Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Color-Quality Invariance for Robust Medical Image Segmentation

Ravi Shah, Atsushi Fukuda, Quan Huu Cap

TL;DR

Single-source domain generalization for medical image segmentation suffers when color distribution and image quality shift across domains. The authors propose Dynamic Color Image Normalization (DCIN) with Global and Local Reference Image Selection (GRIS/LRIS) and a Color-Quality Generalization (CQG) loss to enforce consistent segmentation across color/quality variants. Across throat-pharyngeal segmentation tasks, these components yield substantial cross-domain gains (up to 32.3 Dice points) and are model-agnostic. This approach improves robustness to unseen domains and holds promise for broader color-sensitive medical imaging applications.

Abstract

Single-source domain generalization (SDG) in medical image segmentation remains a significant challenge, particularly for images with varying color distributions and qualities. Previous approaches often struggle when models trained on high-quality images fail to generalize to low-quality test images due to these color and quality shifts. In this work, we propose two novel techniques to enhance generalization: dynamic color image normalization (DCIN) module and color-quality generalization (CQG) loss. The DCIN dynamically normalizes the color of test images using two reference image selection strategies. Specifically, the DCIN utilizes a global reference image selection (GRIS), which finds a universal reference image, and a local reference image selection (LRIS), which selects a semantically similar reference image per test sample. Additionally, CQG loss enforces invariance to color and quality variations by ensuring consistent segmentation predictions across transformed image pairs. Experimental results show that our proposals significantly improve segmentation performance over the baseline on two target domain datasets, despite being trained solely on a single source domain. Notably, our model achieved up to a 32.3-point increase in Dice score compared to the baseline, consistently producing robust and usable results even under substantial domain shifts. Our work contributes to the development of more robust medical image segmentation models that generalize across unseen domains. The implementation code is available at https://github.com/RaviShah1/DCIN-CQG.

Color-Quality Invariance for Robust Medical Image Segmentation

TL;DR

Single-source domain generalization for medical image segmentation suffers when color distribution and image quality shift across domains. The authors propose Dynamic Color Image Normalization (DCIN) with Global and Local Reference Image Selection (GRIS/LRIS) and a Color-Quality Generalization (CQG) loss to enforce consistent segmentation across color/quality variants. Across throat-pharyngeal segmentation tasks, these components yield substantial cross-domain gains (up to 32.3 Dice points) and are model-agnostic. This approach improves robustness to unseen domains and holds promise for broader color-sensitive medical imaging applications.

Abstract

Single-source domain generalization (SDG) in medical image segmentation remains a significant challenge, particularly for images with varying color distributions and qualities. Previous approaches often struggle when models trained on high-quality images fail to generalize to low-quality test images due to these color and quality shifts. In this work, we propose two novel techniques to enhance generalization: dynamic color image normalization (DCIN) module and color-quality generalization (CQG) loss. The DCIN dynamically normalizes the color of test images using two reference image selection strategies. Specifically, the DCIN utilizes a global reference image selection (GRIS), which finds a universal reference image, and a local reference image selection (LRIS), which selects a semantically similar reference image per test sample. Additionally, CQG loss enforces invariance to color and quality variations by ensuring consistent segmentation predictions across transformed image pairs. Experimental results show that our proposals significantly improve segmentation performance over the baseline on two target domain datasets, despite being trained solely on a single source domain. Notably, our model achieved up to a 32.3-point increase in Dice score compared to the baseline, consistently producing robust and usable results even under substantial domain shifts. Our work contributes to the development of more robust medical image segmentation models that generalize across unseen domains. The implementation code is available at https://github.com/RaviShah1/DCIN-CQG.

Paper Structure

This paper contains 14 sections, 1 equation, 3 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. METHODS
  3. Dynamic Color Image Normalization
  4. Global Reference Image Selection
  5. Local Reference Image Selection
  6. Ensembling Both Selected Reference Images
  7. The Color-Quality Generalization Loss
  8. EXPERIMENTAL RESULTS
  9. Data Collection
  10. Reference Image Selection
  11. Training Throat Segmentation Models
  12. Results of Throat Segmentation Models
  13. DISCUSSION
  14. CONCLUSION

Figures (3)

  • Figure 1: Comparison of segmentation results on HQ domain (source domain) and LQ domain (other domains). While performing well on the HQ domain (first row), the model was unable to capture semantic information from other domains (second to last rows).
  • Figure 2: The overview of our proposed dynamic color image normalization (DCIN) and color-quality generalization (CQG) loss.
  • Figure 3: Visual comparison of the results on the $\mathrm{LQ}_{seg}$ (top) and $\mathrm{SP}_{seg}$ (bottom) datasets from different segmentation models. Results with DCIN are from the full DCIN setting (i.e., GRIS + LRIS).