[Edaily Reporter Kim Seung-kwon ] The Health Insurance Review and Assessment Service (HIRA) has applied AI-generated medical imaging interpretation results to its medical expense review processes for the first time. This marks the first instance in which a government agency has begun using AI interpretations not merely as “reference material” but as “the basis for decision-making.” Observers suggest that this development could fundamentally alter the domestic revenue structure of Korean AI medical imaging companies.

Since 2018, HIRA has been developing a deep learning-based interpretation system and has been using AI interpretation results for spinal, knee joint, and urinary tract stone images as reference materials for reviews. It has now applied the system to actual review operations, starting with the knee joint (osteoarthritis) sector, where verification has been completed. HIRA plans to gradually expand the scope of application to include spinal and urinary tract stone cases in the future.

◇Crescom for the Knee, DEEPNOID Inc. for the Infrastructure…Estimated Supply Structure

HIRA has not disclosed which company’s algorithm it is using. However, based on a synthesis of publicly available information, the list of potential suppliers has narrowed considerably.

A leading candidate in the knee field is Crescom, an AI company specializing in musculoskeletal conditions. Crescom was selected for HIRA’s “Knee Osteoarthritis Insurance Reimbursement Review Support System” project and developed “MediAI-OA,” an AI system for automatic analysis of KL (Kellgren-Lawrence) grades; this solution is currently being used in HIRA’s knee osteoarthritis reimbursement review processes. MediAI-OA is a solution that quantitatively provides KL grades, the presence of osteophytes, and the degree of joint space narrowing; it received a Class II medical device manufacturing license in April 2024. In June of this year, it also obtained U.S. Food and Drug Administration (FDA) 510(k) clearance. This is why it is highly likely that MediAI-OA will serve as the core engine for this knee joint application.

In the imaging infrastructure sector, DEEPNOID Inc.(315640)is being mentioned. It is reported that HIRA has been utilizing DEEPNOID Inc.’s AI image interpretation support solution, “DeepFax,” to build an AI image storage and transmission system. Given that DEEPNOID Inc. possesses AI solutions for spinal disorders, it is possible that additional opportunities may arise as HIRA expands into the spinal field, as previously announced. Lunit Inc.(328130), VUNO, Inc.(338220), and other companies possessing AI medical imaging diagnostic technologies are also expected to see growth in their domestic sales.

◇The Real Variable Is “Defensive Demand from Healthcare Providers”

The core of the rationale for these benefits does not lie in HIRA’s procurement itself. The scale of HIRA’s contracted services typically amounts to several hundred million won, which is unlikely to significantly impact the earnings of listed companies.

The key area to watch is healthcare institutions. Major reimbursement criteria, such as for artificial joint replacement surgery, are linked to imaging indicators like KL grade and joint space width. If HIRA begins using the same AI to batch-screen submitted images, hospitals and clinics will have an incentive to conduct their own pre-submission verification using the same criteria to avoid reimbursement cuts. This creates a structure where medical institutions have no choice but to adopt the same standards used by the review agency, meaning that domestic demand for imaging AI—which “sells even without reimbursement rates”—could emerge for the first time.

The “reference effect” from government agencies is also difficult to ignore. A track record of adoption in NHIS reviews serves as strong evidence in overseas public tenders and sales to large hospitals.

However, institutional bottlenecks remain. Of the 549 AI medical devices approved by the Ministry of Food and Drug Safety (MFDS), only 28 have undergone integrated review and been officially included in the reimbursement list, while 21 have been granted temporary reimbursement under an evaluation deferral. More than 91% of the remaining devices have not yet been incorporated into the national health insurance system.

An official in the AI medical industry stated, “The fact that HIRA refines its models independently using accumulated data is also a factor that reduces dependence on suppliers,” adding, “We must also consider the possibility that stricter reviews could reduce reimbursement revenue for medical institutions, thereby actually diminishing the incentive to adopt AI.”