Construction of enhanced MRI-based radiomics models using machine learning algorithms for non-invasive prediction of IL7R expression in high-grade gliomas and its prognostic value in clinical practice.

BACKGROUND

High-grade gliomas are among the most aggressive and deadly brain tumors, highlighting the critical need for improved prognostic markers and predictive models. Recent studies have identified the expression of IL7R as a significant risk factor that affects the prognosis of patients diagnosed with high-grade gliomas (HGG). This research focuses on investigating the prognostic significance of Interleukin 7 Receptor (IL7R) expression and aims to develop a noninvasive predictive model based on radiomics for HGG.

METHODS

We conducted an analysis using data from The Cancer Genome Atlas (TCGA) and The Cancer Imaging Archive (TCIA), focusing on a group of 310 patients diagnosed with high-grade gliomas. To evaluate prognosis, we applied both univariate and multivariate Cox regression analyses alongside Kaplan-Meier survival analysis. Radiomics features were extracted from specific regions of interest, which were outlined by two physicians using 3D Slicer software. For selecting the most relevant features, we utilized the Minimum Redundancy Maximum Relevance (mRMR) and Recursive Feature Elimination (RFE) algorithms. Following this, we developed and assessed Support Vector Machine (SVM) and Logistic Regression (LR) models, measuring their performance through various metrics such as accuracy, specificity, sensitivity, positive predictive value, calibration curves, the Hosmer-Lemeshow goodness-of-fit test, decision curve analysis (DCA), and Kaplan-Meier survival analysis.

RESULTS

The survival analysis encompassed a total of 310 patients diagnosed with high-grade glioma, sourced from the TCGA database. Patients were stratified into high and low expression groups based on the levels of IL7R expression. Kaplan-Meier survival curves and Cox regression analysis revealed that an increase in IL7R expression correlated with a decline in overall survival (OS). The median Intraclass Correlation Coefficient (ICC) for the assessed radiomic features was determined to be 0.869, with 93 features exhibiting an ICC of 0.75 or greater. Utilizing the mRMR and RFE methodologies led to the identification of a final set comprising eight features. The Support Vector Machine (SVM) model recorded an Area Under the Curve (AUC) value of 0.805, whereas the AUC derived from fivefold cross-validation was noted to be 0.768. Conversely, the Logistic Regression (LR) model produced an AUC of 0.85, with an internal fivefold cross-validation AUC of 0.779, indicating a more robust predictive capability. We developed Support Vector Machine (SVM) and Logistic Regression (LR) models, with the LR model demonstrating a more robust predictive capability. Further Kaplan-Meier analysis underscored a significant association between elevated risk scores from the LR model and OS malignancy, with a P value of less than 0.001. GSVA analysis showed the enrichment pathway of KEGG and Hallmark genes in the high RS group. Moreover, expression levels of the LOX gene and the infiltration of M0 macrophages were significantly heightened in the high-risk score group, alongside an increase in tumor mutation burden (TMB). Interestingly, the mutation frequencies of TP53 and PIK3CA were found to be lower in the high-risk score group when compared to their low-risk counterparts.

CONCLUSION

IL7R expression is a vital prognostic marker in high-grade gliomas. The radiomics-based LR models demonstrate strong predictive capabilities for patient outcomes. Future investigations should aim to incorporate these insights into clinical practice to enhance personalized treatment approaches for patients with high-grade glioma.