Assessing the Accuracy of Deep Brain Stimulation and Responsive Neurostimulation Electrode Localization Using Synthetic MRI Scans

Introduction: MRI is essential for planning and localizing electrodes in deep brain stimulation (DBS) and responsive neurostimulation (RNS). However, many patients are ineligible for MRI due to metallic implants or limited access. This study assesses the accuracy of pre-operative electrode planning for DBS and RNS using SynthSR, a convolutional neural network converting CT scans to T1 MRI.

Methods: We acquired pre- and post-implantation imaging for all patients, including pre-op CT, pre-op MRI, and post-op CT with electrodes implanted. SynthSR created a synthetic MRI from the pre-op CT. The post-op CT was co-registered to the synthetic pre-op MRI and real pre-op MRI. Electrodes were localized using Lead-DBS (for depth electrodes) and LeGUI (for cortical strip electrodes) software. Electrode positions were projected to Montreal Neurological Institute (MNI) template space by calculating non-linear warps using the pre-op real or synthetic MRI. Differences in localizations were compared using Euclidean distances of each contact in 3D space and visualizing anatomical targeting accuracy.

Results: 27 patients undergoing subthalamic nucleus (STN)-DBS for Parkinson’s disease were included, with a mean age of 60.7 ± 6.9 years, and mean disease duration of 10.4 ± 3.9 years. For RNS, 12 patients with drug-resistant epilepsy were included with a mean age of 45.5 ± 16.2 years, mean duration of epilepsy of 17.2 ± 13.7 years, and mean seizure onset age of 28.3 ± 16.6 years. Targets for RNS patients included the hippocampus, centromedian nucleus of thalamus, temporal lobe, amygdala, and anterior nucleus of thalamus. For DBS, average Euclidean distance (in MNI space) between the same electrode contacts localized using synthetic and real MRI was 1.34 ± 0.7 mm (similar to the diameter of the electrode itself). For RNS electrode localization, this distance was 2.99 ± 1.9 mm. For DBS localization, all electrodes localized using synthetic MRI were accurately placed within the STN upon anatomical reconstruction.

Conclusion : We demonstrate potential for pre-operative neurostimulation surgical planning and post-operative evaluation using CT scans and SynthSR-generated synthetic MRIs, addressing cases where MRI may be unavailable or unsuitable.