Seoul National University Hospital Presents Possibility of Automated AI-Based Measurement of Key Knee Prognostic Indicator

AI Model Enables Rapid and Standardized Measurement of Posterior Tibial Slope
Expected to Improve Knee Surgery Prognosis and Enhance Clinical Efficiency

A domestic research team has developed an artificial intelligence model that calculates a key indicator influencing knee surgery prognosis ten times faster than manual measurement. This advancement enables rapid and accurate measurement of the 'posterior tibial slope'-a metric that previously lacked standardized measurement criteria. It is expected that this will contribute to improved medical efficiency by providing unified standards for clinical practice and research in the future.

Professor Doohyun Noh, Department of Orthopedic Surgery, Seoul National University Hospital, Research Professor Seongeun Kim, Seoul National University Hospital

On August 26, Professor Doh Doohyun and Research Professor Kim Seongeun of the Department of Orthopedic Surgery at Seoul National University Hospital, together with researchers from the University of Minnesota in the United States and the University of Bergen in Norway, announced that they have developed a deep learning model capable of quickly and reliably measuring the posterior tibial slope. This model is based on more than 10,000 lateral knee joint X-ray images taken between 2009 and 2019.

The 'posterior tibial slope' refers to the angle at which the articular surface of the tibia tilts backward when viewed from the side. This angle has a decisive impact on knee joint stability and the lifespan of artificial joints after knee surgery. In particular, a larger angle increases the risk of cruciate ligament injuries and can worsen the prognosis of artificial joint surgeries.

However, due to variations in the length and magnification of knee X-rays at different medical institutions, there has been no standardized method for measuring the posterior tibial slope. As a result, even for the same patient, measurement values could differ between institutions, limiting the clinical application of research findings.

The model developed by the research team automatically recognizes six anatomical landmarks on the knee bone. Using these points, it determines the joint line and central axis of the tibia, and then calculates the slope based on this information. This method offers the advantage of being applicable under various clinical conditions, including when X-ray images are short or when actual distance measurements are not possible.

When comparing the performance of this model with manual measurements by specialists, the deep learning model was found to be more than ten times faster. Additionally, the 'inter-observer correlation coefficient' was at least 91%, indicating that the deep learning model's measurements closely matched those of specialists. In contrast, the 'intra-observer correlation coefficient', which represents measurement consistency, was up to 95% for manual measurements, whereas the deep learning model achieved perfect consistency at 100%.

Furthermore, in a follow-up study validating knee images from 289 Norwegian patients, the inter-observer correlation coefficient between the deep learning model and specialists was 80%. This result suggests that the deep learning model for measuring the posterior tibial slope can be broadly applied to diverse ethnic groups.

Research Professor Kim Seongeun stated, "This result is a case in which medical AI technology developed in Korea has been successfully validated across multiple ethnicities," and added, "We will seek ways to expand the universality of this model through further studies so that it can become the standard for measuring the posterior tibial slope."

Meanwhile, the results of this study were recently published in the international journal 'Orthopaedic Journal of Sports Medicine.'

This content was produced with the assistance of AI translation services.