Deep learning-based survival prediction model for adult diffuse low-grade glioma: a multi-cohort validation study.

Area of Science:

• Neuro-oncology
• Artificial Intelligence in Medicine
• Biostatistics

Background:

• Prognostication for adult diffuse low-grade glioma (DLGG) is complex.
• Clinical and molecular factors interplay, challenging accurate survival prediction.
• A need exists for robust predictive models in DLGG management.

Purpose of the Study:

• To develop and validate a deep learning model for DLGG patient survival prediction.
• To identify key prognostic factors influencing DLGG patient outcomes.
• To provide a tool for personalized prognostication in DLGG.

Main Methods:

• Analysis of 1,079 DLGG patients across three cohorts (training, internal, external validation).
• Development of a deep learning model (DeepSurv) using seven clinicopathological variables.
• Performance assessment via C-index and integrated Brier scores (IBS); feature importance analysis using permutation importance and SHAP values.

Main Results:

• The deep learning model demonstrated strong predictive performance (C-indices: 0.81-0.87) across all cohorts.
• Low IBS values (0.03-0.04) confirmed high predictive accuracy.
• Age and IDH mutation status were identified as the most significant prognostic factors, with age exhibiting non-linear effects.

Conclusions:

• A validated deep learning model offers reliable prognostication for DLGG patients.
• Age and IDH status are critical determinants of survival in DLGG.
• The model, available as a web platform, aids in personalized predictions and optimizing treatment strategies.