Decoding GBM Tumor Dynamics: AI-Driven Segmentation with RANO Criterion Validation for the Prediction of Radiotherapy Outcomes

Abstract

Glioblastoma Multiforme (GBM) is a very aggressive brain tumor which has poor prognosis despite wide range of treatment modalities with specific enhancements in radiotherapy. Correct evaluation of tumor response to treatment is crucial for guiding treatment decision-making for patients. Despite the wide application of deep learning models for tumor segmentation and evaluation, their fundamental complexity has cast doubt on whether a simpler, traditional approach can yield insights of comparable reliability. A retrospective analysis was performed using MRI data from 18 GBM patients who had radiotherapy.  An experienced radiologist evaluated all pre- and post-treatment MRI’s and provided RANO scores to determine the tumor response.  Multiparametric MRI sequences were segmented using Otsu's thresholding and GMM methods across sagittal, coronal, and axial planes.  Dice Similarity Coefficients (DSC) and Intensity Distribution Scores (IDS) were used to evaluate tumor changes, with low DSC and high IDS values indicating successful treatment.  The segmentation and statistical results were then compared with RANO scores to confirm the findings. The results demonstrated different tumor dynamics among patients, highlighting the variability in treatment outcomes. DSC and IDS offered additional insights into tumor alterations, where low DSC and high IDS values were determined as signs of successful radiotherapy. Both techniques effectively predicted outcomes with notable alterations, showcasing their capability for evaluating radiotherapy effectiveness in GBM treatment. This method provides a more straightforward, budget-friendly option compared to deep learning, yielding valuable understanding of tumor dynamics. Future research should prioritize confirming these results in more extensive groups by integrating advanced AI techniques.