Polyp SAM 2: Advancing Zero shot Polyp Segmentation in Colorectal Cancer Detection

TL;DR

The paper investigates zero-shot polyp segmentation for colorectal cancer detection using the Segment Anything Model (SAM) and its successor SAM 2. It conducts image- and video-based experiments across multiple public datasets, comparing performance under diverse prompts and reporting mDice and mIoU. The results show SAM 2 consistently outperforms SAM and achieves state-of-the-art performance without fine-tuning, with bounding-box prompts yielding the strongest gains, including a 31.4% mIoU improvement on video polyp segmentation. The findings suggest that SAM-2 can streamline clinical workflows by reducing annotation requirements for polyp segmentation and enable robust zero-shot performance in real-world colorectal cancer screening tasks.

Abstract

Polyp segmentation plays a crucial role in the early detection and diagnosis of colorectal cancer. However, obtaining accurate segmentations often requires labor-intensive annotations and specialized models. Recently, Meta AI Research released a general Segment Anything Model 2 (SAM 2), which has demonstrated promising performance in several segmentation tasks. In this manuscript, we evaluate the performance of SAM 2 in segmenting polyps under various prompted settings. We hope this report will provide insights to advance the field of polyp segmentation and promote more interesting work in the future. This project is publicly available at https://github.com/ sajjad-sh33/Polyp-SAM-2.

Figures (1)

• Figure 1: Qualitative Assessment of Segmentation Outcomes on Kvasir-SEG and CVC-300 Datasets using SAM kirillov2023segment and SAM 2.