He began by declaring that an internist’s duty is to take a comprehensive view of their patient.<\/p>\n
“The field of internal medicine covers a large number of diseases, so comprehensive study is very difficult. Partly due to this, internal medicine has become increasingly specialized. However, the patient in front of an internist often shows a wide range of symptoms. Given that an internist is usually the first physician most patients consult, internists must first examine the patient comprehensively, before drawing on their specialist expertise.”<\/p>\n
One method by which clinicians can engage in comprehensive study of internal medicine is to learn from a diverse array of cases. Nagai had long focused on JSIM regional meetings as an opportunity for this kind of study.<\/p>\n
“Back in 1990 or thereabouts, I found the presentation abstracts published by JSIM’s Kanto Regional Meeting to be a valuable source of information,” he explains. “While the abstracts were only around 500 characters, they followed a fairly standardized format and all the cases had confirmed diagnoses. What was more, whereas medical textbooks contain cases from overseas, these abstracts focused on Japanese patients that physicians had actually encountered in clinical practice. As I felt they were the ideal teaching materials for studying a diverse range of cases, I built a database of 3,000 cases on my computer in 1993 or so, back when I was an associate professor and chief ward physician at the University of Tokyo Hospital. Then I distributed it as an appendix to a book for residents.”<\/p>\n
A decade or so later, case reports from regional meetings across Japan were digitized, and the number of reports eventually grew to exceed 60,000.<\/p>\n
“It occurred to me that if I used these as a database, I could support diagnosis in general practice,” he says. “Partly because I was serving as JSIM president at the time, in 2009, I asked the University of Tokyo’s Professor Kazuhiko Ohe (then Director of the Department of Healthcare Information Management at the University of Tokyo Hospital, now a specially appointed professor at Juntendo University) and Assistant Professor Eiji Aramaki (now a professor at Nara Institute of Science and Technology) to create a high-speed search system for medical terminology in case reports, which we called Shorei-kun” (Figure 1).<\/p>\n
![\"\"](\"https:\/\/healthist.net\/en\/wp-content\/uploads\/sites\/3\/2025\/08\/292_en_feature01_02_fig01.jpg\")
<\/small><\/p>\nFigure 1. <\/span>The case report search system by the Japanese Society of Internal Medicine (2009)<\/span><\/strong>“Shorei-kun” is a system that displays case report titles (abstracts), the presenter, their affiliation, and the classification of the disease based on a keyword search. For example, when the user inputs “Fever, Thrombocytopenia, Impaired consciousness” into the search field (top), a list of similar cases is displayed (bottom).<\/p>\n<\/div>\n Still in operation today, this search system continues to be used by many internists. However, although the search system generates hits for a lot of information, it cannot distinguish between positive and negative findings, nor can it understand the context of the cases.<\/p>\n “I realized that what I wanted was to create a mechanism that enabled the user to trace the context, using the accumulated case reports,” Nagai says.<\/p>\n Upon learning about mind mapping software, which presents information in a radial structure, he began working diligently. In 2016, his project titled “Development of an Artificial Intelligence-Based General Practice Diagnostic Support System” was selected by the Japan Agency for Medical Research and Development (AMED) for funding under its Program on ICT Infrastructure Development for Clinical Research. In addition, in FY2019, he received support from the New Energy and Industrial Technology Development Organization (NEDO) as part of its Development Project on Data Sharing in Collaborative Areas and AI System to Achieve the “Connected Industries,” for a project titled “Development of a Medical Care Support AI System Based on Cross-Cutting Integration of Medical Information.” Nagai served both as writer and editor-in-chief, while development of the search algorithm was led by Keita Oda (Specially Designated Associate Professor at Jichi Medical University, former Google engineer), Takeshi Imai (Specially Designated Professor at Jichi Medical University; serving concurrently as an associate professor at the University of Tokyo), and Hisahiko Sato (M.D., AI Researcher, and CEO).