Supervise Less, See More: Training-free Nuclear Instance Segmentation with Prototype-Guided Prompting

Accurate nuclear instance segmentation is a pivotal task in computational pathology, supporting data-driven clinical insights and facilitating downstream translational applications. While large vision foundation models have shown promise for zero-shot biomedical segmentation, most existing approaches still depend on dense supervision and computationally expensive fine-tuning. Consequently, training-free methods present a compelling research direction, yet remain largely unexplored. In this work, we introduce SPROUT, a fully training- and annotation-free prompting framework for nuclear instance segmentation. SPROUT leverages histology-informed priors to construct slide-specific reference prototypes that mitigate domain gaps. These prototypes progressively guide feature alignment through a partial optimal transport scheme. The resulting foreground and background features are transformed into positive and negative point prompts, enabling the Segment Anything Model (SAM) to produce precise nuclear delineations without any parameter updates. Extensive experiments across multiple histopathology benchmarks demonstrate that SPROUT achieves competitive performance without supervision or retraining, establishing a novel paradigm for scalable, training-free nuclear instance segmentation in pathology. Code is available at https://github.com/Y-Research-SBU/SPROUT.

Nuclear segmentation helps pathologists analyze tissue images for disease diagnosis and research, but building reliable models typically requires large numbers of manually annotated nuclei, a process that is time-consuming and costly. We ask whether this task can be performed without new annotations or model training, using the visual cues in pathology slides with a general-purpose segmentation model. We developed SPROUT, a training-free method that automatically creates prompts for nuclear instance segmentation. In common H&E-stained pathology images, nuclei usually appear dark blue or purple, while the surrounding tissue appears pink. SPROUT uses these stain patterns to pick out reliable nuclear and non-nuclear regions within each image, allowing the image to serve as its own reference. It then turns these reliable regions into positive and negative point prompts, while reducing the influence of uncertain areas. These prompts guide the Segment Anything Model to separate individual nuclei without updating the model. Across multiple pathology datasets, SPROUT achieves strong performance comparable to or better than methods that require training, annotations, or model adaptation. This makes nuclear segmentation easier to apply to new pathology datasets where annotations or retraining are impractical, and shows how foundation models can be made more useful in medical AI when guided by simple domain knowledge.