“From EMR documentation to diagnostic imaging”… ‘Big 5’ hospitals begin full-scale adoption of in-house developed ‘AI’ solutions

by Kang Hye-won

11. 27. 2025. 08:37

Implementation of customized technology through in-house AI development;

Reduction of medical staff workload through EMR automation;

AI in the field of radiology utilized for kidney CT analysis

Seoul St. Mary's Hospital is currently piloting 'CMC GenNote,' an enhanced version of its existing AI voice recognition system, within the hospital.

Recently, the "Big 5" tertiary hospitals have been introducing internally developed artificial intelligence (AI) solutions. This decision stems from the realization that relying on technology from AI medical diagnostic companies inevitably leads to fragmented system operations. Consequently, hospitals are now directly developing and operating AI technologies to reduce the risk of patient personal information leakage and enhance clinical efficiency by automating Electronic Medical Records (EMR) and reducing CT image analysis times.

According to the hospital industry on the 27th, the adoption of in-house AI-based EMRs is spreading, particularly among tertiary hospitals. Introducing AI-based systems enables the automatic recording and summarization of EMR entries, which account for a significant portion of medical staff workloads. Furthermore, by integrating with the Hospital Information System (HIS), medical staff can rapidly assess a patient's condition during emergencies.

Seoul National University Hospital developed a single platform called "SNUH.AI" by linking previously scattered AI systems to the hospital's computer network. SNUH.AI officially launched on the 3rd and is currently being utilized within the hospital. This platform supports real-time automatic EMR generation based on 'hari-q3,' a large-scale language model (LLM) developed by Seoul National University Hospital Healthcare AI. Currently, it is being utilized for pre-anesthesia status assessments and discharge records, which comprehensively evaluate the underlying diseases, test results, and risk levels of patients scheduled for surgery. It is also scheduled to be applied to a verification system that reviews pathology test results in real time.

Yonsei University Medical Center developed an AI model called 'Y-Knot' and introduced it to Severance Hospital last November. Y-Knot was developed under the leadership of Professor Seung-Chan Yoo of the Department of Biomedical Systems Informatics at Yonsei University College of Medicine, utilizing Llama3-8B, an LLM series with a Transformer (neural network) structure. It automatically generates drafts of documents within the EMR, making it primarily used in the Departments of Emergency Medicine and Anesthesiology and Pain Medicine, where time savings are critical.

Seoul St. Mary's Hospital has further enhanced its existing AI voice recognition system in collaboration with the AI ​​specialist company 'Puzzle AI.' 'CMC GenNote' is a system that not only records conversations between medical staff and patients in real-time but also generates summaries, formats, and chart drafts. It has been undergoing pilot operation since the 29th of last month in the hospital's Department of Pediatrics, Obstetrics and Gynecology, Endoscopy, and operating rooms. It can recognize the voices of specific medical staff even in noisy hospital environments and supports documentation that incorporates specialized terminology by utilizing LLM.

In-house AI development is also underway in the field of radiology. Kidney CT scans, in particular, are considered an area where the need for AI models is significant, as their complex location and structure make it time-consuming to distinguish between normal tissue and lesions.

To address this, Asan Medical Center developed an AI model that analyzes kidney CT images. This model was measured to have an accuracy of disease classification (AUC) of 0.99, which is close to 1. Furthermore, by applying homomorphic encryption (CKKS scheme), it can distinguish between normal tissue, cysts, and tumors even while the data is encrypted, eliminating the risk of personal information leakage. The hospital plans to expand the scope of application to all medical imaging in the future once support for high-performance Graphics Processing Units (GPUs) becomes available.

Samsung Seoul Medical Center has developed an AI model for monitoring kidney prognosis. By checking GFR and serum creatinine levels using this model, the prognosis of the remaining kidney in patients who have donated a kidney can be assessed. Rapid verification is possible by inputting diagnostic results into web-based questionnaires without the need for separate devices. The hospital stated, "We have filed a patent for commercialization and plan to integrate it into the hospital's Electronic Medical Records (EMR) system."

There is a trend among hospitals to implement customized technologies through their own proprietary AI programs. An official from Asan Medical Center said, "Since medical environments and clinical data differ from hospital to hospital, customized AI technology suitable for these needs is required, so we are developing and applying our own AI programs." The official added, "If the accuracy and speed of diagnosis and treatment improve through the future introduction of AI, it will contribute not only to ensuring patient safety but also to the early detection of severe diseases and improved prognosis."

Source: Asia Today(https://www.asiatoday.co.kr/)