<\/p>\n “With our research partners, we succeeded in creating J-CaseMap, a searchable diagnostic support system that graphically displays the context of case reports and their relationship to key medical terms,” Nagai says.<\/p>\n For example, the mechanism is such that if the user inputs keywords like “fever,” “thrombocytopenia,” and “disturbance of consciousness,” the AI analyzes the relationships between the keywords in the database and displays candidates for a differential diagnosis (Figure 2).<\/p>\n Figure 2. <\/span>The J-CaseMap: an AI-based diagnostic support system<\/span><\/strong>When the user enters, for example, “Fever, Thrombocytopenia, Impaired consciousness” (top), J-CaseMap displays a list of potential differential diagnoses (middle), and the user can browse relevant case reports and view a tree structured display of chief complaints, physical observations, and other information (bottom).<\/p>\n<\/div>\n J-CaseMap is used to search for similar cases when physicians are unsure about a differential diagnosis, or want to confirm heading in the right direction.<\/p>\n “Above all, it’s quite helpful in the case of very rare diseases, because it can jog the physician’s memory in a wide range of ways ——<\/span> everything from ‘I had no idea this could happen’ to ‘this disease never occurred to me’,” Nagai explains. “This system is also useful for organizing possible diagnoses when a physician encounters a difficult-to-diagnose case in community medicine.”<\/p>\n Nagai explains that this is because an internist needs to be a general practitioner. But at the same time, physicians responsible for providing medical care in communities where it is basically unfeasible to team up with specialist physicians and advanced medical institutions must also have some degree of specialist medical knowledge, in addition to engaging in general practice. This means they require an extremely wide range of clinical knowledge, conventionally reliant on the expertise and experience of the individual physician. However, it is hoped that a diagnostic support system like J-CaseMap could help to complement a physician’s knowledge.<\/p>\n Nevertheless, Nagai does not forget to issue some words of warning.<\/p>\n “To avoid misunderstanding, I should note that this AI-based diagnostic support system using case reports is employed solely to assist physicians in recalling comparatively rare diseases,” he stresses. “And it’s the physician who reaches the differential diagnosis. J-CaseMap doesn’t make judgments on behalf of the physician.”<\/p>\n In J-CaseMap, past case reports are used as the database. Nagai says case reports are a veritable treasure trove for the discerning reader, because they contain information about the disease name, symptoms, findings, and patient’s complaints.<\/p>\n “Case reports are the physician’s equivalent of a detective’s case book,” he explains. “The main disease can be likened to the real culprit, while complications are the accomplices. Case reports describe what kind of harm the culprit and accomplices respectively caused to the patient ——<\/span> what the symptoms and observations were ——<\/span> as well as how the case developed and what the outcome was. Clinicians use their expertise to interpret these medical narratives.”<\/p>\n Nagai gives the example of a case report involving vertebral artery dissection presenting with occipital headache. Interpreting the symptoms and observations reveals part of the context, namely that dissection of the vertebral artery caused a hematoma that obstructed the posterior inferior cerebellar artery, which branches off from the vertebral artery. This then triggered a cerebellar infarction.<\/p>\n “And that’s not all,” he continues. “One also needs to learn another piece of context: pain in the occipital region was caused by the vertebral artery hematoma compressing a nerve.”<\/p>\n When this information is structured, the main storyline ——<\/span> i.e. the disease that is the principal cause, along with its complications ——<\/span> becomes clear, and one can trace the progression of the disease in the patient’s body. Conversely, it is also possible to deduce the conceivable causes from what is happening.<\/p>\n “Case reports let us carry out this kind of training,” Nagai says.<\/p>\n While clinicians apply insights from case reports to future patients, interpreting such reports also helps enhance clinicians’ skills, Nagai explained.<\/p>\n Nevertheless, however much of a treasure trove the reports might be, merely amassing them in a disorderly fashion is a waste of their value.