Lung Grounded-SAM (LuGSAM): A Novel Framework for Integrating Text
prompts to Segment Anything Model (SAM) for Segmentation Tasks of ICU
Chest X-Rays
Abstract
Chest radiography is a commonly utilized imaging technique for acquiring
Chest X-Ray (CXR) images due to its cost-effectiveness and its role in
diagnosing lung?related disorders. Nevertheless, interpreting CXR images
can be challenging, and the process of separating the lung field from
CXR images can be a valuable tool for assessing and diagnosing lung
diseases. While various segmentation methods exist, this study primarily
focuses on META’s latest Segment Anything Model (SAM). SAM is an
Artificial Intelligence (AI) model designed to segment objects within an
image. This research aims to harness SAM’s capabilities for segmenting
CXR images. Additionally, we explore the potential of another novel
model called Grounding DINO. Grounding DINO is a zero-shot object
detection model that utilizes a Swin (Shifted Windows) transformer for
extracting image features and BERT (Bidirectional Encoder
Representations from Transformers) for extracting textual information.
It is primarily employed to detect objects in an image based on a
provided text prompt, creating bounding boxes around the objects when
certain text and box thresholds are met. These bounding boxes are then
used as prompts for SAM to generate segmentation masks. The proposed
framework has been assessed on CXRs obtained from patients at Emory
Hospital in Atlanta, Georgia, USA and further evaluated using NIH
clinical center’s CXR image dataset.