Seoul National University Hospital, Bundang Seoul National University Hospital, and Boramae Hospital—institutions actively adopting and operating medical artificial intelligence (AI)—have shared their on-site challenges regarding performance validation and sustainable operations.

The joint international symposium "AI, Unlimited," held on the 14th at the Woobong Hall of Seoul National University College of Medicine, featured a panel discussion addressing the criteria for medical AI adoption and operational challenges across the three hospitals.

The discussion was moderated by Yoon Hyung-jin, Director of the Seoul National University Medical Big Data Research Center, with participation from Lee Ho-young, Chief Information Officer (CIO) of Seoul National University Hospital; Yoo Hyung-won, Director of the AX Promotion Center at Bundang Seoul National University Hospital; and Kim Hak-ryung, CIO of Boramae Hospital.

Even Approved AIs Undergo Performance Checks Using In-House Data

Regarding the adoption of medical AI, the three hospitals operate procedures to verify not only product performance, but also whether the solutions fit their actual hospital data and work environments.

Seoul National University Hospital formed a red team two months ago to verify whether AI solutions deliver the performance claimed by vendors using the hospital's own data.

CIO Lee Ho-young stated, "We have established an additional process to check whether the promised performance is demonstrated using hospital data," adding, "Only after passing this stage is on-site application approved."

The hospital employs a dual-track approach: evaluating user suitability within a research Electronic Medical Record (EMR) environment, while designing comparative studies using existing data for the red team to assess.

Since early last year, Bundang Seoul National University Hospital has operated a Software Medical Device Review Subcommittee under its Equipment Deliberation Committee to review AI adoptions.

The subcommittee includes clinical department professors alongside representatives from biomedical engineering, health insurance review, IT, and medical record departments. Through this process, 21 items have passed review, and 9 have completed purchase contracts and been brought in.

Director Yoo Hyung-won noted, "Even if the technology itself is exceptional, we must minimize potential issues when it enters the hospital," explaining that they operate through a review process and staging servers.

Boramae Hospital has also been evaluating commercial product adoption through its Software Medical Device Working-level Subcommittee starting this year, while research-purpose solutions receive final decisions from the IT Promotion Committee.

CIO Kim Hak-ryung shared, "First and foremost, ensuring that the Hospital Information System (HIS) operates with maximum stability is key," and explained, "We step-by-step operate development and staging servers to confirm suitability before final application."

"Sustainability" of AI Services in Hospitals Is Not Solely Positive

Meanwhile, opinions emerged that continuous utilization of medical AI requires consideration of operational costs, staffing constraints, and adoption limitations depending on the presence of health insurance reimbursements.

CIO Lee Ho-young pointed out, "I do not think the sustainability of many AI services currently in widespread use is that high," citing heavy GPU dependency, high operating costs, and HIS structures that are ill-suited for AI utilization.

Director Yoo Hyung-won stated, "Software entering review with a reimbursement code has a structural advantage in receiving positive evaluations," adding, "Products without reimbursement face constraints during evaluation due to the hospital's purchase cost burden."

He also explained, "Because post-evaluation of software medical devices is a new operational domain, the expansion of manpower and resources dedicated to it is necessary."

CIO Kim Hak-ryung mentioned, "If we turn on AI solutions while conducting outpatient clinics, it feels like clinics might run even slower," noting that "we must also consider the burden of usage within actual clinical workflows."

Furthermore, addressing the heavy investment burden faced by public hospitals, Kim stated, "I think that with financial support and system improvements, we could utilize AI to contribute even more to vulnerable populations and public healthcare."

[Copyright (c) Global Economic Times. All Rights Reserved.]