A robotic assistance system for transcranial magnetic stimulation and its application to motor cortex mapping

To overcome the problems of conventional image guided transcranial magnetic stimulation (TMS), we developed a robotic system to place the coil at a target position and to keep it in place even if the head of the patient moves. The system is able to position the coil tangentially at any point on the head with any predefined rotation about the coil's vertical axis. Motions of the head are detected using a tracking system and compensated by steering the robot to the updated target position. The introduction of a robot for TMS coil positioning does not only solve the problems of exact targeting and motion compensation, it also enables a number of novel features: Firstly, it allows for precise continuation of the stimulation in consecutive sessions. Secondly, it makes it simple to stimulate in a predefined grid pattern. Thirdly, it separates treatment planning from treatment execution as it is known from radiotherapy and radiosurgery. Finally, it might offer new possibilities of treatment, e.g. by controlled coil movement over the scalp during repetitive stimulation. The system's main components are a serial robot arm with six joints and a marker-based optical tracking system. Both devices are linked by a control computer, which converts the target coordinates from the medical image data into real world coordinates and commands robot joint settings to place the TMS coil there. Hereby, real-time tracking data from the camera system is used to determine the current position of the head. Two major challenges had to be overcome for robot aided TMS. Firstly, we developed a new way to deal with robot redundancy, i.e. an algorithm how to chose a configuration from the set of possible joint settings encoding the same pose of the TMS coil. When we command the robot to a new position, we first calculate all possible target joint settings and choose the one closest to the actual joint setting according to the Euclidean metric in joint space. The second challenge was to define a safe, robust, and fast heuristic for the trajectory fromthe actual position to the target position for the coil. Our solution is to combine a foolproof trajectory for the centre of the coil -- namely a circular motion with some distance from the head from the start to the target spot -- with an optimised orientation change along the way. The resulting heuristic is safe, because of the simple trajectory of the coil's centre. It is also robust, because the freedom in orientation allows to find a realisable coil path for most of the start and target points around the head. The heuristic can be implemented in a fast way; it usually needs about 100 ms to compute a trajectory. Evaluation of the system yielded a mean accuracy of the system of about 1 mm. The motion compensation module was evaluated to work with a latency of about 100 ms, i.e. it takes about 0.1 seconds before the robot starts to compensate for a head movement. Our experiments with 20 subjects show that these data allow for successful stimulation in the areas of tinnitus treatment, treatment of chronic pain, and motor cortex mapping. One major application of our robot aided TMS system has been the precise localisation of cortical representations of muscles. The use of the robot for the procedure has two advantages compared to manual (image guided) TMS: it makes it easy to define and approach target points, e.g. along a grid-like pattern, and it keeps the TMS coil at the target position, even if the head moves, so that several MEPs can be averaged. The second point is of special importance as TMS evoked MEPs have been shown to possess great variability. Thus, averaging strongly improves the mapping results. We approach the problem of motor cortex mapping from a biophysical point of view. Our model assumes a monotonic functional relationship between the electric field strength at a cortical representation of a muscle and its MEP. Thus, we find the cortical representation by identifying the point on the cortex where such a relationship is most likely given the experimental data. In detail, for a point p on the cortex we approximate the electric field strength Ep for each stimulation point (coil position) i. We then estimate the likelihood of a monotonic functional relationship by looking at the list of pairs (Ep(i);MEP(i)). We show in this thesis that if we use Kendall's tau as an estimator for the monotonic relationship, we get a clear maximum site for the cortical representation. Furthermore, applied to a mapping with a figure-of-eight coil the predicted representation agrees with data from fMRI within the resolution of the fMRI data (4 mm). For the first time, our method allows mapping with non-focal coils like circular coils. ...