POSSIBILITIES OF USING NEURAL NETWORKS FOR THE NEEDS OF PATHOMORPHOLOGY

In the last decade, digital pathomorphology has been transformed from a simple storage tool into a platform for comprehensive analysis, with an important contribution from artificial intelligence-based tools.The switch from individual glass slides to complete digitization of whole-slide images (WSI) has enabled scalable, reproducible, and standardized morphological evaluation that can serve as a basis for incorporating machine learning into routine diagnostic and prognostic work.The purpose of this narrative review was to synthesize and structure the available evidence regarding the use of artificial neural networks in pathomorphology, including their diagnostic and prognostic utility, advantages, limitations, and prospects for application.Original papers, reviews, and meta-analyses on neural network-based analysis of histological and cytological images, biomarker assessment, risk stratification, and outcome prediction were searched in PubMed, Scopus, Web of Science, and Google Scholar (2010-2025) according to predefined inclusion and exclusion criteria.The reviewed literature demonstrates that deep learning models can reach or even exceed expert level in tumour grading, prediction of tumour microenvironment characteristics and mutations, as well as personalized risk estimates across a range of cancer types.Pretrained foundation models on large, clinically curated image collections further expand these capabilities, allowing efficient reuse for diverse diagnostic and prognostic tasks with minimal retraining.Successful translational use cases typically integrate neural networks into well-defined clinical workflows, combine whole-slide image analysis with clinical, genomic, and radiological information, and position algorithms as decision-support tools rather than autonomous systems.Nevertheless, there remains considerable heterogeneity in datasets, study designs, and validation approaches across the field, alongside issues of explainability, regulatory approval, and unequal access to digital pathology infrastructure, all of which continue to limit widespread routine use.Neural networks may facilitate not only diagnosis, but also standardized quantification of tissue changes and optimization of Журнал «Перспективи та інновації науки» № 12(58) 2025 1980 treatment decisions in liver and kidney pathology, as well as in bladder, breast, lung, gastric, and colorectal cancers.In summary, neural networks should be considered an evolving adjunct to expert pathologists within multimodal integrated diagnostic environments, whose safe and effective implementation requires rigorous validation, standardized data and workflows, and sustained interdisciplinary collaboration.