Predicting survival in malignant glioma using artificial intelligence.

Area of Science:

• Neuro-oncology
• Medical imaging
• Artificial intelligence

Background:

• Malignant gliomas, such as glioblastoma, are aggressive brain tumors with poor prognoses.
• Traditional survival prediction methods (Kaplan-Meier, Cox Proportional Hazards) have limitations in accuracy.
• Accurate survival prediction is crucial for glioma management and research, utilizing overall survival (OS) and progression-free survival (PFS).

Purpose of the Study:

• To compare the effectiveness of imaging-based, non-imaging, and combined AI models for glioma survival prediction.
• To highlight advancements in AI, machine learning (ML), and deep learning (DL) for integrating multimodal data.
• To address challenges and propose solutions for AI implementation in glioma prognosis.

Main Methods:

• Utilizing radiomics from imaging data to identify tumor features.
• Leveraging clinical and molecular biomarker data for non-imaging AI models.
• Developing combined AI models that integrate multimodal data sources (imaging, clinical, molecular).

Main Results:

• Imaging-based AI models demonstrate high predictive accuracy through radiomics.
• Non-imaging AI models provide complementary insights using clinical and genetic data.
• Combined AI models, integrating multiple data modalities, show the greatest potential for accurate survival prediction.

Conclusions:

• AI, particularly combined approaches, offers significant improvements in glioma survival prediction.
• Advanced AI techniques can enable personalized treatment strategies and enhance prognostic accuracy.
• Addressing limitations like data heterogeneity and interpretability is key for widespread AI adoption in glioma management.