As the adoption of artificial intelligence (AI) in hospitals expands, experts suggest shifting the focus from 'what to implement' to 'how to manage and integrate' these technologies.
Barun Law Firm, in collaboration with Caleb & Company, announced on September 9 that they co-hosted the '1st AI Innovation Leaders Forum – Transforming and Advancing Hospital AI: From Data and Governance to Clinical Solutions' on September 8 at the Melon Hall on the 5th floor of the L Tower in Seocho-gu, Seoul.
The event, attended by over 100 practitioners from general hospitals and related medical institutions, featured partners including the Korea Health and Medical Information Service and AI Nation & Jeongwon Ensys.
In the first session focused on 'Transforming AI Governance,' Professor Yang Kwang-mo of Sungkyunkwan University and head of the AI Research Center at Samsung Medical Center diagnosed structural disconnections in the field through his presentation titled 'The Core of Hospital AI Transformation – A Hospital Where Data Flows.' He noted, "Even if hospital AI passes performance verification, there remains a significant gap before it translates into actual work." He identified key issues as disconnections between institutions (limitations in standardizing and sharing images), between tasks (scattered channels and manual classification), and between evaluations (discrepancies in defined metrics and performance data).
Professor Yang pointed out that changes in regulations, such as Germany's Hospital Reform Act, the UK's NHS App-centered 10-year plan, and domestic support projects for the structural transformation of advanced general hospitals, all demand a common need for 'defining and connecting necessary information.'
As a solution, he proposed a 'data-flowing operational system' that connects standards, data, AI judgment, task execution, and performance evaluation. He shared examples from the Samsung Medical Center consortium, including patient-centered medical image sharing (938 cases exchanged among six institutions), automated classification of medical referrals using an on-premises LLM (analyzed 6,624 electronic referrals with a 75.4% assignment accuracy rate, improved to 84.7% after expert re-evaluation), and the integrated referral and return system 'S-CARENet.'
He emphasized the need to pre-design who, where, and how to process data after AI judgment, connecting data, AI, tasks, and evaluations under the premise of human final verification (Human-in-the-loop), and highlighted the division of roles among management, planning, and IT departments.
Attorney Ahn Juhyun of Barun presented on 'Domestic AI Regulations and Hospital Responses – Checkpoints for AI Implementation through Regulations and Law,' diagnosing the regulatory landscape faced by hospitals. She stated, "Hospital AI is subject to multiple regulations, including the Basic AI Act, regulations on digital medical products, and the Personal Information Protection Act," emphasizing that the focus of regulation is not the model itself but the combination of data processing and clinical usage. She also noted that even similar clinical support functions could face different regulatory pathways depending on their context of use, such as medical device approvals for advertising versus actual use.
As a response, she proposed inspection criteria based on usage phases. When utilizing generative AI for medical record writing or patient guidance, hospitals must adhere to privacy checkpoints, including removing patient identification information, using secure internal networks, and notifying patients of AI usage. Procedures for processing, exporting, combining, and managing pseudonymized information and health data must also be pre-designed. Ahn stressed that hospital AI governance should operate continuously before implementation, during operation, and during changes, highlighting the need for comprehensive legal and institutional responses that encompass AI regulations, medical devices, personal information, and medical laws.
In the second session on 'AI Strategies and Solutions,' Bae Gi-won, head of the Innovation Center at Caleb & Company, diagnosed patterns of failure in AI implementation in his presentation titled 'Hospital AX Strategy and Governance – A to Z.' He noted, "While implementation is increasing, it remains unclear who approves and who is responsible," identifying common causes of the 'seven major failures' as governance gaps, including verification-field discrepancies, alert fatigue, performance degradation (drift) due to lack of post-management, bias, vendor dependency, and gaps in responsibility and governance. He warned that if all AIs are reviewed with the same intensity, it could lead to an increase in 'shadow AI' as the field seeks to circumvent regulations.
