefficient of Variation (ih.cov): A measure of metabolic heterogeneity.
- Intensity Skewness (stat.skew): A measure of the symmetry of the uptake distribution.
Among IDH-positive patients, females exhibited significantly lower metabolic heterogeneity (ih.cov of 0.031) and more negative skewness (stat.skew of -0.327) compared to their male counterparts (ih.cov of 0.068 and stat.skew of -0.123). Furthermore, TBR-based models demonstrated superior performance in discriminating between sexes and IDH statuses compared to raw SUV models.
Discussion and Conclusion
While this study is limited by a relatively small sample size ($n=35$), it provides compelling evidence that the radiomic signature of gliomas is sex-dependent. The results indicate that "one-size-fits-all" diagnostic models may be insufficient for the complexities of neuro-oncology.
Future research must focus on validating these sex-specific features in larger, multi-center cohorts. Transitioning from mixed-sex models to sex-specific radiomic biomarker signatures is a critical step toward achieving truly personalized cancer care and more accurate non-invasive diagnostics.
Summary updated