Background

Drug-resistant focal epilepsy can be caused by a range of structural brain abnormalities, from large tumours to small, difficult-to-detect cortical malformations. Resective brain surgery is a potentially curative treatment, but only leads to seizure freedom in around 70% of patients. Identification of a focal lesion on presurgical MRI, complete resection of the lesion and disconnection of the epileptogenic network have all been identified as significant predictors of seizure freedom. This PhD will combine diffusion-weighted microstructural mapping, tractography and AI to aid surgical planning in patients with focal epilepsy.

Novelty & Importance

To date, epilepsy lesion detection methods have relied solely on structural MRI data (T1, T2), but there is significant evidence that diffusion imaging can help to localise subtle lesions. This project would pioneer the integration of diffusion-weighted brain imaging into AI lesion detection approaches. 

Furthermore, seizures are propagated through white matter tracts, and there is new evidence that alongside surgical resection of the lesion, disconnection of the white matter tracts is an important predictor of post-surgical seizure freedom. This thesis will integrate presurgical tract estimation with post-surgical resection cavities to build predictive models of surgical seizure freedom.

Through our Multicentre Epilepsy Lesion Detection project, we have collected a multimodal MRI dataset of a wide variety of focal epilepsy pathologies for over 2000 patients and 1000 controls. Our previously MELD project epilepsy lesion detection tools (Spitzer et al., Brain, Ripart et al., Annals of Neurology, Ripart et al., JAMA Neurology), have been applied at over 70 epilepsy surgical centres internationally through open source tools (https://github.com/MELDProject/). The improvement of these tools through the novel incorporation of multimodal DWI therefore has the potential to have a widespread and rapid clinical impact.  

Aims and objectives

1.Characterize microstructural abnormalities: Harmonize and process DWI data to identify microstructural changes in focal epilepsy pathologies through MD (mean diffusivity) and FA (fractional anisotropy) maps.

2.Develop AI for lesion detection: Train a multimodal lesion segmentation model integrating DWI microstructural maps with T1-weighted, FLAIR, and T2-weighted MRI sequences to enhance lesion detection.

3.Map connectivity for surgical planning: Use tractography to investigate brain connectivity abnormalities in epilepsy patients and assess how surgical disconnection of tracts impacts outcomes.

4.Predict surgical outcomes: Identify imaging features that predict poor surgical outcomes, enabling development of predictive models for surgery planning.

References

Spitzer, H., Ripart, M., Fawaz, A., Williams, L. Z. J., Robinson, E. C., Iglesias, J. E., Adler, S., & Wagstyl, K. (2023). Robust and Generalisable Segmentation of Subtle Epilepsy-Causing Lesions: A Graph Convolutional Approach. Medical Image Computing and Computer Assisted Intervention – MICCAI 2023, 420–428.

Ripart, M., DeKraker, J., Eriksson, M. H., Piper, R. J., Gopinath, S., Parasuram, H., Mo, J., Likeman, M., Ciobotaru, G., Sequeiros-Peggs, P., Hamandi, K., Xie, H., Cohen, N. T., Su, T.-Y., Kochi, R., Wang, I., Rojas-Costa, G. M., Gálvez, M., Parodi, C., … MELD HS study group. (2024). Automated and Interpretable Detection of Hippocampal Sclerosis in temporal lobe epilepsy: AID-HS. Annals of Neurology. https://doi.org/10.1002/ana.27089

Ripart, M., MELD consortium, Adler, S., & Wagstyl, K. (2024). Multi-pathology MRI lesion segmentation in a multi-centre cohort of patients with focal epilepsy: a MELD study. In https://openreview.net › forumhttps://openreview.net › forum. https://openreview.net/pdf?id=LrihI0cqZm