SAM3-AgeSeg: an adaptive segmentation model for bone tumors in the aging population.

BACKGROUND: Accurate segmentation of bone tumors from X-ray images is crucial for clinical diagnosis and treatment planning. However, elderly patients pose a significant challenge to general-purpose segmentation models. METHODS: To address this aging-related medical challenge, we propose SAM3-AgeSeg, an adaptive segmentation model specifically designed for the aging population. Built on the powerful foundation model SAM3, our approach uses a lightweight fine-tuning technique, Low-Rank Adaptation (LoRA), to efficiently learn and adapt to the imaging characteristics of elderly bones. RESULTS: We conducted a systematic evaluation of the public BTXRD bone tumor X-ray dataset. The experimental results demonstrate that SAM3-AgeSeg outperforms existing state-of-the-art methods in overall segmentation accuracy and exhibits superior boundary segmentation robustness, particularly for a subset of elderly patients. DISCUSSION: This study validates the effectiveness of adaptive strategies in enhancing the performance of medical image analysis for aging-related challenges, offering a potential research direction for further investigation into age-adaptive medical image segmentation.