<\/p>\n In order to extract and use the requisite information, it is necessary to thoroughly systematize and classify it, and develop a well-structured dictionary.<\/p>\n “Even now, I use gaps in my schedule each day to add cases to J-CaseMap and organize the dictionary,” Nagai reveals. “Effective use of case reports requires comprehensive knowledge of the framework, classification scheme, and system structure. In particular, we must maintain consistency when interpreting the context and correcting the descriptions and terminology. That’s why, as the editor-in-chief ever since version 1, I have continued to carry out this meticulous finalization work myself.”<\/p>\n Having started off with around 6,000 registered cases, J-CaseMap has been updated annually, with Version 5 released on September 1, 2024. At the time of release, it had grown to include 20,268 cases, and has now reached 25,000. In addition, the algorithm has been improved to ensure that the increased number of cases does not slow down the search performance.<\/p>\n Although he has recently begun using AI to produce drafts, Nagai still continues to carry out the task of revising all the cases himself. If he has any free time at all, such as when he is traveling from one place to another, Nagai carries out this diligent proofing work every day on his laptop computer. One might imagine that this would be tough going, but Nagai says he certainly does not find it a hardship.<\/p>\n “That’s because doing this work helps me maintain my clinical intuition. Reading through case reports is almost like actually dealing with patients face to face ——<\/span> kind of like a simulation of clinical practice. It’s just the same as lawyers studying precedents. As long as I’m a physician, I feel that I mustn’t lose the mindset of a practicing clinician.”<\/p>\n A medical AI aimed at improving the quality of medical care may seem revolutionary.<\/p>\n However, “While the tools and means have changed, the development of medical science has been an unbroken process involving the accumulation, classification, search, and use of information,” Nagai points out.<\/p>\n By information, he means patient medical records, case reports, and the latest medical literature. Nagai first began compiling a medical database in the 1970s, he says. And it was in 1981 that he started using a computer.<\/p>\n “The catalyst was my former professor’s instruction to organize some medical charts dating back to the Meiji period (1868-1912),” he explains. “Back then, the accumulation, organization, and use of medical information was carried out entirely by hand using index cards.”<\/p>\n Physicians would write a heading on the card to serve as a keyword. Next, they would write a concise summary of the information. And then they would classify the card they had created according to their system, and search for and use it when needed. Without realizing, this task served as a kind of learning process, and the information would be structured in the physician’s head, Nagai says.<\/p>\n “Having to carry out this manual task for my professor gave me the idea to do it on the computer. I was just moving with the times. However, I only had 28 KB of memory, so the data had to be stored on tapes, and writing the program was hard, too.”<\/p>\n As well as the widespread adoption of electronic medical records, advances underway in the digital transformation (DX) of wards ——<\/span> initiatives aimed at using digital technology to improve ward operations and systems in an effort to increase efficiency ——<\/span> include efforts to enhance patient safety and boost the operational efficiency of medical professionals. Such moves seem set to continue.<\/p>\n “Using J-CaseMap enables a physician to reach a differential diagnosis earlier, and therefore to treat the patient sooner,” Nagai explains. “Increasing the speed of the process in this way allows physicians to use the time they spend on diagnosis more efficiently, which will lead to work style reforms for physicians, who currently face such issues as long working hours. For patients, it has the advantage of reducing how long they have to wait for a diagnosis and treatment.”<\/p>\n If used skillfully, a medical AI has advantages as a tool. But Nagai points out that it also presents issues.<\/p>\n “Hallucinations will likely be a major issue with medical AI going forward,” he cautions.<\/p>\n Referring to the experience of perceiving something that does not exist, hallucination is a phenomenon whereby an AI generates information that differs from reality or does not actually exist. The phenomenon is so named because it is as though the AI is experiencing an illusion, and outputs false information that appears to be correct.