PL01.6.A Radiomics outperforms semantic features for prediction of response to stereotactic radiosurgery in brain metastases

Abstract Background Brain metastases show different patterns of contrast enhancement, potentially reflecting hypoxic and necrotic tumor regions with reduced radiosensitivity. An objective evaluation these might allow a prediction response to radiotherapy. We therefore investigated the potential MRI radiomics in comparison visual assessment semantic features predict early stereotactic radiosurgery patients brain metastases. Material Methods In this retrospective study, 150 308 from solid tumors (NSCLC 53% patients) treated by (single dose 17-20 Gy) were evaluated. The each metastasis (partial or complete remission vs. stabilization progression) was assessed within 180 days after radiosurgery. Patterns enhancement pre-treatment T1-weighted MR images either visually classified (homogenous, heterogeneous, ring-like) subjected analysis. Random forest models optimized cross-validation evaluated hold-out test data set (30% metastases). Results total, 221/308 (72%) responded optimal model comprised 10 outperformed solely based on (AUC, 0.71 0.56; accuracy, 69% 54%). diagnostic performance could be further improved combining resulting an AUC 0.74 accuracy 75% set. Conclusion developed allowed enhancement.