The segmentation of medical images is becoming increasingly important in the daily clinical practice, particularly with regard to tumor segmentation. Artificial intelligence, which is becoming more and more widespread, is already able to achieve good results in segmentation, making the task of medical staff easier. The U-net is a widely popular convolutional neural network designed to be involved in image segmentation. However, when training the U-net, we may encounter difficulty not having enough memory due to the large size of the images. The main goal of this paper is to investigate the effects of reducing the resolution of the image data using a resizing layer on the input of the U-net and then convert it back at the output also using a resizing layer. The main question is whether we are able to achieve fine segmentation quality. Also it is investigated, what resolution is needed to achieve acceptable segmentation on the dataset. As a result, we managed to achieve an F1-score of 96.9%. Furthermore, we have demonstrated that it is possible to segment the images well with U-net network at lower resolution. Also, it is shown that neither too large nor too small resolution supports good-quality segmentation.