Leveraging Self-Supervision to Bridge the Data Gap in Acetabular Version Estimation

The acetabular version is a critical factor in total hip arthroplasty planning, and it is typically measured from computed tomography (CT) scans as the standard procedure. However, the substantial radiation exposure and financial cost of CT scans make anterior-posterior (AP) pelvic radiographs an appealing alternative. Manually quantifying the acetabular version from radiographs is laborious and error-prone for clinicians. Deep learning methodologies offer an automated solution, but their performance is constrained by the paucity of large, annotated datasets, which are resource-intensive to acquire for medical imaging applications. In this work, we propose a novel self-supervised pre-training approach to learn effective representations from unlabeled pelvic radiographs to facilitate accurate acetabular version estimation using limited labeled data. Our dataset comprised 924 pelvic radiographs, including 624 unlabeled and 300 labeled with corresponding CT scans. The acetabular version angles for 300 patients were computed from their CT images. The pelvic radiographs were cut in half and divided into left and right sides, doubling the dataset size. We employed a self-supervised pre-training method to learn representations from unlabeled images on a separate task before fine-tuning on labeled data for the main acetabular version estimation task. Our approach achieved a notable reduction in mean absolute error, from 3.97° to 3.50°, compared to the ImageNet-initialized baseline, demonstrating the potential of self-supervised learning to enhance model performance when labeled data is scarce.