<\/p>\n “Experts say that we’ll never get the probability of this phenomenon’s occurrence down to zero,” Nagai says. “But the question is how to prevent it. It will likely be essential for physicians making good use of AI as a tool to have the ability to determine the authenticity of AI-generated information. To put it the other way around, physicians won’t be able to make good use of AI as a tool unless they have this ability to distinguish true from false. Consequently, they’ll need to get into the habit of checking the veracity of the information for themselves.”<\/p>\n Along with the evolution of medical AI, physicians must develop new competencies to effectively utilize these tools.<\/p>\n We understand the concept of disease in terms of pathology and observations. The concept of disease is created from the combination of these. This kind of system was first adopted in Ancient Greece during the time of Hippocrates, and reached its finished form in 18th century France as what was termed “Paris medicine.” It principally developed through observation and clinical autopsy in hospitals.<\/p>\n A scientific revolution subsequently took place in Germany when the focus turned to understanding the human body through its mechanisms. It was around this time that Japan ended its policy of national seclusion. As Germany led the world in medical science at the time, the Meiji government introduced “German medicine.”<\/p>\n Meanwhile, in the U.S., physician William Osler brought back what was then the world’s most advanced knowledge in the field of medical science from his studies in Europe, and established the foundations of “American medicine,” which incorporated a good balance of this knowledge. American medicine is an approach to medical science characterized by both the systematic approach of Paris medicine and the mechanisms of German medicine.<\/p>\n After World War II, the occupying forces turned the Hibiya parlor of Nittoh Tea into a library for the people of Japan. Students, workers, and scholars alike visited this library, where they came into contact with information about the U.S. The library also received the latest American medical literature. Our predecessors at the medical department had vied with each other to visit this library and would spend hours there poring over the latest medical journals. They would also summarize the information they had gained there on cards and share them with each other. It was through such zealous study that American medicine entered Japan anew. Mostly in the form of case reports, this knowledge found its way into the Journal of the Japanese Society of Internal Medicine<\/i>, which had had no articles to publish immediately after the war ended.<\/p>\n For J-CaseMap, we used collections of case reports from Japanese Society of Internal Medicine regional meetings. The forerunner to these regional meetings were the Department of Internal medicine at the Tokyo Imperial University (now the University of Tokyo), starting in the Taisho period (1912-1926). These continued to be held even during the war, when it was hard to conduct research; in the final year of the war, they took place once a month in the Department auditorium ——<\/span> with participants listening out for the air raid siren ——<\/span> right up until the Saturday (March 3) before the Great Tokyo Air Raid (March 10). The seminars featured presentations of cases and pathologic specimens, with participants undertaking case research informed by both Paris medicine and German medicine.<\/p>\n When turning these case reports into a database, I can sense the hopes of my predecessors, who conducted research based on the perspectives of systematization and classification, even amid tough times and difficult environments, and so studied the literature available to them with intense dedication. I regard today’s new tool of medical AI as an extension of their endeavors.<\/p>\n<\/div>\n When consulting a medical institution, the first place most people approach for an examination is a department or clinic specializing in internal medicine. At the same time, internists ——<\/span> physicians specializing in internal medicine ——<\/span> must conduct a comprehensive differential diagnosis process to reach an accurate judgment. However, as a diverse array of symptoms may occur in a patient, the limits of a physician’s experience and knowledge may not encompass diseases with few cases. The Japanese Society of Internal Medicine has published the J-CaseMap case search database to serve as a diagnostic tool. But as this database covers only rare diseases, physicians need to have the ability to distinguish the false from the genuine in the information generated by the AI.