10016- NI-2 RADIOMICS ANALYSIS OF NON-ENHANCING LESIONS AFTER BEVACIZUMAB ADMINISTRATION IN RECURRENT GLIOBLASTOMA

Abstract INTRODUCTION Identifying non-enhancing lesions (nCETs) in glioblastomas within the high-intensity area on T2/FLAIR is challenging. Bevacizumab (BEV) administration can reduce the size of contrast-enhancing lesions (CETs), which may correspond to the nCET. Elucidating these imaging features could enable visualization of non-enhancing lesions in glioblastomas. The current study aimed to investigate imaging features useful for identifying nCET by performing radiomic analysis on areas where CETs were transformed into nCET after BEV administration. MATERIALS AND METHODS The study included 25 patients with recurrent glioblastoma in which contrast-enhancing lesions on contrast-enhancing T1-weighted images shrank after BEV administration at the Asahikawa Medical University Hospital. FLAIR images, T2-weighted images, and ADC images were linearly transformed to the standard brain coordinate system with recommended signal correction. Regions of interest were defined as non-contrast T2/FLAIR areas (T2FL-H) before BEV administration (Pre-BEV) and reduced contrast enhancement areas (nCET) after BEV administration (Post-BEV). A total of 147 image features were extracted from Pre-BEV FLAIR images, T2-weighted images, and ADC images with T2FL-H and Post-BEV with nCET. The relationship between each imaging feature and T2FL-H or nCET was investigated. RESULTS Univariate analysis showed 81 statistically significant imaging features (p<0.05). FLAIR_GLRLMRln_SD exhibited the highest ability to discriminate between T2FL-H and nCET (n=40, AUC=0.9825, p<0.0001). An ideal cut-off for detecting nCET was FLAIR_GLRLMRln_SD < 332.4-354.0, with a sensitivity of 95% and specificity of 95%. CONCLUSION FLAIR_GLRLMRln_SD was able to most accurately identify areas where the contrast effect disappeared after BEV administration. It may be possible to accurately detect the nCET in glioblastomas by applying this image feature.