Addressing correct polyp regions in colonoscopy images to support computer-aided clinical systems has recently gained significant research attention. A key problem is making polyp segmentation methods robust against complex textures, background/foreground transitions, and size diversity. Designing an accurate model to identify polyps remains a challenge. To tackle these difficulties, this paper proposes a novel end-to-end model called MSCR \(^{**}\) for an exact segmentation in endoscopic images. The MSCR \(^{**}\) model takes advantage of the Mix Transformer architecture in an encoder module; a new parallel spatial/deep decoder architecture called SpaDeep-CBAM is given; the channel-wise feature pyramid module (CFP) and reverse attention (RA) module are combined into a refine module, and then to adapt with the SpaDeep-CBAM module. The proposed model is assessed on the Kvasir, CVC-ClinicDB, CVC-ColonDB, EndoScene, and ETIS-Larib Polyp datasets. The results are evaluated with the state-of-the-art models for polyp segmentation. Our model performs better than those on the Kvasir, CVC-ColonDB, ETIS-Larib Polyp datasets, and it is competitive with others on the remaining datasets.