In this paper, we introduce the new hybrid architecture based on ResNet and UNet for detecting brain tumors- the standard network for segmentation recently, with much higher sensitivity in detection. ResNet employs deep architecture to facilitate feature extraction of the brain MRI images. These are finally passed on to the UNet architecture, specially designed for the task of semantic segmentation, which would therefore correctly outline the regions of tumors. It is able to differentiate between glioma, meningioma, pituitary tumor and no tumor, which therefore would enhance the reliability of both classification and segmentation. As experimentally proven, the proposed model is efficient in type tumor detection and accurately segmented the area of the tumor along with the ability to surpass the performance and accuracy acquired by the traditional deep learning techniques. This will ultimately help clinicians with a treatment process of personalization and early diagnosis.