BACKGROUND
Transcranial magnetic stimulation (TMS) is a widely used technique in neuroscience research and clinical practice. It allows scientists and doctors to stimulate the human brain in a non-invasive way. “Non-invasive” means that the brain can be stimulated without surgery and without direct contact with the skin. Because of this, TMS is considered a safe and well-tolerated method to study brain function and to treat certain neurological and psychiatric conditions.
In many TMS applications, an important step is called motor hotspot identification (often referred to as “motor hotspot hunting”). This involves finding the exact position on the scalp where stimulation reliably produces a small muscle response in a hand muscle.
For many years, researchers have known that not only the position of the stimulation coil matters, but also its orientation (the angle at which it is held). Even small changes in position or angle can influence the strength of the muscle response. But how can we find the optimal position and orientation in a way that is both efficient and precise?
In everyday practice, clinicians and researchers often adjust the coil manually, testing different combinations of position and angle. Together, these combinations form what is known as a parameter space. Even within a small brain region, this space can be very large. As a result, traditional manual hotspot identification is often time-consuming, subjective, and imprecise.
Mathematical approaches known as acquisition functions are specifically designed to search large parameter spaces efficiently. These methods can help identify the optimal stimulation settings more quickly, more objectively, and with greater precision.
OUR FINDINGS
In a study published in PLOS Computational Biology, David Luis Schultheiss and Joschka Bödecker from the Department of Computer Science, together with Zsolt Turi and Andreas Vlachos from the Department of Neuroanatomy, systematically compared five state-of-the-art acquisition functions to evaluate how efficiently and precisely they identify the optimal stimulation settings. The team also introduced a novel approach for motor hotspot hunting. Their method uses a mathematical framework called Gaussian process–based Bayesian optimization, which is designed to search large parameter spaces in an intelligent and efficient way. Unlike previous approaches, their newly developed method can simultaneously optimize both the coil location and its orientation during hotspot identification. This allows for a faster, more precise, and more automated procedure compared to traditional manual methods.
KEY FINDINGS
- Simultaneous optimization of coil position and angle significantly improved both the speed and the spatial precision of motor hotspot hunting compared to conventional manual approaches.
- The different acquisition functions showed clear differences in how they explored the search space and how well they performed.
- Strategies that focused too strongly on “exploitation” (repeatedly testing what already seemed best) often got stuck in suboptimal solutions and stopped improving too early.
- More balanced strategies that combined exploration and exploitation—especially Thompson sampling—consistently produced the best results.
- Artificially restricting the possible coil rotation angles reduced flexibility and did not account for individual differences in brain anatomy, leading to less optimal outcomes.
Overall, the results demonstrate that Bayesian optimization using Gaussian processes can substantially improve TMS motor hotspot identification. By efficiently navigating the complex combination of coil position and orientation, this approach makes the procedure faster, more precise, and less demanding for participants.
The study also highlights how important it is to carefully choose the right acquisition strategy and avoid unnecessary restrictions on coil orientation. By improving precision and reducing testing time, this method may enhance safety, reproducibility, and overall effectiveness in both research and clinical TMS applications—potentially extending beyond the motor cortex to other brain regions as well.
Schultheiss DL, Turi Z, Boedecker J, Vlachos A (2025) Efficient Gaussian process-based motor hotspot hunting with concurrent optimization of TMS coil location and orientation. PLoS Comput Biol. doi: 10.1371/journal.pcbi.1013994.
The original article can be found here.
