OBJECTIVES/GOALS: Trigeminal Neuralgia (TN) is a debilitating neuropathic condition characterized by electric-shock-like pain attacks. TN is considered a clinical diagnosis, and few proposed objective markers exist. This work studies the ability of advanced MRI techniques to diagnose and classify TN. METHODS/STUDY POPULATION: Anatomical MRI data from patients undergoing radiosurgery to treat TN was collected. A custom deep-learning UNet algorithm was trained to segment trigeminal nerves from the pons to the anterior wall of Meckels cave using segments drawn by an expert in neuroanatomy. 108 radiomics features related to nerve shape, voxel intensity, and image texture were extracted from the segmented nerves. A 2 layer neural network was trained to distinguish TN affected nerves from the pain-free contralateral nerves. Feature selection was performed within a cross-validation scheme to prevent model overfitting. Mean model performance over the validation sets was used to estimate model generalizability. RESULTS/ANTICIPATED RESULTS: 134 patients and 268 nerves were included. The average number of years with TN was 8. The average validation set accuracy was 78% [range: 75-80%]. The average validation set sensitivity and specificity were 0.82 [range: 0.79-0.84] and 0.76 [range: 0.70-0.79]. 34% of patients had undergone a prior invasive procedure to treat their TN. To evaluate whether the model detected signal changes relating to the previous treatment, those patients were excluded and the model was retrained on the surgically naive patients. Model performance in a reduced cohort of patients was similar to the model trained on all the patients, with accuracy of 77% [range: 73-82%]. DISCUSSION/SIGNIFICANCE: This study suggests that radiomics features calculated from MRIs of trigeminal nerves correlate with anatomical changes in TN affected nerves. This technique will need to be verified in a larger, more heterogeneous cohort of TN patients with a range of MRI acquisition parameters.