Radiomics and clinical data predict pseudoprogression after radiotherapy in high-grade glioma

Area of Science:

• Neuro-oncology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Distinguishing pseudoprogression (PsP) from true tumor progression in high-grade glioma (HGG) post-radiotherapy is challenging with conventional MRI.
• Misdiagnosis can lead to suboptimal patient management, including unnecessary surgeries or delayed effective treatments.

Purpose of the Study:

• To develop and internally validate a multivariable model for individualized pseudoprogression risk estimation in HGG patients.
• To improve diagnostic accuracy and inform clinical decision-making for HGG patients undergoing radiotherapy.

Main Methods:

• Retrospective analysis of 222 HGG patients (WHO CNS grade 3 or 4) treated with surgery and radiotherapy.
• Radiomic feature extraction and selection using LASSO modeling, retaining 17 features for the RadScore.
• Development of an integrated model combining RadScore with clinical-imaging (rCBV, ADC), inflammatory (NLR), molecular (MGMT promoter methylation), and treatment (TMZ) data.

Main Results:

• The integrated model, incorporating RadScore and clinical-imaging, inflammatory, molecular, and treatment factors, achieved an AUC of 0.811 in the validation cohort.
• RadScore alone demonstrated an AUC of 0.771, while a clinical-imaging model without RadScore achieved an AUC of 0.744.
• The model provides a framework for risk estimation by integrating diverse data types.

Conclusions:

• The developed integrated model shows promising internal validation for individualized pseudoprogression risk estimation in HGG.
• The model effectively combines radiomic, perfusion-diffusion, inflammatory, molecular, and treatment data for improved diagnostic capability.
• External validation using standardized imaging protocols is recommended to confirm generalizability.