Аннотация
Neural networks demonstrate exceptional efficiency in processing and analyzing big data. In medicine, these advantages open up great opportunities for early diagnosis. However, their implementation is limited by the “black box” problem: doctors cannot trust a diagnosis without understanding the logic of decision-making. To eliminate this barrier, Explainable Artificial Intelligence (XAI) is used. These methods not only indicate the signs on the basis of which the forecast was made, but can also help identify previously unknown correlations. We systematically screen and then critically analyze recent studies XAI methods applied to Alzheimer's disease screening and diagnosis that integrate machine learning and deep learning models with post-hoc and embedded explainability techniques across multiple data modalities. The review highlights how XAI methods contribute to biomarker identification, model transparency, and clinical trust, while also exposing persistent challenges related to data bias, computational complexity, and limited clinical validation.