Evaluating Applicability of Machine Learning Models in Navigated Transcranial Magnetic Stimulation Systems

The article describes the results of assessing the possibility of using machine learning models as part of robotic complexes for navigational transcranial stimulation (nTMS). It is assumed that they will be used to maintain a given position of the radiation source above the patient’s head. To track its movement, the applicability of models was assessed to detect the patient’s head and then obtain information about the angles of its rotation relative to the camera of the technical vision system. This information is used to recalculate the coordinates to which the radiation source must be moved to affect a given area. The use of machine learning models in this case can reduce the time of image processing and software complexity. The work was carried out using Wider Face and Youtube Face datasets and images obtained using a Vzence DCAM710 ToF sensor. At the same time, the compliance of the model results with annotated values from data sets (head detection) and three-dimensional coordinates of the irradiation point on the human head when working with real images was assessed. The three-dimensional coordinates of the point were calculated using the rotation angles obtained from the models. Research results have shown that this approach can provide a manipulator positioning accuracy of up to 3 cm, which is not enough for nTMS systems.