Transcranial magnetic stimulation (TMS) is FDA-cleared for the treatment of depression, obsessive compulsive
disorder, and smoking addiction, and there are ongoing clinical trials for new mental health indications. However,
variable response across patients is a significant limitation of TMS. A contributing factor may be a lack of proper
individualization and reproducibility of the TMS targeting to relevant cortical regions. Accurate individualized
targeting requires neuronavigation systems that track the position of the TMS coil relative to the patient's head.
While a small minority of the FDA-cleared TMS treatment devices incorporate neuronavigation, it has significant
drawbacks, including uncomfortable tracking headgear, reduced accuracy due to headgear movement relative
to the head, time-consuming registration, and high cost of dedicated optical or electromagnetic tracking devices.
Novel technologies that could address the limitations of conventional neuronavigation have recently become
feasible, including inexpensive consumer-grade depth cameras and advanced computer vision algorithms
allowing accurate tracking of natural objects such as faces and heads. Our goal is to leverage these advances
to develop an accurate, low-cost, and trackerless system for TMS computer-vision-based neuronavigation
(CVN). Aim 1 is to use consumer-grade depth cameras to detect keypoints on the subject's head comprising
either conventional reflective markers attached to the head or anatomical landmarks for trackerless navigation.
The algorithms will leverage several features of the cameras, including visible light video feed, infrared depth
scanning, and multi-camera synchronization that can be processed together to robustly extract 3D spatial
information. Aim 2 is to localize the head keypoints in 3D space. To this end, CVN will pair two cameras to
acquire visible and infrared light stereo data. The stereo data will be combined with the less accurate raw depth
information provided by each camera to localize the keypoints in 3D space. Aim 3 is to track the position of the
subject's head relative to the TMS coil. Combining the sparse keypoints, the less accurate but dense surface
information generated by each camera, and multi-frame temporal information, CVN will automatically register
the head position to an MRI-based individual head model or, if one is unavailable, a personalized head template
from a model library. The head position will be computed relative to the TMS coil, which will be tracked with the
same methods and permanently mounted reflective markers. We will fine tune the head tracking algorithms with
data from a diverse sample of human subjects, and the complete CVN system will be tested and compared to a
conventional neuronavigation device both with bench-top measurements and in a study of healthy volunteers to
determine accuracy and reproducibility. Overall, the proposed neuronavigation technology could synergize with
current trends toward fMRI-based personalization of TMS targeting to enable more precise and efficacious
interventions for mental health disorders.