Wrist joint synovial hypertrophy and effusion detection in musculoskeletal ultrasound images using self-attention U-Net

Skeletal muscle ultrasound has emerged as a pivotal imaging modality in rheumatology clinics, offering unparalleled advantages such as radiation-free imaging, safety, and dynamic examination capabilities. However, its reliance on operator expertise often leads to inconsistent interpretations and dia...

Full description

Saved in:
Bibliographic Details
Published in:Multimedia tools and applications Vol. 83; no. 41; pp. 89317 - 89334
Main Authors: Chang, Chuan-Wang, Chang, Chuan-Yu, Zhu, Yu-Xian, Wang, Sz-Tsan
Format: Journal Article
Language:English
Published: New York Springer US 01.12.2024
Springer Nature B.V
Subjects:
1237: Advanced Deep Learning for Computer Vision and Multimedia Applications
Computer Communication Networks
Computer Science
Data Structures and Information Theory
Diagnostic systems
Image segmentation
Joints (anatomy)
Lesions
Medical imaging
Multimedia Information Systems
Semantic segmentation
Sensitivity
Special Purpose and Application-Based Systems
Ultrasonic imaging
Wrist
Synovial hypertrophy
Effusion
U-Net
Skeletal muscle ultrasound
SEAT-UNet
ISSN:1573-7721, 1380-7501, 1573-7721
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Skeletal muscle ultrasound has emerged as a pivotal imaging modality in rheumatology clinics, offering unparalleled advantages such as radiation-free imaging, safety, and dynamic examination capabilities. However, its reliance on operator expertise often leads to inconsistent interpretations and diagnostic variability. In this study, we present a novel diagnostic system aimed at detecting rheumatoid arthritis (RA) in the wrist joint, with a focus on identifying synovial hypertrophy and effusion using musculoskeletal ultrasound images. Leveraging deep learning techniques, specifically semantic segmentation models, we introduce SEAT-UNet, which combines the U-Net architecture with a self-attention mechanism to enhance the accuracy of lesion classification and localization. SEAT-UNet addresses the challenge of discontinuous dispersion encountered in conventional segmentation models, particularly when delineating lesion areas. Our experimental results demonstrate exceptional performance, achieving a sensitivity and Dice coefficient of 100% and 84%, respectively, in synovial hypertrophy detection, and 86% sensitivity with an 84% Dice coefficient in effusion detection. These findings underscore the potential of SEAT-UNet as a valuable tool for early RA diagnosis, offering improved patient outcomes and facilitating more effective disease management strategies.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1573-7721
1380-7501
1573-7721
DOI:10.1007/s11042-024-19910-5