Estimating Progression-Free Survival in Patients with Primary High-Grade Glioma Using Machine Learning

Area of Science:

• Neuro-oncology
• Medical Imaging
• Artificial Intelligence

Background:

• High-grade gliomas are aggressive brain tumors with poor prognoses.
• Tumor genetic diversity complicates treatment response prediction.
• Early prediction of treatment response is crucial for advancing targeted and immune therapies.

Purpose of the Study:

• To evaluate radiomics and artificial intelligence (AI) for predicting progression-free survival (PFS) in highest-grade glioma (CNS WHO 4) patients.
• To assess the utility of AI models in stratifying patients for personalized treatment plans.

Main Methods:

• Retrospective study of 51 highest-grade glioma patients.
• Extraction of 109 radiomic features from preoperative MRI scans.
• Integration of clinical data (sex, weight, age, tumor location) with radiomic features.
• Development and validation of AI models (random forest, decision tree, gradient booster, artificial neural network) to predict PFS.

Main Results:

• The random forest model achieved the highest predictive performance in the test set (1-MAPE = 92.27%, C-index = 0.9544).
• Other AI models also demonstrated satisfactory predictive capabilities.
• The study successfully predicted time to recurrence using radiomic and clinical data.

Conclusions:

• AI models combined with radiomic features show significant potential for predicting PFS in high-grade glioma.
• These findings support the use of AI in risk stratification and personalized treatment strategies.
• Further validation with larger, multicenter datasets is recommended to confirm these promising results.