NIMG-52. CORRELATION OF BIDIMENSIONAL AND VOLUMETRIC TUMOR SIZING IN NEWLY DIAGNOSED AND RECURRENT GLIOBLASTOMA

Abstract

Abstract BACKGROUND Glioblastoma (GBM) is the most common malignant primary nervous system tumor and remains incurable with a poor prognosis despite current therapy. The standard for monitoring response to treatment, as defined by the Response Assessment in Neuro-Oncology (RANO) criteria, relies on a bidimensional (2D) measurement product of T1-gadolinium enhancing disease. However, due to the variability of bidimensional tumor characterization, there is considerable interest in volumetric segmentation to assess tumor burden, which could be implemented using machine learning models offering workflow automatization. METHODS In this study, we evaluate a commercially available automated tumor segmentation tool and correlate its volumetric output to the manual assessment of bidimensional 2D enhancing tumor burden by experts. MRI examinations during systemic treatment for two retrospective cohorts of patients with either newly diagnosed (nGBM) or recurrent GBM (rGBM) were assessed. The 2D diameter product in cm^2 of enhancing tumor was determined by blinded readers and volumetric segmentations in cm^3 were automatically generated. Spearman's correlation between the manual 2D and automated volume measurements was calculated for each cohort. RESULTS 315 patients with nGBM, and 482 with rGBM were evaluated, comprising 3073 total brain MRIs in the nGBM cohort and 2439 MRIs in the rGBM cohort. Spearman's correlation between the manual 2D and automated volumetric measurements was 0.7463 (0.730, 0.762) in the nGBM cohort and 0.8598 (0.849, 0.870) in the rGBM cohort. CONCLUSION Automated segmentation tools have the potential to revolutionize radiology workflows and increase sensitivity to subtle changes on imaging. However, the differential performance of this volumetric assessment confirms the need for further refinement and customization of such automated tools to account for tumor- and patient-specific factors.