Artificial neural network–based classification of gliomas using molecular and clinical data

Gliomas are the most common primary brain tumors, characterized by substantial heterogeneity in clinical course and prognosis. Accurate tumor grading remains essential for therapeutic decision-making and survival prediction. The integration of molecular and clinical variables into predictive models may improve the diagnostic accuracy beyond histology alone. This retrospective study analyzed 856 patients with glioblastoma (GBM) and lower-grade glioma (LGG) from The Cancer Genome Atlas (TCGA) database. Clinical and mutational data were processed using IBM SPSS Neural Networks (v29.0). A Multilayer Perceptron (MLP) artificial neural network was developed to classify glioma grade. Twenty-two genetic predictors (including IDH1, IDH2, TP53, EGFR, PDGFRA, and others) and demographic variables (age, sex, race) were incorporated. Model performance was assessed on an independent test set using accuracy, sensitivity, specificity, precision, F1-score, Matthews correlation coefficient (MCC), and area under the ROC curve (AUC). The model achieved an overall accuracy of 88.2%, with a sensitivity of 91.5% for GBM and specificity of 85.4% for LGG in the test set. The AUC was 0.945, indicating excellent discriminatory power. Variable importance analysis identified IDH1 mutation as the strongest predictor (normalized importance = 100%), followed by age at diagnosis (70.5%) and IDH2 mutation (54.4%). In contrast, RB1 and BCOR mutations had minimal contributions. The proposed ANN model demonstrated robust accuracy and strong discriminatory performance in glioma grading. The integration of molecular markers such as IDH1/2 with clinical variables enhances diagnostic precision and supports the potential role of artificial intelligence in clinical decision support. Future validation in prospective and multi-center cohorts, as well as multimodal data integration, may facilitate personalized treatment strategies for glioma patients.