US framework combines hospital-wide governance, performance registries and temporary reimbursement
Launching the Series
Medical imaging artificial intelligence is evolving from a diagnostic aid that detects abnormalities into an operational foundation connecting hospital workflows, national screening programs, health insurance and medical data systems. Its clinical value depends on reliable integration, effective oversight, appropriate reimbursement and measurable patient outcomes.
Korea IT Times will publish this eight-part series every Friday from September 4 through October 23, 2026. The series will compare policies, hospital deployments and corporate strategies in South Korea, the United States and Europe, examining the conditions required for medical imaging AI to become national healthcare infrastructure and for Korean companies to compete globally.
Series Overview
September 4 | HIRA Takes Its First Step in AI Claims Review as Medical Imaging AI Enters National Infrastructure
September 11 | The Competition to Build Integrated Medical Imaging AI Platforms
September 18 | Lung Cancer Screening AI and Post-Screening Patient Management
September 25 | How Breast Cancer and Stroke AI Improve Patient Outcomes
October 2 | DICOM Standardization and the Expansion of Medical Imaging AI
October 9 | Performance Monitoring and Security for Medical Imaging AI
October 16 | Linking Medical Imaging AI Reimbursement to Clinical Outcomes
October 23 | A Global Market Strategy for Korean Medical Imaging AI
AI Enters the Health Insurance Claims Review Process
The industrial success of medical imaging AI increasingly depends on whether algorithms can be integrated reliably into hospital systems and evaluated in clinical practice. Although the number of image-analysis products has grown rapidly, deployment is now shaped by PACS, RIS and EMR integration, performance monitoring, clinician response procedures, accountability for errors and reimbursement.
South Korea has taken an important step. On September 1, the Health Insurance Review & Assessment Service (HIRA) announced that it had applied AI-generated medical image readings to claims reviews for knee osteoarthritis. The move extends medical imaging AI beyond diagnostic support and commercial products into the administration of the national health insurance system.
HIRA has operated a deep-learning medical image analysis system since 2018, using AI readings of spinal, knee-joint and urinary-stone images as reference material for claims reviews. After improving its models with accumulated images and interpretation data and conducting expert validation, the agency connected the verified knee-joint model to its review workflow.
Review officers examine the AI results alongside other relevant materials, while cases requiring medical judgment are referred to review committee members. Under HIRA’s internal operational and ethical standards, AI results alone cannot determine the outcome, and people remain responsible for the final decision.
The agency plans to expand the system gradually to spinal and urinary-stone cases while monitoring accuracy and operational results. Its stated objective is to reduce the time spent on repetitive image checks and allow reviewers to concentrate on cases requiring greater professional judgment.
HIRA President Hong Seung-won said the initiative was significant because AI-generated readings previously used for reference had now been applied directly to claims review. He added that HIRA would pursue AI transformation under operational and ethical principles to improve the professionalism and efficiency of its reviews.
The program’s impact will eventually need to be assessed through measurable operating data. Relevant indicators include average review time, cases handled per reviewer, agreement between AI and human assessments, committee re-review rates, error patterns and the outcomes of provider appeals. Model accuracy, update histories, performance-degradation thresholds and suspension procedures will also be essential to public confidence.
Clinical AI and insurance-review AI serve different purposes and carry different risks. Clinical systems support disease detection and treatment decisions, while claims-review systems assist assessments of billing appropriateness and administrative efficiency. Both require representative data, human oversight, traceable decisions, correction procedures and a clear route for appeals.
US health systems are addressing similar challenges through hospital-level operating frameworks. James Thannickal of Northwell Health argued in an August 5 commentary in the Journal of Imaging Informatics in Medicine that imaging AI should be managed as a core operational capability rather than as a collection of isolated products.
His six-part framework covers clinical intent, workflow orchestration, integration architecture, governance and safety, measurement and analytics, and continuous improvement. Before deployment, hospitals should define the patients and clinical stages to which an AI system will apply and design how its results will move through PACS, RIS and EMR systems. They must then measure clinical outcomes, working time, errors and costs to decide whether a model should be maintained, updated or withdrawn.
The framework addresses a growing operational problem. When emergency CT triage, lung-nodule detection, mammography analysis and stroke detection products connect to PACS in different ways, clinicians may have to navigate multiple screens and alert systems. Hospitals also face mounting costs in managing product updates, outages, access controls, audit logs and cybersecurity.
The strength of an imaging AI platform will therefore be judged by whether it can route images to the appropriate applications, return results to existing viewers and worklists, and connect them with relevant patient information. Hospitals also need the freedom to add or replace algorithms without rebuilding the entire integration architecture, as well as contingency procedures for system failures.
Post-Deployment Performance Will Shape the Quality of National Healthcare
The American College of Radiology (ACR) approved its first imaging AI practice parameter with the Society for Imaging Informatics in Medicine (SIIM) in May. Effective October 1, 2026, the guidance covers product selection, local testing, real-world performance monitoring and quality improvement.
