OBJECTIVES/GOALS: Patient-specific definition of extent of surgical excision is foundational to the safety offered by computer assisted interventions. Consequently, this study aims to develop a pipeline for automated segmentation of bone removed during cortical mastoidectomy, a technically complex otologic surgery. METHODS/STUDY POPULATION: A simulator, previously developed in our lab, allows fully immersive simulation of mastoidectomy using segmented temporal bones generated from CT data. Using the simulator, one attending surgeon will perform three trials of mastoidectomy on 20 different temporal bones. From the simulator we will obtain data on the volume of bone removed for a specific anatomy, averaged between trials. No new U-net (nnU-net), an open-source three-dimensional segmentation network, will then be trained to predict the volume of bone removed using segmented pre-operative CT imaging. Segmentation accuracy will be evaluated with the Dice coefficient, modified Hausdorff distance (mHD), sensitivity and specificity. RESULTS/ANTICIPATED RESULTS: We expect the mean pairwise Dice coefficient to be high indicating relative similarity of volume removed between trials. Moreover, we predict that following five-fold cross-validation the best model will result in a Dice coefficient, mHD, sensitivity, and specificity indicative of volume removed predictions consistent with surgeon-generated data. Finally, given that network training will penalize overlap of the predicted excised bone segment and previously segmented anatomic structures, we expect that no critical anatomical structures will be marked as tissue removed. DISCUSSION/SIGNIFICANCE: We hope to show that deep learning architectures can accurately predict bone removed during mastoidectomy. These predictions can be used for preoperative planning, as clinical endpoints in surgical simulators, or be used in conjunction with surgical robots, all ultimately improving patient safety.