As a solution, he suggested that the first step is to inventory the scattered AIs within the institution, proposing differentiated reviews based on risk and a dedicated organization with a multi-layered governance system. He stated, "Well-designed governance does not slow down the pace of implementation; rather, it accelerates it," urging management to view regulatory compliance as a matter of implementation speed rather than cost.
Lee Se-ra, team leader at AI Nation, shared insights on 'AI Born in Clinical Settings – Co-developing a Pressure Ulcer Solution,' diagnosing challenges faced in the field. She explained, "Pressure ulcer management is a labor-intensive task that involves repeated measurement, imaging, stage assessment, and documentation for each lesion," sharing difficulties in refining data collected in the field and the challenge of distinguishing 'healed pressure ulcers' from similar lesions like dermatitis.
She introduced the journey of developing an AI detection solution for pressure ulcer stages in collaboration with Yonsei Severance Hospital over 21 months, explaining the three-step processing of labeling and classifying over 30,000 lesion data, defining evaluation criteria, and confirming results through imaging and inference (lesion area segmentation and stage classification). Lee emphasized that this is a record created in collaboration with the field, not a product sold by a vendor.
Kwon Chan-young, head of the AI Strategy Office at Jeongwon Ensys, addressed the essence of real-time diagnostic support AI in his presentation on 'Real-time Diagnostic Support AI.' He noted, "Video is ultimately a series of still images, and if it is 30 frames per second, the detection AI makes independent judgments 30 times per second." He diagnosed limitations in relying on single-frame detection, which can lead to false positives when markers flicker or frames are blurred.
As a solution, he proposed tracking whether the same lesion is continuously detected across frames in the temporal axis to ensure reliability, and skipping blurred frames to achieve smooth detection without flickering, introducing examples of real-time polyp detection (CADe) and diagnostic support (CADx) in endoscopy.
During the panel discussion, Lee Jun-hee, head of Barun's Corporate Strategy Research Institute, moderated a discussion with panelists including Kim Kyung-yoon, senior AWS Healthcare executive, Min Gyu-hong, head of the Smart Hospital Information Strategy Team at Seoul St. Mary's Hospital, Lee Eui-kyu, attorney at Barun, and Han Eun-jin, specialized nurse at Severance Hospital's Clinical Support Nursing Team, focusing on the evolving role of hospitals and challenges in data and governance.
Min Gyu-hong emphasized, "Hospitals must transform from places that treat illnesses to comprehensive health management centers that encompass prevention and proactive management," stating that personal medical information should be provided to individuals while also being utilized for AI research through lawful procedures, creating a virtuous cycle back into clinical practice. He stressed the urgent need for 'AI-Ready medical data' that has undergone standardization, refinement, and quality control, identifying the common challenge of transitioning dispersed data into an accessible and manageable structure while reinforcing the structure and meaning of existing data.
Lee Eui-kyu highlighted that the core of AI governance is identifying the AIs used by hospitals and establishing approval, verification, and monitoring procedures based on risk levels, ensuring the final judgment authority of medical staff, and maintaining a compliance system that records verification, changes, and incident responses. He added, "Such governance is not a regulation but a foundation for the rapid introduction of safe AI."
Yum Min-seob, director of the Korea Health and Medical Information Service, emphasized that the competitiveness of medical AI is not solely created by algorithms but requires high-quality data, standards, and a safe linkage and utilization system to lead to real innovation in medical practice. He introduced the government's basic medical AI strategy, the public medical AI highway, the health information highway (medical MyData), and the national integrated bio big data project, stating, "The Korea Health and Medical Information Service will work to create medical innovations that citizens can feel through medical data and AI."
Lee Dong-hoon, representative attorney at Barun, stated, "For the successful implementation of medical AI, comprehensive responses to various regulations related to AI, digital medical products, medical devices, personal information protection, and medical data utilization are essential." He added, "Barun has accumulated expertise and experience in digital healthcare, personal information and data, AI regulations, and consulting for medical institutions, and will faithfully fulfill its role as a strategic partner in proactively addressing and resolving legal issues faced by hospitals and medical institutions at each stage of AI transformation."
