CalFrac: A Calcaneal Fracture Dataset for Medical Image Segmentation

In recent years, the incidence of calcaneal fractures, a complex and common type of foot trauma, has risen sharply due to an increase in traffic accidents, falls from heights, and other injuries. However, research in computer-aided diagnosis of such fractures remains limited, primarily due to the lack of a precisely annotated dataset for training and validating segmentation algorithms. To address this gap, we have developed a new dataset, CalFrac, which includes CT scan data from 140 calcaneal fracture patients at Ruijin Hospital in Shanghai. The data is stored in the widely-used DICOM format, and under the guidance of expert clinicians, all cases were meticulously annotated using 3D Slicer software. This process yielded high-resolution 3D mask images that accurately capture the fracture regions in shape and extent. We further analyzed and evaluated the CalFrac dataset by testing six classical and advanced medical image segmentation models. The results highlighted the limitations of these models and validated the dataset's value for improving segmentation accuracy in calcaneal fracture diagnosis.