Researchers have developed a method to fine-tune the MedSAM3 medical image segmentation model using Low-Rank Adaptation (LoRA) with as few as 10 annotated cases. This approach achieves performance competitive with specialist systems trained on much larger datasets, including reliable gallbladder segmentation where existing tools fail. The method requires only 3-5 hours of training per organ on a single GPU and is approximately 2-3 times faster than nnU-Net. The findings suggest that 10 annotated cases may be sufficient for clinically useful segmentation, reducing bottlenecks in image annotation and training time. --- Why it matters: This matters to researchers in AI because it provides a more efficient way to fine-tune medical image segmentation models with limited annotated data, which is crucial for clinical workflows. This approach can help reduce the burden of annotating large datasets and speed up model development. Source: https://arxiv.org/abs/2608.18731