Deep learning and pathomics analyses predict prognosis of high-grade gliomas

Area of Science:

• Oncology
• Radiomics
• Computational Pathology

Background:

• High-grade gliomas are aggressive brain tumors requiring accurate prognostic tools.
• Traditional prognostic factors may not fully capture tumor heterogeneity and patient outcomes.

Purpose of the Study:

• To develop and validate a prognostic model for high-grade gliomas using pathomics.
• To integrate pathological and clinical data for improved predictive accuracy.

Main Methods:

• Whole-slide images (WSIs) of tumor regions were analyzed using a deep learning model.
• Pathological features were extracted and correlated with clinical data.
• Three predictive models were constructed: pathomics-based, clinical, and combined.

Main Results:

• The combined pathomics and clinical model achieved a C-index of 0.847 (training) and 0.739 (testing).
• High-risk patients identified by the model had a median progression-free survival (PFS) of 10 months (p<0.001).
• IDH status stratification further refined PFS predictions.

Conclusions:

• A combined pathomics and clinical model demonstrates significant efficacy in predicting high-grade glioma prognosis.
• This approach offers a powerful tool for stratifying patients and guiding treatment decisions.