Brain tumors was a very crucial disease and the important part is to segmenting the tumor regions of brain. It is very difficult for radiologist to identify the exact the regions of brain tumor and segmenting them. It will assist MRI labs and making correct diagnosis of patient, the required treatment and maintaining a surveillance of the alterations of the illness. Because of the excision tumor region boundary demarcation is a colossal nightmare, it is time consuming, individuals end up differing and due to the intermittency, hence our encouragement to automate this procedure. Although, recent medical imaging caused to tremble because of deep-learning specially CNNs. They are confined to local so it is called receptive fields, it will not see the big picture which is very important with fuzzy or diffusing tumors. For solving these issues we have introduced a CNN-Transformer hybrid model which takes CNNs layers and transformer encoders. In which 3 CNN layers and two transformer layers, CNNs layers help to extract fine-grained and hierarchical features out of every 2D slice and then the image goes to transformer layers which uses multi-head self attention to extract long range spatial attachment. At last, a decoder makes it up-sampling using transposed convolutional and slices retrieve again the high-resolution segmentation masks with close boundary reconstruction.