It recommends that healthcare organizations establish AI governance groups involving clinicians, radiologists, IT and security specialists, data experts, compliance officers and executives. Hospitals should maintain an inventory of every AI product, its version and intended use, and conduct local acceptance testing before deployment. They should also define thresholds for suspending a system when performance deterioration or safety risks emerge.
Tessa Cook of the University of Pennsylvania, who chaired the writing committee, said the parameter provides steps covering AI selection, implementation, monitoring and improvement. SIIM Board Chair Nabile Safdar emphasized that responsible medical AI use is an ongoing process requiring a dedicated team and consistently applied procedures.
The ACR is also operating Assess-AI, a quality registry that tracks clinical performance after deployment. It compares AI outputs with surrogate labels extracted from radiology reports and combines them with de-identified patient characteristics and examination metadata. Participating institutions can compare their results with aggregated benchmarks from other sites and investigate discordant cases locally.
Assess-AI currently supports 12 imaging use cases, including intracranial hemorrhage, pulmonary embolism, pneumothorax, large-vessel occlusion, bone age, cervical-spine fracture and breast density. Such registries can reveal differences between performance demonstrated during regulatory review and results produced in routine care.
Patient populations, disease prevalence, equipment manufacturers, software versions and imaging protocols can all affect the performance of the same algorithm. A national professional registry can help identify error patterns that may be too rare or dispersed for one hospital to detect.
The US Food and Drug Administration is similarly focusing on the total product life cycle of AI-enabled devices, including real-world performance and model changes. Its AI-enabled medical device list provides authorization and review information, although the agency cautions that it is not comprehensive. The FDA is also exploring ways to identify devices incorporating foundation models, large language models and multimodal architectures.
South Korea is strengthening its regulatory foundations. In June, the Ministry of Food and Drug Safety issued guidance on predetermined change-control plans for AI-enabled digital medical devices. It followed in August with guidance for certifying quality-management systems for AI digital medical devices, extending oversight from initial authorization to software changes and continuing quality management.
Linking regulatory and change-management information with hospital quality data and HIRA’s claims-review records could provide the basis for national performance monitoring. Organizing these data by product, version, institution, imaging equipment, patient group, error type and clinical outcome would allow emerging problems to be identified across multiple sites. Rules governing patient privacy, trade secrets and institutional data rights would need to be established first.
Reimbursement also influences how imaging AI is adopted and operated. Beginning October 1, 2026, the Centers for Medicare & Medicaid Services (CMS) will provide a New Technology Add-on Payment (NTAP) for qualifying Medicare fee-for-service inpatient cases using Aidoc’s CARE Multi-Triage CT Body.
The system analyzes chest, abdominal and pelvic CT scans and flags multiple potentially urgent findings. The maximum additional payment for fiscal 2027 is $137.53 per eligible case. It is not a fixed payment for every CT scan; reimbursement is limited to the lesser of 65% of the technology’s cost or 65% of the amount by which the case cost exceeds the standard inpatient payment.
NTAP is designed to offset part of the cost of introducing qualifying technologies during their first two or three years in the market. CMS will eliminate the streamlined alternative pathway for fiscal 2028 applications, requiring future applicants to demonstrate substantial clinical improvement as well as novelty and cost criteria.
South Korea’s reimbursement model should likewise measure patient outcomes and system-wide efficiency. Useful indicators include reporting time, prioritization of emergency patients, reductions in additional testing, time to treatment, follow-up completion, interval cancers, readmissions and changes in total spending per patient. Temporary support could be adjusted as multicenter real-world evidence accumulates.
Competitive standards for companies are changing accordingly. Sensitivity and specificity remain basic market-entry requirements, while multicenter evidence, cybersecurity and lifecycle management increasingly determine long-term competitiveness. An accurate product may still deliver limited value if it complicates clinical work or increases alert fatigue.
Korean companies have accumulated specialized algorithms and overseas regulatory experience in lung cancer, thoracic disease, neurological disorders and breast cancer. Their next task is to connect individual results reliably with radiology worklists, EMRs, national screening follow-up and insurance systems. Hospital and government contracts should define integration and maintenance costs, data-access rights, performance reporting, update responsibilities, incident response times and the handling of data after a product is discontinued.
Medical imaging AI does not operate like fixed equipment whose performance remains unchanged after installation. Patient populations and clinical environments evolve, while software updates can alter both performance and risk. Regular evaluation and accountable governance are essential to preserving clinical value and public trust.
South Korea has a unified national health insurance system, nationwide screening programs, extensive digital imaging infrastructure and competitive AI companies. Connecting these assets through an integrated system of deployment, monitoring and reimbursement could transform medical imaging AI from a hospital-level tool into core national healthcare infrastructure.
Part 2 will examine the emerging platform competition to integrate multiple imaging AI applications into hospital workflows and assess where Korean companies stand.