Barun Law Firm, in collaboration with Caleb & Company, announced on September 9 that they co-hosted the '1st AI Innovation Leaders Forum – Transforming and Advancing Hospital AI: From Data and Governance to Clinical Solutions' on September 8 at the Melon Hall on the 5th floor of the L Tower in Seocho-gu, Seoul.
The event, attended by over 100 practitioners from general hospitals and related medical institutions, featured partners including the Korea Health and Medical Information Service and AI Nation & Jeongwon Ensys.
In the first session focused on 'Transforming AI Governance,' Professor Yang Kwang-mo of Sungkyunkwan University and head of the AI Research Center at Samsung Medical Center diagnosed structural disconnections in the field through his presentation titled 'The Core of Hospital AI Transformation – A Hospital Where Data Flows.' He noted, "Even if hospital AI passes performance verification, there remains a significant gap before it translates into actual work." He identified key issues as disconnections between institutions (limitations in standardizing and sharing images), between tasks (scattered channels and manual classification), and between evaluations (discrepancies in defined metrics and performance data).
Professor Yang pointed out that changes in regulations, such as Germany's Hospital Reform Act, the UK's NHS App-centered 10-year plan, and domestic support projects for the structural transformation of advanced general hospitals, all demand a common need for 'defining and connecting necessary information.'
As a solution, he proposed a 'data-flowing operational system' that connects standards, data, AI judgment, task execution, and performance evaluation. He shared examples from the Samsung Medical Center consortium, including patient-centered medical image sharing (938 cases exchanged among six institutions), automated classification of medical referrals using an on-premises LLM (analyzed 6,624 electronic referrals with a 75.4% assignment accuracy rate, improved to 84.7% after expert re-evaluation), and the integrated referral and return system 'S-CARENet.'
He emphasized the need to pre-design who, where, and how to process data after AI judgment, connecting data, AI, tasks, and evaluations under the premise of human final verification (Human-in-the-loop), and highlighted the division of roles among management, planning, and IT departments.
Attorney Ahn Juhyun of Barun presented on 'Domestic AI Regulations and Hospital Responses – Checkpoints for AI Implementation through Regulations and Law,' diagnosing the regulatory landscape faced by hospitals. She stated, "Hospital AI is subject to multiple regulations, including the Basic AI Act, regulations on digital medical products, and the Personal Information Protection Act," emphasizing that the focus of regulation is not the model itself but the combination of data processing and clinical usage. She also noted that even similar clinical support functions could face different regulatory pathways depending on their context of use, such as medical device approvals for advertising versus actual use.
As a response, she proposed inspection criteria based on usage phases. When utilizing generative AI for medical record writing or patient guidance, hospitals must adhere to privacy checkpoints, including removing patient identification information, using secure internal networks, and notifying patients of AI usage. Procedures for processing, exporting, combining, and managing pseudonymized information and health data must also be pre-designed. Ahn stressed that hospital AI governance should operate continuously before implementation, during operation, and during changes, highlighting the need for comprehensive legal and institutional responses that encompass AI regulations, medical devices, personal information, and medical laws.
In the second session on 'AI Strategies and Solutions,' Bae Gi-won, head of the Innovation Center at Caleb & Company, diagnosed patterns of failure in AI implementation in his presentation titled 'Hospital AX Strategy and Governance – A to Z.' He noted, "While implementation is increasing, it remains unclear who approves and who is responsible," identifying common causes of the 'seven major failures' as governance gaps, including verification-field discrepancies, alert fatigue, performance degradation (drift) due to lack of post-management, bias, vendor dependency, and gaps in responsibility and governance. He warned that if all AIs are reviewed with the same intensity, it could lead to an increase in 'shadow AI' as the field seeks to circumvent regulations.
