{"id":34336,"date":"2026-10-23T15:30:14","date_gmt":"2026-10-23T12:30:14","guid":{"rendered":"https:\/\/ukrmedia.news\/?p=34336"},"modified":"2026-10-23T15:36:33","modified_gmt":"2026-10-23T12:36:33","slug":"microsoft-shi-diahnostyka-khvorob-likari","status":"publish","type":"post","link":"https:\/\/ukrmedia.news\/en\/science-tech\/microsoft-shi-diahnostyka-khvorob-likari\/","title":{"rendered":"Microsoft\u2019s AI outperformed doctors: 83% accuracy versus 36% in a study"},"content":{"rendered":"<p>In a controlled study, Microsoft\u2019s artificial intelligence demonstrated more than twice the diagnostic accuracy for complex conditions compared to general practitioners. The MAI-DxO system correctly identified the diagnosis in 83.2% of cases, whilst doctors using reference books and the internet achieved a rate of 36.1%. The results were published on 22 October 2026 in the scientific journal *Science*. However, the trials were conducted using historical clinical cases rather than during the treatment of actual patients.<\/p>\n<div class=\"news-summary-box\">\n<p><strong>Briefly about the main points<\/strong><\/p>\n<ul>\n<li>Microsoft has presented the results of its MAI-DxO medical artificial intelligence study.<\/li>\n<li>In cases from A&amp;E departments, the AI\u2019s accuracy was 80.3%, compared with 46.8% for doctors.<\/li>\n<li>The system analyses the symptoms, identifies the necessary tests and progressively refines the diagnosis.<\/li>\n<li>The results were obtained during simulations. The effectiveness of the technology in real hospitals has yet to be tested.<\/li>\n<\/ul>\n<\/div>\n<h2>Microsoft has tested the AI on hundreds of complex medical cases<\/h2>\n<p>Microsoft researchers have tested a new system <strong>MAI-DxO<\/strong> based on 304 complex clinical cases published in <a href=\"https:\/\/www.nejm.org\/\" target=\"_blank\" rel=\"nofollow noopener\">medical journal<\/a> The New England Journal of Medicine. This refers to diseases and conditions whose diagnosis requires a thorough analysis of symptoms, test results and medical history.<\/p>\n<p>Artificial intelligence has determined that <strong>correct diagnosis in 83.2% cases<\/strong>. General practitioners who were able to use medical reference books and search engines gave the correct answer in 36.1% of cases. Without additional resources, their result was 22.3%.<\/p>\n<p>Consequently, under the conditions of this trial, MAI-DxO proved to be approximately 2.3 times more accurate than doctors with access to reference information.<\/p>\n<p>The researchers also tested the system on 325 cases from the MIMIC-IV-ED emergency department database. Here, the artificial intelligence achieved an accuracy of 80.3%, whilst doctors achieved 46.8%.<\/p>\n<h2>How MAI-DxO medical artificial intelligence works<\/h2>\n<p>Unlike systems that receive all the patient\u2019s information at once, MAI-DxO replicates the sequence of a standard diagnostic process.<\/p>\n<p>First, the programme analyses the patient\u2019s initial symptoms and medical history. It then determines which additional questions need to be asked, which tests need to be carried out, and how to interpret the results.<\/p>\n<p>To do this <a href=\"https:\/\/ukrmedia.news\/en\/science-tech\/microsoft-unifies-copilot-versions\/\">Microsoft<\/a> It uses several AI agents that act as virtual medical experts. They suggest possible diagnoses, cross-check each other\u2019s findings and assess the need for further tests.<\/p>\n<p>Their work is coordinated by the Diagnostic Orchestrator system. The main findings of the study were obtained using a model <a href=\"https:\/\/ukrmedia.news\/en\/science-tech\/openai-gpt-5-6-obmezhennyi-zapusk\/\">GPT-5<\/a> from OpenAI.