In addition to helping doctors discover and measure tumors, it also helps them develop better recovery and treatment plans. Recent MRI brain tumor segmentation algorithms have focused on U-Net design to combine high-level and low-level features for improved accuracy. Fully convolutional networks, which are also used for this purpose, are unable to successfully reconstruct the image through the decoder path because of the insufficient and low-level information from the encoder path. More effort needs to be done to optimise the low-level information flow from the encoder path to the decoder path in order to improve image reconstruction. In this study, we suggested a transfer learning residual U-Net model that combines the U-Net and VGG-16 architectures. To improve image reconstruction, VGG-16 is combined with the encoder. Additionally, a residual path in skipping connection is included to highlight key feature details while muting noisy and unnecessary feature replies. It is trained using The Cancer Imaging Achieve (TCIA) and Brats 2018 datasets, and It makes it easier to segment small brain tumors. When compared to previous brain tumor segmentation techniques, the suggested model performs competitively. .