The Segment Anything Model (SAM) is a widely used tool for image processing, but its application in ultrasound image segmentation has limitations due to the noisy nature of ultrasound images. This project presents ClickSAM, a tool that fine-tunes SAM specifically for ultrasound images using click prompts. ClickSAM has two stages of training; the first stage uses single-click prompts centered in ground-truth contours, while the second stage improves model performance through additional positive and negative click prompts. The enhanced model performance is achieved by comparing the first stage’s predictions to the ground-truth masks and using a Centroidal Voronoi Tessellation algorithm to gather positive and negative click prompts in each segment. ClickSAM demonstrates superior performance in ultrasound image segmentation compared to other existing models.
Publication date: 9 Feb 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2402.05902
