Paper Title
Automatic Segmentation of the Brain Tumors by Results of Radiation Diagnostics using a Modified Neural Network UNET

Abstract
In this paper described using Convolution Neural Network UNET with specification of boundaries. The modelhastrained for segmentation and detection tasksof glial brain tumors. Mixed datasetconsists from BraTS’2019 and own data has used for training.The article also highlights the data pre-processing and formation of data.The following metrics of the quality of neural networks were obtained: Sørensen–Dice coefficient= 0.708,Sensitivity = 0.937; Specificity = 0.903. Final model realization will be presented in medicine diagnostic software by authors and “NeuroSoft Diagnostic” LLC. Keywords - Convolutional Neural Networks, UNET, Border Refinement, Dicom Images, Image Segmentation, Glial Tumors, Oncology, Radiation Diagnostics, Brats Dataset, Machine Learning, UNET Network, Image Recognition, Computer Vision, Data Pre-Processing.