A recent study accepted by ISBI 2025 introduces a novel model that improves cancer grade classification and gland segmentation. The model uses a Vision Transformer (ViT) for classification and a modified Segment Anything Model (SAM) for segmentation. By incorporating prior knowledge of gland benignity or malignancy, the model makes accurate predictions for segmentation. A heat map generated by Grad-CAM++ guides the segmentation process, and a prompt adapter integrates this information effectively. The model's performance was tested on the Gland Segmentation Challenge (GlaS) dataset, showing significant improvements across various scales and achieving top results compared to benchmarks. An arxiv version of the paper is available at https://arxiv.org/pdf/2501.14718Link opens in a new window