Alzheimer's disease represents not only a clinical but also a systemic healthcare challenge, associated with an exponential increase in the number of patients, limited diagnostic resources, and the high cost of late detection. In overloaded clinical workflows, the key task becomes not maximizing classification accuracy, but minimizing the risk of missing characteristic structural brain changes in the early stages of the disease. This work proposes a screening-oriented three-dimensional convolutional neural network designed for primary filtering of structural MRI studies to identify neurodegenerative patterns, with an explicit prioritization of sensitivity. The model is considered a clinical decision support tool that redirects specialists' attention towards patients with the highest risk of structural changes associated with neurodegeneration. The proposed architecture is a lightweight 3D CNN with progressive regularization and a weighted loss function reflecting the asymmetric cost of classification errors. Experimental evaluation on a held-out test set of 162 subjects demonstrates an AUC-ROC of 0.824 and a sensitivity of 0.943, missing only 3 out of 53 cases with structural changes characteristic of Alzheimer's disease. The obtained results confirm the applicability of the model as a first-level automated screening tool for MRI data, aimed at reducing the systemic risk of missing structural biomarkers of neurodegeneration and optimizing the use of clinical resources.
Georgii Erokhin (Mon,) studied this question.