From Pre- to Intra-operative MRI: Predicting Brain Shift in Temporal Lobe Resection for Epilepsy Surgery

TL;DR

This work tackles brain shift during temporal lobe epilepsy surgery by predicting intraoperative deformation from preoperative MRI using NeuralShift, a U-Net-based model trained with a multi-task loss and supervised by registration-derived surrogates. The model outputs a dense displacement field, an intraoperative brain mask, and its signed distance function, enabling preoperative-to-intraoperative alignment without real-time intraoperative scans. On 98 patient cases with 9-fold cross-validation, the method achieved a Dice similarity of about 0.97 for the brain mask and reduced landmark TRE to around 1.1–1.2 mm at midline landmarks and ~2.8–3.0 mm at resection-side landmarks. This suggests potential improvements in accuracy and safety of neuronavigation during temporal lobe surgery, with public code to follow.

Abstract

Introduction: In neurosurgery, image-guided Neurosurgery Systems (IGNS) highly rely on preoperative brain magnetic resonance images (MRI) to assist surgeons in locating surgical targets and determining surgical paths. However, brain shift invalidates the preoperative MRI after dural opening. Updated intraoperative brain MRI with brain shift compensation is crucial for enhancing the precision of neuronavigation systems and ensuring the optimal outcome of surgical interventions. Methodology: We propose NeuralShift, a U-Net-based model that predicts brain shift entirely from pre-operative MRI for patients undergoing temporal lobe resection. We evaluated our results using Target Registration Errors (TREs) computed on anatomical landmarks located on the resection side and along the midline, and DICE scores comparing predicted intraoperative masks with masks derived from intraoperative MRI. Results: Our experimental results show that our model can predict the global deformation of the brain (DICE of 0.97) with accurate local displacements (achieve landmark TRE as low as 1.12 mm), compensating for large brain shifts during temporal lobe removal neurosurgery. Conclusion: Our proposed model is capable of predicting the global deformation of the brain during temporal lobe resection using only preoperative images, providing potential opportunities to the surgical team to increase safety and efficiency of neurosurgery and better outcomes to patients. Our contributions will be publicly available after acceptance in https://github.com/SurgicalDataScienceKCL/NeuralShift.

Figures (4)

• Figure 1: Preoperative-to-intraoperative MRI preprocessing pipeline. Preoperative MRI (pMRI) is rigidly aligned to the MNI template, skull stripped, and affinely registered. Intraoperative MRI (iMRI), containing the post-resection cavity, is first reoriented using a subject-specific AC--PC--IH coordinate system defined by manually annotated landmarks, and then rigidly registered to the corresponding pMRI in native space. The pMRI-to-MNI transformations are subsequently propagated to iMRI so that both modalities share a common MNI space. Finally, intensity normalisation, bias-field correction, and cropping are applied to produce standardised network inputs.
• Figure 2: Overview of the proposed brain shift prediction framework. The inference module takes as input the preoperative MRI (pMRI) and a hemisphere indicator ("half mask") encoding resection laterality, and predicts (i) a dense displacement field mapping pMRI to iMRI space, (ii) the intraoperative brain mask, and (iii) its signed distance function (SDF) using a U-Net. The supervision builder constructs training targets via non-rigid registration (NiftyReg F3D) between pMRI and iMRI, yielding a dense displacement field used to warp pMRI. The intraoperative mask is obtained by thresholding, and the corresponding SDF is computed from the mask. Training employs separate loss terms for displacement field regression, mask prediction, and SDF supervision.
• Figure 3: Two representative examples of preoperative and intraoperative MRI normalisation into MNI space. For each case, pMRI and iMRI are shown in native space and after transformation to MNI space. The rightmost column overlays pMRI and iMRI in MNI space, demonstrating improved spatial correspondence after applying the preprocessing pipeline (Fig. 1) while preserving the post-resection cavity in iMRI.
• Figure 4: Qualitative visualisation of predicted brain deformation in axial, coronal, and sagittal planes. The first three columns show the preoperative MRI, the intraoperative MRI, and the deformed preoperative MRI obtained using the predicted displacement field. The remaining columns compare registration-derived and predicted displacement fields, visualised using displacement vectors and optical flow.