feature-segmentation-based registration for fast and accurate deep brain stimulation targeting

Objects Deep brain stimulation (DBS) has turned out to be the surgical technique of choice for the treatment of movement disorders, e.g. Parkinson’s disease (PD), the usual target being the subthalamic nucleus (STN). The targeting of such a small structure is crucial for the outcome of the surgery. Unfortunately the STN is in general not easily distinguishable in common medical images. Material and Methods Eight bilaterally implanted PD patients were considered (16 STNs). A three-dimensional MR T1-weighted sequence and inversion recovery T2-weighted coronal slices were acquired pre-operatively. We study the influence on the STN location of several surrounding structures through a proposed methodology for the construction of a ground truth and an original validation scheme that allows evaluating performances of different targeting methods. Results The inter-expert variability in identifying the STN location is 1.61 ± 0.29 mm and 1.40 ± 0.38 mm for expert 1 and 2 respectively while the best choice of features using segmentation-based registration gives an error of 1.55 ± 0.73 mm. Conclusions By registering a binary mask of the third and lateral ventricles of the patient with its corresponding binary mask of the atlas we obtain a fast, automatic and accurate pre-operative targeting comparable to the expert’s variability.