Development of Neural Network Models Using Explorative Artificial Intelligence Methods for Analyzing and Classifying Diseases in the Iris

The article presents a reviewmethodological framework for the automated analysis of iris images using neural network models and explainable artificial intelligence (XAI) techniques. The reproducible pipeline components are described: data sources and imaging requirements; preprocessing and segmentation of iris/pupil (exclusion of noniris regions, unwrapping, normalization); selection of architectures (pretrained CNNs, CNN+ViT hybrids) and training on limited datasets. The role of XAI as a quality component is highlighted: GradCAM/ GradCAM++, Integrated Gradients, SHAP, EnsembleCAM; quantitative validation (IoU/pointinggame, Deletion/Insertion, Infidelity) and probability calibration (ECE, Brier). Clinical scenarios are discussed (localized iris pathologies; to a limited extent – systemic correlations), as well as limitations of evidence and privacy requirements for biometric data. A «practical minimum» for prototyping and external validation is formulated, with an emphasis on stratified domainspecific reporting. Future directions are outlined: standardized datasets, open benchmarks, multicenter validation, and domain shift monitoring.