ИСТИНА

Classifying tissue types in whole slide histological images (WSIs) is a critical task in digital pathology, with significant potential to enhance diagnostic accuracy. While traditional patch-based ap- proaches using convolutional neural networks (CNNs) have shown success, they often struggle to capture the spatial and topological relationships inherent in histological structures. To address these limitations, we propose a novel graph-based method leveraging graph convolutional neural networks (GCNNs) for tissue type clas- sification in WSIs. In our approach, a graph is constructed where nodes represent image patches and edges encode spatial relation- ships, allowing the network to effectively aggregate information from neighboring patches. Additionally, we introduce a new sam- pling strategy that selectively builds graphs from annotated regions, improving the model’s ability to utilize spatial dependencies. Evalu- ations on the free publicly available PATH-DT-MSU dataset reveal that our method achieves a significant improvement in Macro F1- score compared to state-of-the-art patch-based and graph-based methods. These results demonstrate that the proposed GCNN model not only enhances classification accuracy but also provides a more robust framework for addressing the complex spatial structure of histological images.