Convolutional Neural Networks have emerged as a predominant tool in musculoskeletal medical image segmentation. It enables precise delineation of bone and cartilage in medical images. Recent developments in image processing and network architecture desire a reevaluation of the relationship between segmentation accuracy and the amount of training data. This study investigates the minimum sample size required to achieve clinically relevant accuracy in bone and cartilage segmentation using the nnU-Net methodology. In addition, the potential benefit of integrating available medical knowledge for data augmentation, a largely unexplored opportunity for data preprocessing, is investigated. The impact of sample size on the segmentation accuracy of the nnU-Net is studied using three distinct musculoskeletal datasets, including both MRI and CT, to segment bone and cartilage. Further, the use of model-informed augmentation is explored on two of the above datasets by generating new training samples implementing a shape model-informed approach. Results indicate that the nnU-Net can achieve remarkable segmentation accuracy with as few as 10–15 training samples on bones and 25–30 training samples on cartilage. Model-informed augmentation did not yield relevant improvements in segmentation results. The sample size findings challenge the common notion that large datasets are necessary to obtain clinically relevant segmentation outcomes in musculoskeletal applications.

Patient-specific angle and distance measurements are essential for evaluating joint alignment in orthopaedic diagnosis, surgical planning, and postoperative follow-up. Traditional methods often rely on 2D standing X-rays, which fail to capture the full 3D dynamics of the lower limb ¹ . This limitation has driven interest in 3D measurement techniques on computed tomography (CT). Furthermore, weight-bearing CT (WBCT) enables 3D measurements under physiological loading, a critical factor for alignment measurements ² . Manual anatomical landmarking to obtain, especially in 3D image annotation, is labour-intensive and prone to error due to the added level of complexity. Automating this process can significantly enhance speed, accuracy, and consistency of 3D measurements. Current state-of-the-art methods combine image segmentation, geometric operations, and statistical shape models but are time-consuming, computationally expensive, and depend on accurate bone segmentations ³ . To address these challenges, we propose a deep learning-based approach to automate anatomical landmarking directly at the image level. Our method employs a cascaded U-Net framework for various lower-limb measurements. In the first experiment, we utilized a dataset of 96 CT scans and 31 WBCT images to automate landmarking for 28 key knee and ankle measurements. In the second experiment, a separate dataset of 77 WBCT foot images, including 22 flatfoot cases, was used to extract 8 critical foot and ankle measurements. Landmark localization accuracy was assessed using Euclidean distance to ground truth landmarks. Average localization errors ranged from 1 mm to 3 mm, aligning with manual error ranges and image resolution. This approach demonstrates fast, reliable, and clinically applicable automation of anatomical landmarking, significantly reducing annotation time. While further fine-tuning is required, deep learning offers enhanced standardization and efficiency, improving decision-making and patient outcomes in orthopaedics.