Analysis of radiometric data of MRI images of patients with diffuse midline gliomas with different mutational statuses

Background. Radiomics and radiogenomics are methods of analyzing graphical data that have great potential in the diagnosis of oncological diseases. Aspects of the diagnosis of neoplasms of the central nervous system in children and adults are particularly relevant. In particular, histological verification of brain stem tumors is a huge problem due to their anatomical inaccessibility. At the same time, the histological diagnosis, immunohistochemistry and molecular genetic profile data provide valuable information about the biological properties of the tumor, which are important for determining treatment tactics and the prognosis of the disease. This article is devoted to the problems of the application of radiomics and radiogenomics for diagnosis in pediatric neuro- oncology.Research objectives. To determine the possibility of a connection between the features of the textural characteristics of the tumor image and the presence of a mutation of the H3F3A K27M gene in it. To determine the prognosis of diffuse midline glioma in children using machine learning methods, based on quantitative data obtained as a result of textural image analysis, as well as clinical data from patients.Materials and methods. Textural analysis of MRI studies of 223 children diagnosed with diffuse brainstem tumor. Training of machine learning models in order to solve the problem of classifying patient groups based on the mutation status of the H3F3A K27M gene.Results. The study demonstrated the possibilities of using classical and deep machine learning models in the analysis of radiomic features. At the same time, satisfactory results were achieved in determining the mutation status of H3K27M according to MRI data using deep machine learning technology (neural networks) with the ability to generate synthetic data.