<\/p>\n","protected":false},"author":2,"featured_media":2500,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[14],"tags":[],"class_list":["post-2501","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-medicine"],"acf":{"author":"text by Toshiko Mogi","intro":" When consulting a medical institution, the first place most people approach for an examination is a department or clinic specializing in internal medicine. At the same time, internists ——<\/span> physicians specializing in internal medicine ——<\/span> must conduct a comprehensive differential diagnosis process to reach an accurate judgment. However, as a diverse array of symptoms may occur in a patient, the limits of a physician’s experience and knowledge may not encompass diseases with few cases. The Japanese Society of Internal Medicine has published the J-CaseMap case search database to serve as a diagnostic tool. But as this database covers only rare diseases, physicians need to have the ability to distinguish the false from the genuine in the information generated by the AI.<\/p>","person":[{"acf_fc_layout":"personcontent","personimg":2499,"personsholder":"President, Jichi Medical University","personname":"Ryozo Nagai","persondetail":"Graduated from the School of Medicine at the University of Tokyo’s Faculty of Medicine. He has previously served as lecturer and then associate professor in the Third Division of Internal Medicine at the University of Tokyo Hospital, professor in the Second Department of Internal Medicine at Gunma University School of Medicine, visiting professor at Tokyo Medical and Dental University’s Medical Research Institute, and professor in cardiovascular medicine at the University of Tokyo’s Graduate School of Medicine. Between 2003 and 2007, he held this last post concurrently with the position of Director of the University of Tokyo Hospital. He took up his current position in April 2012. Since May 2019, he has served as Medical Supervisor of the Imperial Household. His fields of specialism are cardiology and vascular biology."}],"issue":2493,"custom_css":".entry-content .v292_feature01_01_column{\r\nborder:#c8161d solid 4px;\r\nbackground-color:#fdebd2;\r\npadding:0 10px;\r\n}\r\n.entry-content .v292_feature01_01_column h3{\r\nbackground-color:#c8161d;\r\nborder-radius:0px 0px 12px 12px;\r\ncolor:#fff;\r\nfont-size:1.6em;\r\nfont-weight:bold;\r\nline-height:1;\r\nmargin:0 auto 5px auto;\r\npadding:4px 4px 8px 4px;\r\ntext-align:center;\r\nwidth:100%;\r\n}\r\n.entry-content .v292_feature01_01_column h4{\r\ncolor:#c8161d;\r\nfont-size:1.2em;\r\n}\r\n.entry-content .v292_feature01_01_column p{\r\nmargin:0 0 1em 0;\r\ntext-indent:1em;\r\n}\r\n.entry-content .v292_feature01_01_column p.no-margin{\r\nmargin:0;\r\n}"},"_links":{"self":[{"href":"https:\/\/healthist.net\/en\/wp-json\/wp\/v2\/posts\/2501","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/healthist.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/healthist.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/healthist.net\/en\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/healthist.net\/en\/wp-json\/wp\/v2\/comments?post=2501"}],"version-history":[{"count":0,"href":"https:\/\/healthist.net\/en\/wp-json\/wp\/v2\/posts\/2501\/revisions"}],"acf:post":[{"embeddable":true,"href":"https:\/\/healthist.net\/en\/wp-json\/wp\/v2\/issue\/2493"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/healthist.net\/en\/wp-json\/wp\/v2\/media\/2500"}],"wp:attachment":[{"href":"https:\/\/healthist.net\/en\/wp-json\/wp\/v2\/media?parent=2501"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/healthist.net\/en\/wp-json\/wp\/v2\/categories?post=2501"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/healthist.net\/en\/wp-json\/wp\/v2\/tags?post=2501"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}“This disease never occurred to me”<\/h2>\n
![\"\"](\"https:\/\/healthist.net\/en\/wp-content\/uploads\/sites\/3\/2025\/08\/292_en_feature01_02_fig02.png\")
<\/small><\/p>\nDatabase has now grown to 25,000 cases<\/h2>\n
The ability to determine the authenticity of information is essential<\/h2>\n
AI in the context of the history of medicine<\/h3>\n
“Paris medicine” and “German medicine” <\/h4>\n
“American medicine” entered Japan after the war<\/h4>\n
Case reports are the fruits of my predecessors’ efforts<\/h4>\n