<\/p>\n<p>This approach makes it possible to verify not only the accuracy of the final diagnosis, but also the sequence of decisions that led to it.<\/p>\n<h2>Artificial intelligence could reduce the cost of medical examinations<\/h2>\n<p>In addition to diagnostic accuracy, the researchers assessed the cost of the tests prescribed.<\/p>\n<p>The trial used estimated prices for medical services. The system was designed to determine which tests and procedures were necessary to reach a diagnosis, without prescribing unnecessary tests.<\/p>\n<p>In the cost-constrained configuration, MAI-DxO maintained an accuracy of approximately 81%, whilst the estimated cost of the diagnostic process was less than half that of the comparable options.<\/p>\n<p>At the same time, the cost savings depended on the system\u2019s settings. Without a budget cap, the AI\u2019s costs for examinations could have been higher. Therefore, the results do not imply that the use of this technology will automatically reduce the cost of diagnosis in every hospital.<\/p>\n<h2>Does this mean that AI is already better than doctors?<\/h2>\n<p>Despite the impressive results, the study does not prove that artificial intelligence is superior to doctors in real-world clinical practice.<\/p>\n<p>The tests were carried out using text descriptions of previously documented cases. The system did not examine patients in person and did not make decisions regarding their treatment.<\/p>\n<p>Furthermore, the AI\u2019s results were compared with the work of individual general practitioners, rather than that of narrow-field specialists or multidisciplinary medical teams. It is precisely such teams that are usually called upon to diagnose particularly complex conditions.<\/p>\n<p>The study was funded by Microsoft and carried out by the company\u2019s staff. The findings require further verification in independent clinical settings.<\/p>\n<h2>When the technology begins to be tested on real patients<\/h2>\n<p>The next stage will be a study at the Beth Israel Deaconess Medical Centre in Boston, with which Microsoft has announced a partnership.<\/p>\n<p>The developers plan to test the AI\u2019s capabilities when working with patients, particularly in A&amp;E departments. The system\u2019s results will be compared with independent assessments by doctors.<\/p>\n<p>Microsoft is also collaborating with the Mayo Clinic on the development of artificial intelligence technologies for healthcare.<\/p>\n<p>For the time being, MAI-DxO remains a research system. Its success in test tasks does not yet mean that it is ready to make diagnoses independently or to replace medical consultations.<\/p>","protected":false},"excerpt":{"rendered":"<p>\u0428\u0442\u0443\u0447\u043d\u0438\u0439 \u0456\u043d\u0442\u0435\u043b\u0435\u043a\u0442 Microsoft \u043f\u0440\u043e\u0434\u0435\u043c\u043e\u043d\u0441\u0442\u0440\u0443\u0432\u0430\u0432 \u0431\u0456\u043b\u044c\u0448 \u043d\u0456\u0436 \u0443\u0434\u0432\u0456\u0447\u0456 \u0432\u0438\u0449\u0443 \u0442\u043e\u0447\u043d\u0456\u0441\u0442\u044c \u0434\u0456\u0430\u0433\u043d\u043e\u0441\u0442\u0438\u043a\u0438 \u0441\u043a\u043b\u0430\u0434\u043d\u0438\u0445 \u0437\u0430\u0445\u0432\u043e\u0440\u044e\u0432\u0430\u043d\u044c, \u043d\u0456\u0436 \u043b\u0456\u043a\u0430\u0440\u0456 \u0437\u0430\u0433\u0430\u043b\u044c\u043d\u043e\u0457 \u043f\u0440\u0430\u043a\u0442\u0438\u043a\u0438, \u043f\u0456\u0434 \u0447\u0430\u0441 \u043a\u043e\u043d\u0442\u0440\u043e\u043b\u044c\u043e\u0432\u0430\u043d\u043e\u0433\u043e \u0434\u043e\u0441\u043b\u0456\u0434\u0436\u0435\u043d\u043d\u044f. \u0421\u0438\u0441\u0442\u0435\u043c\u0430 MAI-DxO \u043f\u0440\u0430\u0432\u0438\u043b\u044c\u043d\u043e \u0432\u0438\u0437\u043d\u0430\u0447\u0438\u043b\u0430 \u0434\u0456\u0430\u0433\u043d\u043e\u0437 \u0443 83,2% \u0432\u0438\u043f\u0430\u0434\u043a\u0456\u0432, \u0442\u043e\u0434\u0456 \u044f\u043a \u043c\u0435\u0434\u0438\u043a\u0438, \u044f\u043a\u0456 \u043a\u043e\u0440\u0438\u0441\u0442\u0443\u0432\u0430\u043b\u0438\u0441\u044f \u0434\u043e\u0432\u0456\u0434\u043d\u0438\u043a\u0430\u043c\u0438 \u0442\u0430 \u0456\u043d\u0442\u0435\u0440\u043d\u0435\u0442\u043e\u043c, \u0434\u043e\u0441\u044f\u0433\u043b\u0438 \u043f\u043e\u043a\u0430\u0437\u043d\u0438\u043a\u0430 36,1%. \u0420\u0435\u0437\u0443\u043b\u044c\u0442\u0430\u0442\u0438 \u043e\u043f\u0443\u0431\u043b\u0456\u043a\u0443\u0432\u0430\u043b\u0438 22 \u0436\u043e\u0432\u0442\u043d\u044f 2026 \u0440\u043e\u043a\u0443 \u0432 \u043d\u0430\u0443\u043a\u043e\u0432\u043e\u043c\u0443 \u0436\u0443\u0440\u043d\u0430\u043b\u0456 Science. \u0412\u043e\u0434\u043d\u043e\u0447\u0430\u0441 \u0432\u0438\u043f\u0440\u043e\u0431\u0443\u0432\u0430\u043d\u043d\u044f \u043f\u0440\u043e\u0432\u043e\u0434\u0438\u043b\u0438 \u043d\u0430 \u0456\u0441\u0442\u043e\u0440\u0438\u0447\u043d\u0438\u0445 \u043a\u043b\u0456\u043d\u0456\u0447\u043d\u0438\u0445 [\u2026]<\/p>\n","protected":false},"author":207433346,"featured_media":34337,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"rank_math_description":"\u0428\u0406 Microsoft \u0434\u043e\u0441\u044f\u0433 \u0442\u043e\u0447\u043d\u043e\u0441\u0442\u0456 83,2% \u0443 \u0434\u0456\u0430\u0433\u043d\u043e\u0441\u0442\u0438\u0446\u0456 \u0441\u043a\u043b\u0430\u0434\u043d\u0438\u0445 \u0437\u0430\u0445\u0432\u043e\u0440\u044e\u0432\u0430\u043d\u044c \u043f\u0440\u043e\u0442\u0438 36,1% \u0443 \u043b\u0456\u043a\u0430\u0440\u0456\u0432. \u0429\u043e \u043f\u043e\u043a\u0430\u0437\u0430\u043b\u043e \u0434\u043e\u0441\u043b\u0456\u0434\u0436\u0435\u043d\u043d\u044f \u0432 Science.","rank_math_title":"\u0428\u0406 Microsoft \u043f\u0435\u0440\u0435\u0432\u0435\u0440\u0448\u0438\u0432 \u043b\u0456\u043a\u0430\u0440\u0456\u0432 \u0443 \u0442\u0435\u0441\u0442\u0430\u0445: 83% \u043f\u0440\u043e\u0442\u0438 36%","rank_math_focus_keyword":"Microsoft","td_subtitle":"","subtitle":"","footnotes":""},"categories":[805620],"tags":[806764],"class_list":["post-34336","post","type-post","status-publish","format-standard","has-post-thumbnail","category-science-tech","tag-806764"],"_links":{"self":[{"href":"https:\/\/ukrmedia.news\/en\/wp-json\/wp\/v2\/posts\/34336","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ukrmedia.news\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ukrmedia.news\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ukrmedia.news\/en\/wp-json\/wp\/v2\/users\/207433346"}],"replies":[{"embeddable":true,"href":"https:\/\/ukrmedia.news\/en\/wp-json\/wp\/v2\/comments?post=34336"}],"version-history":[{"count":2,"href":"https:\/\/ukrmedia.news\/en\/wp-json\/wp\/v2\/posts\/34336\/revisions"}],"predecessor-version":[{"id":34339,"href":"https:\/\/ukrmedia.news\/en\/wp-json\/wp\/v2\/posts\/34336\/revisions\/34339"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ukrmedia.news\/en\/wp-json\/wp\/v2\/media\/34337"}],"wp:attachment":[{"href":"https:\/\/ukrmedia.news\/en\/wp-json\/wp\/v2\/media?parent=34336"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ukrmedia.news\/en\/wp-json\/wp\/v2\/categories?post=34336"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ukrmedia.news\/en\/wp-json\/wp\/v2\/tags?post=34336"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}