Related Experiment Video
Updated: Sep 26, 2025
04:25
Author Spotlight: Bridging Gaps in Anatomy and Establishing a Foundation for Algorithmic Studies
Published on: December 15, 2023
3.0K
Enhancing deep learning based classifiers with inpainting anatomical side markers (L/R markers) for multi-center
Ki Duk Kim1, Kyungjin Cho2, Mingyu Kim1
1Department of Convergence Medicine, University of Ulsan College of Medicine, Asan Medical Center, Seoul 05505, Republic of Korea.
Computer Methods and Programs in Biomedicine
|April 24, 2022
Summary
Inpainting anatomical side markers (L/R markers) in chest X-rays significantly boosts deep learning classifier performance for disease prediction. This method enhances accuracy and robustness in multi-center studies.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiology
Background:
- Anatomical side marker (L/R marker) placement protocols vary across institutions.
- These markers contain valuable signals for deep learning disease prediction models.
Purpose of the Study:
- To enhance deep learning classifier performance in multi-center chest radiograph datasets.
- To investigate the impact of inpainting L/R markers on classifier accuracy and robustness.
Main Methods:
- EfficientDet network for L/R marker detection.
- Generative Adversarial Network (GAN) for inpainting detected marker regions.
- Training and evaluating deep learning classifiers on original, inpainted, and clipped images.
- Comparative analysis using DeLong's and Stuart-Maxwell tests for validation.
Main Results:
- Marker inpainting significantly improved classification performance in both internal and external validation datasets.
- In binary classification, AUCs and accuracies increased notably with inpainted images (e.g., external validation: AUC 0.858 vs. 0.723).
- Inpainting led to superior performance in multi-class and multi-task learning, with improved activation maps.
Conclusions:
- Inpainting L/R markers enhances deep learning classifier performance and robustness for multi-center chest radiograph analysis.
- This preprocessing technique is valuable for developing more accurate AI tools in clinical settings.
- Open-source code for marker detection and inpainting is provided.
