Computational Tools for Evaluation and Programing of Deep Brain Stimulation

Deep brain stimulation (DBS) is a surgical therapy to alleviate symptoms of numerous
movement and psychiatric disorders by electrical stimulation of specific neural tissues
via implanted electrodes. Precise electrode implantation is important to target the
right brain area. After the surgery, DBS parameters, including stimulation amplitude,
frequency, pulse width, and selection of electrode’s active contacts, are adjusted during
programming sessions. Programming sessions are normally done by trial and error.
Thus, they can be long and tiring.
The main goal of the thesis is to make the post-operative experience, particularly
the programming session, easier and faster by using visual aids to create a virtual
reconstruction of the patient’s case. This enables in silico testing of different scenarios
before applying them to the patient.
A quick and easy-to-use deep learning-based tool for deep brain structure segmentation
is developed with 89% ± 3 accuracy (DBSegment). It is much easier to implement
compared to widely-used registration-based methods, as it requires less dependencies
and no parameter tuning. Therefore, it is much more practical. Moreover, it segments
40 times faster than the registration-based method. This method is combined with an
electrode localization method to reconstruct patients’ cases.
Additionally, we developed a tool that simulates DBS-induced electric field distributions
in less than a seconds (FastField). This is 1000 times faster than standard methods
based on finite elements, with nearly the same performance (92%). The speed of the
electric field simulation is particularly important for DBS parameter initialization, which
we initialize by solving an optimization problem (OptimDBS). A grid search method
confirms that our novel approach convergences to the global minimum. Finally, all the
developed methods are tested on clinical data to ensure their applicability.
In conclusion, this thesis develops various novel user-friendly tools, enabling efficient
and accurate DBS reconstruction and parameter initialization. The methods are by far the quickest among open-source tools. They are easy to use and publicly available,
FastField within the LeadDBS toolbox, and DBSegment as a Python pip package and a
Docker image. We hope they can boost the DBS post-operative experience, maximize
the therapy’s efficacy, and ameliorate DBS research.