Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Domain Adaptive Lung Nodule Detection in X-ray Image

Haifeng Zhao, Lixiang Jiang, Leilei Ma, Dengdi Sun, Yanping Fu

TL;DR

This work tackles the domain shift problem in cross-center chest X-ray lung nodule detection by integrating mean teacher self-training with hierarchical region- and pixel-level contrastive learning and a nodule-level domain-invariant feature learning (NDL) module. The method leverages a gradient reversal layer and an adversarial domain classifier to align nodule representations across domains, while refining pseudo-label quality through multi-scale contrastive signals. A new annotated B-Nodule dataset is introduced to support cross-domain evaluation. Experimental results across three datasets show improved cross-domain performance, outperforming strong baselines and demonstrating enhanced robustness in detecting nodules amidst domain variations, which is crucial for real-world, multi-center radiology workflows.

Abstract

Medical images from different healthcare centers exhibit varied data distributions, posing significant challenges for adapting lung nodule detection due to the domain shift between training and application phases. Traditional unsupervised domain adaptive detection methods often struggle with this shift, leading to suboptimal outcomes. To overcome these challenges, we introduce a novel domain adaptive approach for lung nodule detection that leverages mean teacher self-training and contrastive learning. First, we propose a hierarchical contrastive learning strategy to refine nodule representations and enhance the distinction between nodules and background. Second, we introduce a nodule-level domain-invariant feature learning (NDL) module to capture domain-invariant features through adversarial learning across different domains. Additionally, we propose a new annotated dataset of X-ray images to aid in advancing lung nodule detection research. Extensive experiments conducted on multiple X-ray datasets demonstrate the efficacy of our approach in mitigating domain shift impacts.

Domain Adaptive Lung Nodule Detection in X-ray Image

TL;DR

This work tackles the domain shift problem in cross-center chest X-ray lung nodule detection by integrating mean teacher self-training with hierarchical region- and pixel-level contrastive learning and a nodule-level domain-invariant feature learning (NDL) module. The method leverages a gradient reversal layer and an adversarial domain classifier to align nodule representations across domains, while refining pseudo-label quality through multi-scale contrastive signals. A new annotated B-Nodule dataset is introduced to support cross-domain evaluation. Experimental results across three datasets show improved cross-domain performance, outperforming strong baselines and demonstrating enhanced robustness in detecting nodules amidst domain variations, which is crucial for real-world, multi-center radiology workflows.

Abstract

Medical images from different healthcare centers exhibit varied data distributions, posing significant challenges for adapting lung nodule detection due to the domain shift between training and application phases. Traditional unsupervised domain adaptive detection methods often struggle with this shift, leading to suboptimal outcomes. To overcome these challenges, we introduce a novel domain adaptive approach for lung nodule detection that leverages mean teacher self-training and contrastive learning. First, we propose a hierarchical contrastive learning strategy to refine nodule representations and enhance the distinction between nodules and background. Second, we introduce a nodule-level domain-invariant feature learning (NDL) module to capture domain-invariant features through adversarial learning across different domains. Additionally, we propose a new annotated dataset of X-ray images to aid in advancing lung nodule detection research. Extensive experiments conducted on multiple X-ray datasets demonstrate the efficacy of our approach in mitigating domain shift impacts.
Paper Structure (15 sections, 8 equations, 3 figures, 3 tables)

This paper contains 15 sections, 8 equations, 3 figures, 3 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Lung Nodule Detection in Chest X-ray image
  4. Unsupervised Domain Adaptation for Object Detection
  5. Contrastive Learning
  6. METHOD
  7. Hierarchical Contrastive Learning
  8. NDL Module
  9. Teacher-Student Mutual Learning
  10. EXPERIMENTS
  11. Experimental Settings
  12. Results and Comparisons
  13. Ablation Study
  14. CONCLUSION
  15. ACKNOWLEDGEMENT

Figures (3)

  • Figure 1: Samples from multiple X-ray datasets: (a) NODE21 10479589, (b) CXR cxr-dcjlk_dataset, and (c) B-Nodule dataset, which is collected and annotated by ourselves. There exist certain domain differences among chest X-ray images from different datasets regarding illumination, color contrast/saturation, resolution, and nodule quantity.
  • Figure 2: Overview of our model. Left: Teacher-student mutual learning and nodule-level domain-invariant feature learning (NDL) module. The teacher provides pseudo-labels to supervise the student, and the student refines the teacher by Exponential Moving Average (EMA). Right: Hierarchical contrastive learning strategy.
  • Figure 3: The comparison of the detection results of lung nodules between the approaches and our method in the scenario of CXR to NODE21. The small image serves as a magnified view of a specific region. The red boxes and green boxes denote the false positives (FP) and true positives (TP).