P13.11.B CLINICAL-RADIOMICS MODEL FOR LOW-GRADE GLIOMA SURVIVAL PREDICTION

Abstract

Abstract BACKGROUND Low-grade gliomas are by the latest WHO classification exclusively IDH-mutated tumors, and are further divided into astrocytomas (1p19q non co-deleted) and oligodendrogliomas (1p19q co-deleted). Radiomics, that is extraction of mathematical features of images, has prognostic potential and the method is able to capture properties of whole tumors, in contrast to typical biological samples. Our aim was to investigate how a radiomics prognostication model performs in relation to a clinical prognostication model for IDH-mutated tumors. MATERIAL AND METHODS 143 cases (71 astrocytomas and 72 oligodendrogliomas) were included in the analysis. Median follow-up time was 75 months. The data was divided in training and test set with 70:30 ratio. Four different radiomics models were investigated including image features extracted from tumor and peritumor zones respectively. RESULTS Preliminary results have shown best radiomics model performance for prognostication purposes in IDH-mutated tumors were based on the peritumoral zone 0-10 mm around the segmented tumor. This model resulted in a c-index of 0.88 on training set and 0.76 on test set in explaining the overall survival when the cohort of IDH-mutated tumors were analyzed together. The model will also be evaluated in the molecular subgroups and on an external test set, and finally compared to a clinical model. CONCLUSION Our findings will identify the top features in radiomics prognostication of IDH-mutated tumors, and also provide important data on the best area for radiomics extraction in these cases. The clinical implications of our findings will show whether radiomics adds additional value to the clinical model of low-grade glioma prognostication.