A Landmark-Based Approach for Instability Prediction in Distal Radius Fractures

Distal radius fractures (DRFs) are among the most common fractures seen in emergency departments in the world. DRF instability estimation, also known as secondary displacement, is a critical assessment factor in deciding whether to operate or not, which can significantly affect patient recovery and costs of treatment. Clinical assessment of DRF instability heavily relies on the subjective option of the treating specialist. To improve this, we propose a unique deep learning-based landmark detection method that assesses DRF instability upon clinically relevant and reproducible bone landmarks. Using our defined anatomical landmarks in the routinely captured Posterior Anterior and lateral view X-rays, we use a landmark detection model to detect these landmarks and compute landmark-wise distance and radian measurements. These measurements are utilized as input features for an XGBoost classification model to solve the DRF instability classification task. We verify the effectiveness of the proposed method on a large DRF instability dataset collected from eleven hospitals in the Netherlands.