On generalisability of segment anything model for nuclear instance segmentation in histology images

TL;DR

This paper evaluates the generalisability of the Segment Anything Model (SAM) for nuclear instance segmentation in histology images, exploring zero-shot and fine-tuned performance, and proposing a nuclei detection-based prompting approach.

Contribution

It introduces a method combining nuclei detection with SAM prompts for improved nuclear segmentation in histology images and compares SAM's performance with other methods.

Findings

SAM shows promising zero-shot segmentation performance.

Fine-tuning improves SAM's accuracy in nuclear segmentation.

The proposed detection-based prompting enhances segmentation results.

Abstract

Pre-trained on a large and diverse dataset, the segment anything model (SAM) is the first promptable foundation model in computer vision aiming at object segmentation tasks. In this work, we evaluate SAM for the task of nuclear instance segmentation performance with zero-shot learning and finetuning. We compare SAM with other representative methods in nuclear instance segmentation, especially in the context of model generalisability. To achieve automatic nuclear instance segmentation, we propose using a nuclei detection model to provide bounding boxes or central points of nu-clei as visual prompts for SAM in generating nuclear instance masks from histology images.