인기 키워드 순위

    메디게이트 뉴스

    Gil Medical Center Counts 360 AX Champions as Hospital AI Shifts to Integration and Operations

    Speakers at the AX Innovation Division symposium cite data linkage, system integration and cost

    Posted on November 22, 2026 20:46
    Modified on November 22, 2026 21:02

    ※ This article was translated by AI.
    Hyun-jung Kim, director of the AI R&D Center at Gachon University Gil Medical Center, explains the significance of the AX Innovation Division's first anniversary at a symposium on Nov. 18. Photo: Gil Medical Center.
    Hyun-jung Kim, director of the AI R&D Center at Gachon University Gil Medical Center, explains the significance of the AX Innovation Division's first anniversary at a symposium on Nov. 18. Photo: Gil Medical Center.

    [MedigateNews] Speakers at a symposium agreed that using medical artificial intelligence (AI) continuously in clinical practice requires safely connecting data across hospitals and solving the problems of integration with existing clinical systems and operating costs.
     
    Hyun-jung Kim, director of the AI R&D Center at the AX Innovation Division of Gachon University Gil Medical Center and a professor of dermatology, spoke on Nov. 18 at a symposium marking the division's first anniversary. "When we asked all staff what they wanted to try with AI, the number of 'AX (AI Transformation) champions' who brought ideas from the field reached 360," Kim said. "The achievement of the past year is that hospital staff themselves identify and build AI projects, rather than relying only on custom development by outside companies."
     
    The projects staff proposed ranged from reducing repetitive work that comes with patient care to building tools suited to specific tasks. Patients' inconveniences at the hospital and the difficulties staff encounter at work became the starting point for deciding where to apply AI.
     
    "Staff are proposing even everyday problems, such as patient guidance and work flow, as AI projects, and they are actively offering ideas," Kim said. "When staff bring ideas, the AX Innovation Division connects them with the relevant departments or people in charge."
     
    Kim also said the process helps build the hospital's AI capabilities. "Staff on the front line specify what problem they want to solve and work with the relevant departments to find ways to apply it to actual work, and through this we aim to build the AI the hospital needs ourselves," Kim said. Kim emphasized, "It is also an opportunity for outside AI startups to solve real clinical problems together with hospital staff."
     
    For the symposium, the division invited speakers from three government and public organizations, the Ministry of Science and ICT, the National Institute of Medical Device Safety Information (NIDS) and the National IT Industry Promotion Agency (NIPA), and from two companies, Naver and haheho, so that staff could view AX in clinical work within a broader frame.
     
    From left: Su-jung Kim, director of the Artificial Intelligence Data Policy Division at the Ministry of Science and ICT (MSIT); Jung-rim Lee, president of the National Institute of Medical Device Safety Information (NIDS); and Myung-suk Yoon, team leader at the National IT Industry Promotion Agency (NIPA).

    Government and Public Agencies Set Out Conditions for Adoption, from Data Linkage to Regulation and Operations
     
    Su-jung Kim, director of the AI Data Policy Division at the Ministry of Science and ICT, emphasized in the keynote, "Attention is focused on GPU investment, but improving the performance of AI services requires refined data and a system to use it."
     
    "When we began building training datasets in 2018, there was a period when we collected data without knowing whether it was being used properly," Su-jung Kim explained. "We came to realize that a great deal was wasted when we collected it without thinking about demand and use."
     
    On hospital data, Su-jung Kim said, "People often say it is too sensitive to release, but it is in fact an asset. If we do not settle a compensation system, we will go around in circles forever, with nobody releasing data."
     
    As a solution, Su-jung Kim proposed data spaces modeled on cases such as Europe's Gaia-X. Under this approach, each hospital retains control over its data while linking and using it under agreed rules.
     
    The ministry supported its medical data space project for the first time this year. It plans to advance it to an upgrade stage next year and a rollout stage in the final year. Su-jung Kim set out three principles: a distributed, federated structure in which each hospital controls its data; immediate commercialization using existing infrastructure, technology and services as far as possible; and firm compensation proportionate to data volume and resources contributed.
     
    "Currently 27 institutions are participating, and we need to increase the number," Su-jung Kim said. "If the medical field succeeds, other industries will have no excuse."
     
    Jung-rim Lee, head of NIDS, described the relationship between innovative technology and regulation. "Innovative technology must be safe when used on people, the growth of innovative industries must be supported, and regulation should aim for maximum effect with minimum regulation," Lee said.
     
    According to Lee, about 400 applications for innovative medical device designation have been filed so far, and 132 of them (about 33%) were designated. The Ministry of Food and Drug Safety (MFDS) has also issued guidelines on AI medical devices and generative AI proactively, helping companies identify regulatory criteria during development.
     
    "When people hear 'regulator,' it is easy to think of an agency that only creates obstacles or imposes sanctions," Lee said. "If we tell companies in advance what is allowed and what is not, we can reduce their trial and error and give them the chance to make quality-assured products that clinicians can use with confidence."
     
