NIMG-01. PREDICTING POST-STEREOTACTIC RADIOTHERAPY MAGNETIC RESONANCE IMAGE OUTCOMES OF BREAST CANCER METASTASES TO THE BRAIN

Abstract BACKGROUND Stereotactic radiosurgery (SRS) is a cornerstone in the management of Breast Cancer Metastases to Brain (BCMB). While control rates are high following SRS, radiation necrosis rare but potentially devastating long-term toxicity. There clinical need for automated/semi-automated methods assess tumor response and optimize RT plans local with minimal Multiparametric MRI (mpMRI), particularly Apparent Diffusion Coefficient water (ADC) maps, contain information that mechanistically relatable voxel-level RT. We report deep learning-based approach predict post-SRS ADC FLAIR, T2-weighted (T2W), T1-weighted unenhanced (T1W) contrast-enhanced (T1WCE) images, from pre-SRS T1W, T1WCE, T2W FLAIR delivered dose map. These “forward models” will enable oncologist simulate radiologic outcomes iteratively METHODS trained variant pix2pix Generative Adversarial Network (GAN) on map data 18 BCMB patients treated stereotactic confirmed controlled locally recurrent metastases. Patients were 1-40 Gy between 2013-2019. RESULTS On test 6 patients, forward model predicted values within Gross Tumor Volume (GTV) broadly agreement ground truth maps. In expectations, also predicts increasing post-RT GTV simulated doses range 1-71 Gy. have explored an inverse required produce “prescribed” GTV. CONCLUSIONS envision models assist initial plan optimization, while may be useful daily optimization.