Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bridging the Applicator Gap with Data-Doping:Dual-Domain Learning for Precise Bladder Segmentation in CT-Guided Brachytherapy

Suresh Das, Siladittya Manna, Sayantari Ghosh

TL;DR

This work tackles covariate shift in bladder segmentation for CT-guided brachytherapy by introducing dual-domain learning through data-doping of abundant NA (no applicator) CT data with scarce WA (with applicator) data. By systematically evaluating multiple segmentation architectures across axial, sagittal, and coronal planes, the study demonstrates that mixing NA and WA samples—even at low WA proportions, such as $NA:WA = 7:3$—substantially improves performance, achieving dice and IoU scores near WA-only models. The findings quantify the benefits of cross-domain data integration, with improvements up to approximately $0.24$ IoU in the axial plane and consistent gains across other planes, thus enhancing segmentation reliability for treatment planning. Overall, the data-doping approach offers a practical path to robust and clinically reliable bladder segmentation under data scarcity in brachytherapy.

Abstract

Performance degradation due to covariate shift remains a major challenge for deep learning models in medical image segmentation. An open question is whether samples from a shifted distribution can effectively support learning when combined with limited target domain data. We investigate this problem in the context of bladder segmentation in CT guided gynecological brachytherapy, a critical task for accurate dose optimization and organ at risk sparing. While CT scans without brachytherapy applicators (no applicator: NA) are widely available, scans with applicators inserted (with applicator: WA) are scarce and exhibit substantial anatomical deformation and imaging artifacts, making automated segmentation particularly difficult. We propose a dual domain learning strategy that integrates NA and WA CT data to improve robustness and generalizability under covariate shift. Using a curated assorted dataset, we show that NA data alone fail to capture the anatomical and artifact related characteristics of WA images. However, introducing a modest proportion of WA data into a predominantly NA training set leads to significant performance improvements. Through systematic experiments across axial, coronal, and sagittal planes using multiple deep learning architectures, we demonstrate that doping only 10 to 30 percent WA data achieves segmentation performance comparable to models trained exclusively on WA data. The proposed approach attains Dice similarity coefficients of up to 0.94 and Intersection over Union scores of up to 0.92, indicating effective domain adaptation and improved clinical reliability. This study highlights the value of integrating anatomically similar but distribution shifted datasets to overcome data scarcity and enhance deep learning based segmentation for brachytherapy treatment planning.

Bridging the Applicator Gap with Data-Doping:Dual-Domain Learning for Precise Bladder Segmentation in CT-Guided Brachytherapy

TL;DR

This work tackles covariate shift in bladder segmentation for CT-guided brachytherapy by introducing dual-domain learning through data-doping of abundant NA (no applicator) CT data with scarce WA (with applicator) data. By systematically evaluating multiple segmentation architectures across axial, sagittal, and coronal planes, the study demonstrates that mixing NA and WA samples—even at low WA proportions, such as —substantially improves performance, achieving dice and IoU scores near WA-only models. The findings quantify the benefits of cross-domain data integration, with improvements up to approximately IoU in the axial plane and consistent gains across other planes, thus enhancing segmentation reliability for treatment planning. Overall, the data-doping approach offers a practical path to robust and clinically reliable bladder segmentation under data scarcity in brachytherapy.

Abstract

Performance degradation due to covariate shift remains a major challenge for deep learning models in medical image segmentation. An open question is whether samples from a shifted distribution can effectively support learning when combined with limited target domain data. We investigate this problem in the context of bladder segmentation in CT guided gynecological brachytherapy, a critical task for accurate dose optimization and organ at risk sparing. While CT scans without brachytherapy applicators (no applicator: NA) are widely available, scans with applicators inserted (with applicator: WA) are scarce and exhibit substantial anatomical deformation and imaging artifacts, making automated segmentation particularly difficult. We propose a dual domain learning strategy that integrates NA and WA CT data to improve robustness and generalizability under covariate shift. Using a curated assorted dataset, we show that NA data alone fail to capture the anatomical and artifact related characteristics of WA images. However, introducing a modest proportion of WA data into a predominantly NA training set leads to significant performance improvements. Through systematic experiments across axial, coronal, and sagittal planes using multiple deep learning architectures, we demonstrate that doping only 10 to 30 percent WA data achieves segmentation performance comparable to models trained exclusively on WA data. The proposed approach attains Dice similarity coefficients of up to 0.94 and Intersection over Union scores of up to 0.92, indicating effective domain adaptation and improved clinical reliability. This study highlights the value of integrating anatomically similar but distribution shifted datasets to overcome data scarcity and enhance deep learning based segmentation for brachytherapy treatment planning.
Paper Structure (11 sections, 3 equations, 5 figures, 2 tables)

This paper contains 11 sections, 3 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Material and methods
  3. Dataset establishment
  4. Model selection
  5. Evaluation Indicator
  6. Results
  7. Data selection
  8. Training only with NA data
  9. Data doping: Investigation with Assorted Data
  10. Generalization of Data Doping
  11. Conclusion

Figures (5)

  • Figure 1: Block diagram of the process flow for model training and validation with optimally assorted dataset. WA denotes With-applicator images.
  • Figure 2: Depiction of distorted image due to the presence of the brachy-applicator. Top Panel: Artifacts due to applicator insertion have been marked. Bottom Panel: No such artifact.
  • Figure 3: Failure of the training in bladder segmentation with exclusive NA data training. The top, middle and bottom panels represent original CT image for Axial plane, corresponding ground truth masks and segmentation results obtained for test dataset (WA data) respectively.
  • Figure 4: Bladder segmentation in Axial plane with different proportional data doping. The panels (a) and (b) represent original CT image and corresponding ground truth masks respectively. Panels (c)-(g) represent segmentation results for different NA:WA ratio in training data: (c) 1:9, (d) 3:7, (e) 5:5, (f) 7:3, (g) 9:1.
  • Figure 5: Bladder segmentation in Axial plane for different deep models for NA:WA$=7:3$ proportional data doping. The panels (a) and (b) represent original CT image and corresponding ground truth masks respectively. Panels (c)-(h) represent segmentation results for different models, U-Net, Half-UNet, RRDB U-Net, U-Net++, DC-UNet and Attention U-Net with the dice loss.