    Myung-sook Yoon, team leader at NIPA, said, "Until now, the medical AI field has focused on solution development and verifying model accuracy. Now we need to think together about how to connect with existing hospital systems and how the tools will be used within clinicians' workflows."
     
    According to Yoon, NIPA-supported products had obtained 49 regulatory approvals and 637 paid deployments at medical institutions as of the first half of this year. In NIPA's medical AI development program, Dr. Answer, version 1.0 covered diagnosis and interpretation for eight diseases, and version 2.0 covered risk prediction and chronic disease management for 12 diseases. Version 3.0 is developing solutions for 15 diseases that support daily management after treatment.
     
    Yoon explained, "The 'AI-specialized hospital,' which would improve overall hospital operations and the patient journey, is currently at the concept stage, and its first-phase project is being run this year and next under the lead of Seoul National University Bundang Hospital."
     
    Yoon added, "High accuracy does not automatically improve clinical outcomes and patient safety. Without budget, operating staff, governance and an accountable organization, even the best technology cannot be used continuously." Yoon emphasized, "Above all, it is important to agree on the scope of integration, validation criteria and operational responsibility from the demonstration stage."
     
    Han-ju Yoo (left), head of Naver's Applied AI Group, and Han-sol Paeng, CEO of haheho.

    Hospital Use Cases: Clinical Search, Voice Medical Records and Pre-Consultation Questionnaires
     
    Han-ju Yoo, group leader of Naver's Applied AI Group, has worked on applying Naver's large language model (LLM) to healthcare since Naver unveiled it in 2021. Naver is focused on the medical foundation model 'KMed.AI', the clinical search and reasoning support service 'SNUH.AI' that uses it, and voice EMR, which organizes consultation conversations into medical records. It also recently launched a health agent service on the Naver portal.
     
    KMed.AI, which Naver is developing jointly with Seoul National University Hospital (SNUH), is a service that searches domestic and international clinical guidelines and recent papers when a clinical question is entered, answers with supporting evidence and suggests differential diagnoses.
     
    "Unlike faculty with deep subspecialty expertise, residents, nurses and private practitioners who see a wide range of patients have a need to check medical knowledge they are not familiar with," Yoo said. "Many nurses took part in the closed beta at SNUH in early September, and there were many questions about administration and regulations."
     
    Yoo added, "Professors and senior specialists are still at about 40% satisfaction, but more than 70% of nurses and residents said they would want to use it right now. KMed.AI is supplied to SNUH on-premises, and inquiries are also coming from Vietnam, Indonesia and Taiwan. We expect integration with ezCaretech's EMR around January next year." Yoo also noted difficulties in building up materials within a closed network: textbooks are not easy to purchase, and intellectual property (IP) rights for academic society guidelines must be negotiated one by one.
     
    On voice EMR, Yoo said, "If the content comes out after the consultation is over, you have to review it again. A three-minute consultation must not become a three-minute, 10-second consultation." Yoo said, "In last year's demonstration in the emergency room at Samsung Medical Center, conversation time with patients increased in all groups, but further analysis is needed to determine whether those longer conversations improved the quality of care."
     
    Cost is the obstacle to hospital adoption. "Global Big Tech companies that advertise performing better than doctors spend $60 each time they run the AI," Yoo said. "Private practices spend KRW 50,000 to 100,000 a month on EMR, so they cannot spend KRW 60,000 on a single consultation."
     
    Yoo also explained, "The cost of running KMed.AI is about one-thirty-second that of the top outside AI, and our goal is to cut it to one-hundredth. For speech recognition, we are testing processing on laptops or phones instead of servers." Yoo said, "We aim to improve Korean-language performance and release a standard version for laptops around next year."
     
    Han-sol Paeng, CEO of haheho, described cases in which small gaps in patient guidance lead to canceled surgeries or tests. For example, if instructions to remove nail polish or fast before surgery are not conveyed properly, that day's schedule can be delayed or canceled.
     
    "If you look at how operating room turnover or test cancellations change after improving the guidance process, you can gauge the effect small changes have on hospital operations," Paeng said.
     
    haheho sends designers into clinical settings to observe the inconveniences patients and clinicians actually experience, and designs tools to fit the work of each department. When AI pre-consultation questionnaires were introduced, nurses asked to receive the results on paper rather than look at a screen in front of patients. Reflecting this feedback, the company improved the tool so results can be printed on site.
     
    "We compared about 128 new outpatient pulmonology patients at a hospital, divided by whether they used the AI pre-consultation questionnaire," Paeng said. "Average consultation time was similar in the two groups, at about seven minutes, but the group that used the AI questionnaire provided more than twice as much information. Within the same consultation time, clinicians could learn more about the patient."
     
    Paeng added, "We are also considering ways to let information on patients moving among the emergency room, outpatient clinic and wards carry over at each stage of care."
     
    by Sol Lim
     
    ※ This article was translated by AI. If you find an error in the English, please let us know at news@medigatenews.com.