<p>Segmentation of medical images is relevant to medical diagnosis, and deep convolutional neural networks (CNNs) have achieved a lot in this field. Nevertheless, CNNs have failed because they primarily concentrate on localized features. On the contrary, Transformer architecture has the capability of looking at the whole sequence of input and thus can capture global contextual information of medical images better. In this work, we propose a new method to improve the detailed image information by applying three methods: namely, contrast-constrained adaptive histogram equalization (CLAHE), modified binary Otsu-based histogram equalization (MBOBHE), and modified partitioned histogram equalization (MPHE), and secondly, U-Net with ViT Transformer framework is used to process MRI brain tumor medical images further. Three methods are proposed in this study, the preprocessing method using CLAHE is called EF-UVit1, the method using MBOBHE is called EF-UVit2, and the method using MPHE is called EF-UVit3. EF refers to enhanced features, and U and Vit refer to U-Net and ViT Transformer, respectively. EF-UVit refers to the model of fusion of U-Net and ViT Transformer with feature enhancement of input image for MRI brain tumor medical image segmentation. This study achieves particularly outstanding results in the recognized dataset Brats2020, the results of the evaluation metrics of the three methods proposed in this study are all more than 98%, and the results in the MSD dataset, although slightly inferior to the Brats2020 dataset, still compare favorably with other algorithms.</p>