Towards Uncertainty Quantification for Electrode Bending Prediction in Stereotactic Neurosurgery

摘要: Implantation accuracy of electrodes during stereotactic neurosurgery is necessary to ensure safety and efficacy. However, deflect from planned trajectories. Although mechanical models data-driven approaches have been proposed for trajectory prediction, they lack report uncertainty the predictions. We propose use Monte Carlo (MC) dropout on neural networks quantify predicted electrode local displacement. compute image features 23 stereoelectroencephalography cases (241 electrodes) them as inputs a network regress MC with 200 stochastic passes To validate our approach, we define baseline model without compare it using 10-fold cross-validation. Given starting trajectory, bending inferred displacement at tip via simulation. found performed better than non-stochastic provided confidence intervals along electrodes. believe this approach facilitates decision making prediction in surgical planning.