Efficient Semantic Segmentation of Nuclei in Histopathology Images Using Segformer

Segmentation of nuclei in histopathology images with high accuracy is crucial for the diagnosis and prognosis of cancer and other diseases. Using Artificial Intelligence (AI) in the segmentation process enables pathologists to identify and study the unique properties of individual cells, which can reveal important information about the disease, its stage, and the best treatment approach. By using AI-powered automatic segmentation, this process can be significantly improved in terms of efficiency and accuracy, resulting in faster and more precise diagnoses. Ultimately, this can potentially lead to better patient outcomes, making it a vital tool for healthcare professionals. In this paper, a novel method is proposed for semantic segmentation of nuclei using Segformer-b0 and Segformer-b4 on the PanNuke dataset. To achieve the most efficient and accurate segmentation, Segformer architecture is used as it combines the advantages of transformers and convolutional neural networks. To evaluate the performance of the models, dice evaluation metric is used. The proposed method achieved state-of-the-art results on the PanNuke dataset, with Segformer-b4 achieving a mean dice score of 0.845, and Segformer-b0 achieving a mean dice score of 0.82. The findings demonstrate the high efficiency of the Segformer architecture for semantic segmentation in histopathology images, as it also highlights the importance of choosing the appropriate model architecture for the task at hand. The proposed method provides a promising approach for accurate and efficient segmentation of histopathology images. The proposed approach outperformed state of the art methods by achieving better results, as it achieved 85.45% average Dice score which is more accurate compared to the RCSAU-Net which achieved 84.82% average Dice Score. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2024.