As a solution, he suggested that the first step is to inventory the scattered AIs within the institution, proposing differentiated reviews based on risk and a dedicated organization with a multi-layered governance system. He stated, "Well-designed governance does not slow down the pace of implementation; rather, it accelerates it," urging management to view regulatory compliance as a matter of implementation speed rather than cost.
Lee Se-ra, team leader at AI Nation, shared insights on 'AI Born in Clinical Settings – Co-developing a Pressure Ulcer Solution,' diagnosing challenges faced in the field. She explained, "Pressure ulcer management is a labor-intensive task that involves repeated measurement, imaging, stage assessment, and documentation for each lesion," sharing difficulties in refining data collected in the field and the challenge of distinguishing 'healed pressure ulcers' from similar lesions like dermatitis.
She introduced the journey of developing an AI detection solution for pressure ulcer stages in collaboration with Yonsei Severance Hospital over 21 months, explaining the three-step processing of labeling and classifying over 30,000 lesion data, defining evaluation criteria, and confirming results through imaging and inference (lesion area segmentation and stage classification). Lee emphasized that this is a record created in collaboration with the field, not a product sold by a vendor.
Kwon Chan-young, head of the AI Strategy Office at Jeongwon Ensys, addressed the essence of real-time diagnostic support AI in his presentation on 'Real-time Diagnostic Support AI.' He noted, "Video is ultimately a series of still images, and if it is 30 frames per second, the detection AI makes independent judgments 30 times per second." He diagnosed limitations in relying on single-frame detection, which can lead to false positives when markers flicker or frames are blurred.
As a solution, he proposed tracking whether the same lesion is continuously detected across frames in the temporal axis to ensure reliability, and skipping blurred frames to achieve smooth detection without flickering, introducing examples of real-time polyp detection (CADe) and diagnostic support (CADx) in endoscopy.
During the panel discussion, Lee Jun-hee, head of Barun's Corporate Strategy Research Institute, moderated a discussion with panelists including Kim Kyung-yoon, senior AWS Healthcare executive, Min Gyu-hong, head of the Smart Hospital Information Strategy Team at Seoul St. Mary's Hospital, Lee Eui-kyu, attorney at Barun, and Han Eun-jin, specialized nurse at Severance Hospital's Clinical Support Nursing Team, focusing on the evolving role of hospitals and challenges in data and governance.
Min Gyu-hong emphasized, "Hospitals must transform from places that treat illnesses to comprehensive health management centers that encompass prevention and proactive management," stating that personal medical information should be provided to individuals while also being utilized for AI research through lawful procedures, creating a virtuous cycle back into clinical practice. He stressed the urgent need for 'AI-Ready medical data' that has undergone standardization, refinement, and quality control, identifying the common challenge of transitioning dispersed data into an accessible and manageable structure while reinforcing the structure and meaning of existing data.
Lee Eui-kyu highlighted that the core of AI governance is identifying the AIs used by hospitals and establishing approval, verification, and monitoring procedures based on risk levels, ensuring the final judgment authority of medical staff, and maintaining a compliance system that records verification, changes, and incident responses. He added, "Such governance is not a regulation but a foundation for the rapid introduction of safe AI."
Yum Min-seob, director of the Korea Health and Medical Information Service, emphasized that the competitiveness of medical AI is not solely created by algorithms but requires high-quality data, standards, and a safe linkage and utilization system to lead to real innovation in medical practice. He introduced the government's basic medical AI strategy, the public medical AI highway, the health information highway (medical MyData), and the national integrated bio big data project, stating, "The Korea Health and Medical Information Service will work to create medical innovations that citizens can feel through medical data and AI."
Lee Dong-hoon, representative attorney at Barun, stated, "For the successful implementation of medical AI, comprehensive responses to various regulations related to AI, digital medical products, medical devices, personal information protection, and medical data utilization are essential." He added, "Barun has accumulated expertise and experience in digital healthcare, personal information and data, AI regulations, and consulting for medical institutions, and will faithfully fulfill its role as a strategic partner in proactively addressing and resolving legal issues faced by hospitals and medical institutions at each stage of AI transformation."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.