Machine learning-based nomogram for distinguishing between supratentorial extraventricular ependymoma and supratentorial glioblastoma

Objective: To develop a machine learning-based nomogram for distinguishing between supratentorial extraventricular ependymoma (STEE) and supratentorial glioblastoma (GBM).

Methods: We conducted a retrospective analysis on MRI datasets obtained from 140 patients who were diagnosed with STEE (n=48) and GBM (n=92) from two institutions. Initially, we compared seven different machine learning algorithms to determine the most suitable signature (rad-score). Subsequently, univariate and multivariate logistic regression analyses were performed to identify significant clinical predictors that can differentiate between STEE and GBM. Finally, we developed a nomogram by visualizing the rad-score and clinical features for clinical evaluation.

Results: The TreeBagger (TB) outperformed the other six algorithms, yielding the best diagnostic efficacy in differentiating STEE from GBM, with area under the curve (AUC) values of 0.735 (95% CI: 0.625-0.845) and 0.796 (95% CI: 0.644-0.949) in the training set and test set. Furthermore, the nomogram incorporating both the rad-score and clinical variables demonstrated a robust predictive performance with an accuracy of 0.787 in the training set and 0.832 in the test set.

Conclusion: The nomogram could serve as a valuable tool for non-invasively discriminating between STEE and GBM.