Millimetres Matter - Improving registration of DBS MRI using deep learning

Motivation: Precise electrode localisation in DBS surgery determines success or failure of neurostimulation and associated side-effects. Brainshift and electrode artefacts in post-op MRI complicate registration to pre-op data, impacting the study of immediate and longitudinal clinical outcomes. Goal(s): To develop a new DBS MRI registration framework using deep learning and advanced image processing to overcome limitations of current approaches and improve registration. Approach: Post-op MRI is preprocessed to ameliorate artefacts and an artefact-free image is synthesised using deep learning and super resolution, followed by optimised non-linear registration. Results: Proposed method demonstrably outperforms standard approaches, reducing errors near electrodes and improved matching of brain regions. Impact: This work transforms DBS neuroimage processing offering a means for much improved electrode localisation and assessment of DBS outcomes. It's a promising step towards improved patient care and clinical success. Public availability of our tools will benefit the neuroimaging community.