Multimodal Deep Learning-Based Prognostication in Glioma Patients: A Systematic Review

Area of Science:

• Neuro-oncology
• Medical imaging analysis
• Artificial intelligence in medicine

Background:

• Malignant brain tumors, particularly gliomas, significantly impact patient morbidity and mortality.
• Advancements in clinical data collection and analysis offer opportunities for improved prognostic tools.
• Deep learning (DL) presents a promising approach for integrating diverse clinical data modalities.

Purpose of the Study:

• To systematically review the application of DL in predicting glioma prognosis.
• To assess the impact of multimodal data integration on DL model performance for glioma prognostication.

Main Methods:

• A systematic review was conducted following PRISMA guidelines.
• Databases searched included Embase, PubMed MEDLINE, and Scopus.
• Studies focused on DL-based prognostication of gliomas, analyzing predicted outcomes like overall survival and treatment response.

Main Results:

• The review identified studies predicting overall survival (81%), survival and genotype (12.5%), and immunotherapy response (6.2%).
• Multimodal data integration, combining MRI with clinical, histologic, biomarker, or genomic data, showed improved predictive performance compared to unimodal models.
• Risk of bias was noted, primarily due to inconsistent methodological reporting.

Conclusions:

• Multimodal data significantly enhances the accuracy of DL-based overall survival prediction in gliomas.
• Challenges such as data limitations and lack of transparency in reporting hinder the full realization of multimodal DL for brain tumor patients.
• Further research emphasizing transparent reporting is needed to optimize DL applications in neuro-oncology.