In recent times, neural networks using convolutions have introduced the narcotic methods of segmenting a 3D image, which depend on representation of huge features and accomplish decent execution. This is because convolutional neural networks have limited regions in sensory periphery which cannot absolutely replicate the dependencies which are from long ranges present in the image. Newly, Transformer took advantage of worldwide reliant using some automatic-checking techniques and learning of different interpretations which are high. Some experiments were according to transformers, yet current transformers experience severe arithmetical along with memory problems and are unable to fully utilize powerful properties for segmentation of 3D medical pictures.As in the paper, we focus on the parallel combination of various resolution channels, and propose a unique system called High Definition Transformer Based Network using a efficient transformer block that has an adequate depicting attribute even at high attribute resolution. When a 3Dimensional picture is given, first encoder uses neural networks to takeout object manifestations that collect regional data, and intelligently constructs a variety of feature maps for tokens, which are fed in parallel to each Transformer channel, learns global information and iteratively information is shared. Unfortunately channels, the presented standard transformer-based framework requires a substantial computing, so we start a significant and efficient transformer that provide improved performance with fewer variables. The proposed Transformer based high resolution network is evaluated on a brain tumor task dataset.