Lung Segment Anything Model (LuSAM): A Prompt-integrated Framework for Automated Lung Segmentation on ICU Chest X-Ray Images

Abstract — The Segment Anything Model (SAM) is a state-of-the-art deep learning architecture for image seg- mentation that allows users to segment arbitrary objects in images using points, or bounding boxes. SAM uses a novel two-stage approach that first encodes the input image into a high-dimensional embedding, and then generates object masks using the embedding and the input prompt. This ap- proach allows SAM to produce high-quality segmentation even for objects that were not seen during training, and to handle complex input prompts with multiple objects and attributes. We evaluate SAM on the open source Chest X- ray dataset known as Japanese Society of Radiological Technology (JSRT) dataset and another dataset obtained from Emory Hospital, Atlanta, Georgia,USA and show that it achieves state-of-the-art performance in terms of Stability Score and Jaccard Index including the perfor- mance across different demographic data such as age and gender. Overall, SAM provides a powerful tool for image segmentation, enabling efficient and accurate object detec- tion and segmentation for a wide range of applications. It also represents a significant advance in the field of image segmentation and opens up new opportunities for interactive and multi-modal image analysis.