The article provides a comprehensive analysis of the digital transformation of diagnostic processes in modern healthcare, the information and communication technology (ICT) infrastructure in this field, and the application possibilities of artificial neural networks. The relevance of the research stems from the necessity of analyzing large volumes of visual and digital data generated daily in medical institutions with high accuracy. The main objective of the work is to identify optimal mathematical models that ensure the integration of clinical diagnostics with modern technological solutions. The paper examines technical procedures such as pre-processing of medical images, noise reduction and segmentation, as well as the role of Convolutional Neural Networks in diagnostic accuracy. During the research, mathematical metrics such as sensitivity and specificity were used as the basis for evaluating the effectiveness of neural networks, and the system's ability to minimize false negative results was particularly emphasized. Applying an interdisciplinary approach, the prospects of artificial intelligence in early detection of diseases in fields such as radiology, oncology and ophthalmology are demonstrated. In conclusion, it is substantiated that intelligent systems are not intended to replace physicians, but serve as an effective "second opinion" tool that reduces their workload by 30-40% and minimizes subjective errors in decision-making. The multimodal analysis capabilities of modern ICT infrastructure pave the way for the creation of a "unified digital patient profile" and the development of personalized medicine in the future.

Background/Objectives: We compared statistical reporting and interpretation in original quantitative articles from 10 clinical neurology journals in 2015 and 2025. Methods: We audited 200 articles (10 per journal-year) drawn from nonexhaustive pools, assessing 13 reporting and five interpretation/consistency outcomes. Uncorrected journal-stratified Mantel–Haenszel odds ratios (ORs), Robins–Breslow–Greenland 95% confidence intervals, and Cochran–Mantel–Haenszel P values were calculated; Benjamini–Hochberg correction covered 18 comparisons. Two external assessors independently rated 20 articles for reliability; another pair independently reassessed 33 targeted items, with three third-adjudicator decisions and 11 final code changes. Complete full text was unverifiable for 74 articles; 23 historical discrepancies remained unresolved. Results: No primary comparison met the BH-FDR threshold (minimum q = 0.0532). Adequate effect-estimate reporting was 63.0% versus 75.5% (OR 1.95); uncertainty reporting, 49.0% versus 63.0% (OR 1.94); and exact P-value reporting, 68.5% versus 85.3% (OR 2.55), respectively, for 2015 and 2025. Trend language was nominally less frequent (OR 0.27) and causal overstatement more frequent (OR 2.69) in 2025. A post hoc historical stratum-wise 0.5 correction also yielded no FDR discoveries (minimum q = 0.0748). Conclusions: Observed differences remain exploratory and contingent on sampling, source accessibility, and coding reproducibility; they do not establish a field-wide temporal trend.