Brain tumor segmentation from MRI data is a vital task in medical image analysis that involves identifying and delineating regions of interest within brain images that correspond to tumor presence. This paper proposes a solution to the segmentation problem of tumor parts from multi-spectral MRI records, which combines a 2D convolution U-net architecture adapted to work with spatial features with a spatial histogram enhancement method that aims to improve the visibility of brain structures and lesions in the observed volume. The proposed method was trained and tested using the BraTS 2019 high-grade glioma data set, and evaluated using statistical accuracy benchmarks. The segmentation outcome is globally characterized by average Dice scores of 0.7368, 0.8005, 0.7912, and 0.8612, that were obtained in case of edema, enhanced core, tumor core, and whole tumor regions, respectively.