Algorithm for Processing Medical Images to Create a Dataset while Maintaining the Size of Structural Elements

Currently, there is an acute problem of segmentation of medical images and the allocation of structures on them for their further use for practical purposes. Thus, there is a problem of selection and segmentation of different areas in the human body based on MRI or CT images for further use of these data in the construction of 3D models. In turn, these models may not be the final product, but can already be used in the construction of mathematical models that describe various processes of interest to us in this area. Thus, the use of brain regions reconstructed from MRI images allows one to obtain a personalized distribution of electromagnetic field parameters in the brain under the influence of transcranial magnetic stimulation, which can greatly facilitate the task of selecting parameters and the area of influence. Since the selection of these areas manually is a rather laborious and expensive process, there is a tendency to create neural networks and train them for this purpose. However, the files that come to the input of the neural network for its training and use must be of the same size. In the case of using known operations to solve this problem, structural elements can change not only their size, but also their relative position, which is critical for the purposes of the study. In our work, we propose an algorithm for processing medical data to create a dataset while maintaining the size of structural elements.