Deep learning and machine learning in survival prediction of glioblastoma: A scoping review

Abstract

Glioblastoma (GBM) has been reported to be the most common tumor associated with the brain and central nervous system and, despite treatment, has overall poor outcomes. Survival prediction for GBM patients remains a challenge, and to address these challenges, various advancements in machine learning (ML) and deep learning (DL) techniques, utilizing magnetic resonance imaging data for survival, have been tested with variable accuracy and validity. It is anticipated that advanced noninvasive methods, such as radiomics and artificial intelligence, shall offer new possibilities for improving survival estimation and help to create personalized survival predictions into clinical practice to enhance decision-making. The current challenges that need to be addressed include incomplete or missing data from patients, including treatment or molecular information, results in biased predictions and diminished model accuracy, and the need for large-scale, high-quality datasets, which is important for the training of ML and DL models.