Related Experiment Video
Updated: Oct 6, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Global-Local attention network with multi-task uncertainty loss for abnormal lymph node detection in MR images
Shuai Wang1, Yingying Zhu2, Sungwon Lee2
1Imaging Biomarkers and Computer-Aided Diagnosis Laboratory, Radiology and Imaging Sciences, National Institutes of Health Clinical Center, Bethesda, MD 20892, USA; School of Mechanical, Electrical and Information Engineering, Shandong University, Weihai 264209, PR China.
This study introduces an improved Mask R-CNN network for detecting abnormal lymph nodes in MR images. The novel approach uses pseudo masks and attention mechanisms, achieving superior performance in lymph node detection.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Accurate detection of abnormal lymph nodes in MR images is crucial for disease diagnosis and treatment.
- Distinguishing abnormal lymph nodes from other tissues in MR images presents a significant challenge due to their similar appearances.
Purpose of the Study:
- To develop a novel network based on an improved Mask R-CNN framework for enhanced detection of abnormal lymph nodes in MR images.
- To address the challenge of limited pixel-wise annotated data by utilizing pseudo masks generated from RECIST bookmarks for supervision.
Main Methods:
- Proposed a novel network with two key innovations: global-local attention for encoding multi-scale context and extracting discriminative features, and multi-task uncertainty loss for adaptive weighting of objective functions.
- Utilized pseudo masks generated from RECIST (Response Evaluation Criteria in Solid Tumors) bookmarks as supervision, reducing the need for extensive pixel-wise annotations.
- Developed a new dataset comprising 821 RECIST bookmarks of 41 types of abnormal abdominal lymph nodes from 584 patients.
Main Results:
- The proposed network demonstrated superior performance in detecting abnormal lymph nodes compared to existing state-of-the-art approaches.
- The global-local attention mechanism effectively extracted more discriminative features by encoding both global and local context.
- The multi-task uncertainty loss adaptively optimized multiple loss functions, contributing to improved detection accuracy.
Conclusions:
- The developed Mask R-CNN based network offers a promising solution for accurate and reliable abnormal lymph node detection in MR images.
- The utilization of pseudo masks and innovative network components like global-local attention and multi-task uncertainty loss significantly enhances detection capabilities.
- This approach holds potential for improving the diagnosis and treatment planning of various diseases through more precise lymph node assessment in medical imaging.
