SAM-guided feature enhancement network for precise gastric pathology segmentation

Gastric image analysis is a crucial step in diagnosing gastric cancer, but achieving automated and precise segmentation remains challenging due to the diversity of cell morphology and boundaries in pathological sections. Inspired by the visual big model, this paper proposes the GastricSAM network for gastric cancer tissue image segmentation, which utilizes the powerful generalization ability of the visual big model to enhance the model's ability to capture complex pathological features. In addition, we have designed dynamic fusion modules, edge enhancement modules, and multi-scale decoding blocks. Extensive experiments have shown that our GastricSAM has excellent performance and achieves state-of-the-art results. Its Dice is 0.813 (5.0% higher than LinkNet) and 0.708 mIoU(6.3% higher than LinkNet).