The team led by Yi-Sheng Chan, Vice President of Keelung Chang Gung Memorial Hospital, has developed an artificial intelligence (AI) model specifically designed to detect tibial plateau fractures. By analyzing standard knee X-rays, this AI tool helps physicians quickly and accurately assess the likelihood of fractures, especially those that are subtle and easily missed. This innovation promises to significantly reduce the misdiagnosis rate of tibial plateau fractures, ensuring timely treatment for patients and alleviating the burden on the healthcare system.

Traditional Methods Prone to Errors, Delaying Treatment with Severe Consequences

Tibial plateau fractures are common traumatic injuries, particularly prevalent among young adult males and elderly females. Due to the difficulty in diagnosing these fractures, delayed treatment often leads to complications such as joint deformity, arthritis, and even the need for early joint replacement surgery. Current diagnostic methods primarily rely on physical examinations and X-ray interpretations. However, the accuracy of physical examinations is heavily dependent on the physician's experience, and X-ray interpretations can often overlook hidden fractures.

Efficient and Lightweight Model Achieves Accurate Interpretation Through Extensive Data Learning

This AI model utilizes deep learning technology to analyze a vast number of tibial plateau X-ray images, enabling it to identify fracture characteristics that are often imperceptible to the human eye. The model is built on the EfficientNet B3 architecture, known for its excellent performance in image recognition. This architecture maintains high accuracy while reducing the model's complexity and computational resource requirements, allowing for a lightweight deployment.

To ensure the model's effectiveness, the team first collected over 3,800 tibial plateau X-ray images from various Chang Gung Memorial Hospital branches. Orthopedic specialists meticulously annotated the fracture areas to guarantee data quality. The dataset was then divided into training, validation, and testing sets, and the model was trained using the Adam optimizer with an appropriate learning rate. The validation results showed an AUC (Area Under the Curve) of 0.98 on the testing set. To further validate the model's generalization capability, independent datasets from the Chiayi and Keelung branches were used for external validation, achieving an AUC of 0.97, confirming the model's high accuracy across different datasets. Additionally, the team employed Grad-CAM technology to visualize the model's decision-making process, enhancing the interpretability of the AI and allowing physicians to better understand the basis of its judgments.

AI-Assisted System Enhances Diagnostic Efficiency, Reduces Healthcare Burden

Vice President Yi-Sheng Chan stated that this AI model not only improves diagnostic accuracy and speed but also enables non-specialist physicians to achieve diagnostic standards akin to orthopedic specialists. This capability is especially valuable in regions with limited medical resources. Furthermore, the model can seamlessly integrate with existing medical imaging systems, providing real-time interpretation results, thereby further enhancing healthcare efficiency. Looking ahead, the team plans to apply the model in broader clinical settings, such as acute screening and surgical planning, and will continue to optimize the model to handle a wider variety of fracture types, ultimately delivering better medical services to patients.

Resource (mandarin): AI輔助判讀X光片 揪出隱性骨折降低誤診！