Introduction. The effectiveness of screening as one of the strategies for cancer control is beyond doubt. Screening reduces the risk of diagnosing cancer at a late stage and identifies precancerous pathologies, thereby preventing the development of cancer. Potential limitations of screening include the high probability of false positives, false negatives, and overdiagnosis. The consequences are additional examinations and unnecessary and, often, excessive treatment. At the same time, interval cancers, which are characterized by an aggressive course, often do not come into view.The purpose of the study: to explore the data on effectiveness of artificial intelligence (AI) for improving the sensitivity and specificity of cancer screening and reducing the probability of false negative and false positive results, and overdiagnosis.Materials and methods. Review and analysis of published data on a) screening of breast cancer (BC), lung cancer (LC), prostate cancer (PC), cervical cancer (CC) and large bowel cancer (LBC); b) development and application of AI systems to improve the effectiveness of screening. The PubMed and Cochrane Library databases were searched for relevant publications.Results. In mammography screening, AI reduces the number of abnormal interpretations of mammograms, the number of recalls, the number of biopsies with a negative result, and increases the efficacy of mammogram interpretation regardless of the characteristics of the breast (dense breast, calcifications). The use of AI in conjunction with low-dose computed tomography (LDCT) for LC screening not only improves the diagnosis of various types of LC, but also predicts the risk of developing cancer several years in advance. A systematic review and meta-analysis of 12 studies evaluating the effectiveness of AI in tandem with multiparametric magnetic resonance imaging (mpMRI) of the prostate showed high overall effectiveness in the diagnosis of clinically significant PC. The performance of the AI system – based on the multimodal data including demographics, clinical characteristics, laboratory tests and ultrasound reports of patients with PC, was better than the effectiveness of PSA tests in diagnosing clinically significant PC. The effectiveness of AI in tandem with colonoscopy, despite the use of the most advanced AI systems (deep learning system based on a convolutional neural network), remains controversial. The solution to this problem depends on what goal we are pursuing when developing and training the system? Increasing “detection rate” of adenomas, regardless of their size, and removing them, or identifying and removing only large adenomas? The successful use of AI for cytological diagnosis of cervical pathology, including all stages of cervical intraepithelial neoplasia (CIN), is encouraging. The introduction of AI systems developed and trained to interact with a cytopathologist in reading and evaluating cytological material and diagnosing CIN and CC into general practice will reduce the burden on cytopahologists and other medical personnel.Conclusion. The analysis of published data has shown the promising results concerning the use of AI for cancer diagnostics, especially in the setting of population screening programs, which cover many thousands of people. The use of AI significantly increases the effectiveness of diagnostic tool, improves its sensitivity and specificity, and reduces the probability of false negative, false positive results and overdiagnosis. The decision to introduce into practice any of the AIs with proven effectiveness in clinical trials should be made only after its testing in a real world, at the population level. The “informed consent” forms that objectively describe all the advantages and disadvantages of the use of AI compared to current practice has to be developed.
