<p>Bone Age Assessment (BAA) is a crucial diagnostic practice in medical image analysis. With the development of deep learning fueled by big data, the community has witnessed continuous and great improvement in BAA. However, most methods focus on the majority ethnic groups (MaG) using large-scale public datasets at the thousands-level. Considering the physiological differences among different ethnic groups, BAA for minority ethnic groups (MiG) remains a critical challenge. In this paper, we raise a novel task, few-shot bone age assessment (Fs-BAA), which aims for accurate BAA with limited samples. To this end, we propose a contrast relation adaptation framework with a unique in-batch ranking loss. Specifically, a parallel learning strategy is used where the network learns from MiG data while simultaneously acquiring knowledge from the MaG data. To address the domain shift between MaG and MiG, we maintain the seniority order of MaG during fine-tuning and design an in-batch ranking loss that leverages the contrast relationships of MaG as an optimization objective in parallel learning, thus facilitating the MiG task as a supplementary constraint. Additionally, to fully utilize data from diverse ethnic groups, we propose an efficient hybrid adaption method for extracting single or hybrid features in a flexible and lightweight manner. Our method can be flexibly applied to different ethnic groups and backbones. Extensive experiments demonstrate that our method achieves reliable performance across different ethnic groups using limited samples.</p>