Application of MRI-Based Artificial Intelligence Models for Preoperative Molecular Subtyping and Prognostic Assessment of Midline Gliomas: A Multicenter Prospective Clinical Study

This study aims to validate the clinical value of an MRI-based artificial intelligence model for personalized diagnosis and treatment in patients with midline gliomas. The model integrates preoperative MRI features with clinical variables (e.g., age, sex, and other relevant patient characteristics) to predict both molecular subtypes and patient prognosis.

Model workflow. The model takes as input tumor-containing slices from preoperative MRI sequences, along with patient age and sex. By recognizing information within the MRI sequences, the model outputs the predicted molecular diagnosis for the patient.

Primary objective. To evaluate the model's accuracy in preoperative molecular subtyping of midline gliomas (H3K27M, IDH, and 1p/19q status) by comparing its predictions with the gold standard of postoperative or post-biopsy pathology. Diagnostic performance will be assessed using sensitivity, specificity, accuracy, F1 score, and area under the receiver operating characteristic curve (AUC).

Secondary objective. To assess the model's prognostic capability by integrating imaging features with clinical variables to predict patient survival outcomes and treatment response. Prognostic performance will be evaluated using time-dependent AUC and calibration metrics.

Exploratory objective. To explore the model's added value in clinical decision-making, including its potential to guide preoperative treatment planning and risk stratification.