Resolving spatial response heterogeneity in glioblastoma.

Area of Science:

• Neuro-oncology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Glioblastoma (GBM) presents significant challenges for treatment response assessment due to its spatial intratumoral heterogeneity.
• Multimodal imaging and advanced image analysis offer potential solutions for evaluating GBM response heterogeneity.

Purpose of the Study:

• To develop and validate an automated image analysis pipeline for distinguishing true tumor progression (TP) from pseudoprogression (PsP) in glioblastoma using multimodal imaging.
• To assess the spatial heterogeneity of glioblastoma response to treatment.

Main Methods:

• Automated tumor segmentation and longitudinal registration were used to categorize contrast-enhancing voxels in 61 patients.
• Cerebral blood volume (CBV), FET-PET, and contrast-enhanced T1-weighted (T1w) imaging data were grouped into supervoxels for feature extraction.
• A Random Forest classifier was trained and validated using 10-fold cross-validation, with performance evaluated by AUC and classification metrics.

Main Results:

• The image analysis pipeline achieved 80.0% accuracy and a macro-weighted AUC of 0.875 in differentiating TP from PsP on supervoxel level.
• FET-PET-derived features, particularly the 10th/90th percentile and median of tumor-to-background normalized FET-PET, were significant predictors.
• CBV- and T1c-related features also contributed to the model's predictive performance.

Conclusions:

• Automated multimodal image analysis enables reliable spatial assessment of glioblastoma treatment response.
• This approach holds promise for improving the precision of local response evaluation in glioblastoma.
• Findings support the development of more personalized and localized treatment strategies for glioblastoma patients.