Answers are grounded in this paper and cite their sources.
Relevance. The relevance of this study stems from the active implementation of artificial intelligence (AI) technologies in diagnostic imaging and the need for an objective assessment of their effectiveness in real-world clinical practice. For the Khabarovsk Territory – characterized by vast geographical expanse and uneven staffing in the field of diagnostic imaging – the use of AI offers the potential to improve the accessibility and speed of medical image analysis. AI effectiveness depends not only on the diagnostic capabilities of the algorithms but also on technical factors: data transmission stability, error rates, and the proportion of successfully processed examinations. Analyzing the volume, structure, and technical performance of AI usage is essential for improving the regional diagnostic imaging system. Objective. To evaluate the volume, structure, and technical performance of AI-based software used for medical image analysis in the Khabarovsk Territory between January and August 2026. Materials and Methods. A retrospective descriptive study was conducted using aggregated operational data from the regional diagnostic imaging network. The study analyzed the results of automated processing for chest X-rays, chest fluorography, mammography, and computed tomography (CT) scans of the chest and brain. The primary metrics were the number of examinations submitted and successfully processed, the rate of technically successful processing, and the distribution of automated findings. Differences in successful processing rates were assessed using Pearson’s chi-squared test, with a critical significance level of p < 0.05. Results. A total of 195,599 examinations were submitted for automated analysis, with 181,638 (92.9 %) successfully processed. Technical success rates varied by modality (χ² = 6218.7; degrees of freedom = 4; p < 0.001), ranging from 82.2 % for brain CT scans to 97.6 % for chest X-rays. For mammography, the median automated processing time was 23 seconds, with a 95th percentile of 51 seconds. Automated pathology markers represented the algorithm's output and were not considered confirmed diagnoses. Conclusion. Regional AI implementation covered a substantial volume of examinations; however, technical success rates depended on the modality. Key areas for monitoring include the reasons for processing failures in brain CT scans, the completeness of source data, and the comparison of algorithm results with radiologists' reports.
No takes yet. Share an insight, caveat, or question.
Sitnikov et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: