OTHR-03. ENHANCEMENT OF T1W-GAD MRI IS ASSOCIATED WITH POST-SRS LOCAL CONTROL OF NSCLC BRAIN METASTASES

BACKGROUND: Local control (LC) of brain metastasis (BM) is an important clinical endpoint. To date predictors of LC have been limited to patient and treatment related factors. Quantitative imaging features predictive of LC have not been well described for BMs treated by stereotactic radiosurgery (SRS). This study aimed primarily at assessing quantitative imaging predictors of LC that may be used for tailored SRS treatment of BM patients. METHODS: A cohort of non-small cell lung cancer (NSCLC) treated with SRS alone were identified. Post-operative SRS, radiosurgical boost, or prior WBRT cases were excluded. All patients underwent pre-SRS and follow-up T1-Gad MR imaging (termed here T1-SRS and T1-FWU). BM regions were outlined using T1-SRS during treatment planning. LC was assessed for each treated lesion by a Radiation Oncologist. Intensity histograms were normalized to account for inter-individual brain signal heterogeneity. For each BM, computed predictor factors were derived from established LC markers (volume), features associated with tumor shape (compactness, eccentricity), and signal intensity distribution in BM region (percentiles, standard deviation). RESULTS: A total of 106 NSCLC BMs in 82 participants (41 female) were examined. Mean follow-up time was 9±9 months (median 6.5 months). Kaplan-Meier (KM) curves for LC were split by the predictor factors, with split threshold ranging between -0.5 and 0.5 of sample standard deviation, optimized to maximize the difference between lower and upper curves. KM curves for lower volume (p=0.02), lower eccentricity (p=0.004), higher intensity standard deviation (p=0.02), and higher 95th intensity percentile (p=0.05) resulted in significantly higher LC. CONCLUSION: Volume, eccentricity, intensity standard deviation, and 95th intensity percentile were found to predict LC. Intensity standard deviation and intensity percentile as predictors of LC merit validation in larger, independent datasets or in future prospective studies.