Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CTI-Unet: Cascaded Threshold Integration for Improved U-Net Segmentation of Pathology Images

Mingyang Zhu, Yuqiu Liang, Jiacheng Wang

TL;DR

CTI-Unet addresses threshold sensitivity in histopathology segmentation by cascading multi-threshold outputs within a two-stage U-Net framework. The Initial Segmentation Network (Model 1) generates preliminary masks, while the Threshold Integration Network (Model 2) fuses outputs from thresholds such as $0.01$, $0.1$, and $0.6$ along with the grayscale image to refine predictions. Training uses a composite loss $\mathcal{L} = \mathcal{L}_{\text{Tversky}}(\alpha_i,\beta_i) + 0.5\times \mathcal{L}_{\text{CE}}$, with normalization, resizing to $2048\times2048$, and MONAI-based augmentation, evaluated on the KPI2024 PAS-stained KPI dataset, where CTI-Unet outperforms nnU-Net, Swin-Unet, and CE-Net with an overall Dice score of $91.64\%$. Qualitative results show reduced noise and smoother boundaries thanks to the Threshold Integration Network, evidencing improved robustness and detail preservation in kidney pathology segmentation. The approach offers a flexible framework that can be extended with attention or transformer components and applied to other challenging histology segmentation tasks.

Abstract

Chronic kidney disease (CKD) is a growing global health concern, necessitating precise and efficient image analysis to aid diagnosis and treatment planning. Automated segmentation of kidney pathology images plays a central role in facilitating clinical workflows, yet conventional segmentation models often require delicate threshold tuning. This paper proposes a novel \textit{Cascaded Threshold-Integrated U-Net (CTI-Unet)} to overcome the limitations of single-threshold segmentation. By sequentially integrating multiple thresholded outputs, our approach can reconcile noise suppression with the preservation of finer structural details. Experiments on the challenging KPIs2024 dataset demonstrate that CTI-Unet outperforms state-of-the-art architectures such as nnU-Net, Swin-Unet, and CE-Net, offering a robust and flexible framework for kidney pathology image segmentation.

CTI-Unet: Cascaded Threshold Integration for Improved U-Net Segmentation of Pathology Images

TL;DR

CTI-Unet addresses threshold sensitivity in histopathology segmentation by cascading multi-threshold outputs within a two-stage U-Net framework. The Initial Segmentation Network (Model 1) generates preliminary masks, while the Threshold Integration Network (Model 2) fuses outputs from thresholds such as , , and along with the grayscale image to refine predictions. Training uses a composite loss , with normalization, resizing to , and MONAI-based augmentation, evaluated on the KPI2024 PAS-stained KPI dataset, where CTI-Unet outperforms nnU-Net, Swin-Unet, and CE-Net with an overall Dice score of . Qualitative results show reduced noise and smoother boundaries thanks to the Threshold Integration Network, evidencing improved robustness and detail preservation in kidney pathology segmentation. The approach offers a flexible framework that can be extended with attention or transformer components and applied to other challenging histology segmentation tasks.

Abstract

Chronic kidney disease (CKD) is a growing global health concern, necessitating precise and efficient image analysis to aid diagnosis and treatment planning. Automated segmentation of kidney pathology images plays a central role in facilitating clinical workflows, yet conventional segmentation models often require delicate threshold tuning. This paper proposes a novel \textit{Cascaded Threshold-Integrated U-Net (CTI-Unet)} to overcome the limitations of single-threshold segmentation. By sequentially integrating multiple thresholded outputs, our approach can reconcile noise suppression with the preservation of finer structural details. Experiments on the challenging KPIs2024 dataset demonstrate that CTI-Unet outperforms state-of-the-art architectures such as nnU-Net, Swin-Unet, and CE-Net, offering a robust and flexible framework for kidney pathology image segmentation.

Paper Structure

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

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Overview
  5. Initial Segmentation Network (Model 1)
  6. Threshold Integration Network
  7. Implementation Details
  8. Experiments and Results
  9. Dataset
  10. Experimental Setup
  11. Evaluation Metrics
  12. Results and Analysis
  13. Qualitative Results
  14. Conclusion

Figures (3)

  • Figure 1: Overview of the proposed Cascaded Threshold-Integrated Segmentation Framework. The initial segmentation module provides preliminary masks, which are processed at multiple thresholds. The Threshold Integration Network then refines these outputs to produce the final segmentation.
  • Figure 2: Network architecture of our prosed CTI-Unet. Left pink part is the initial segmentaion network while the right cyan part is the threshold integration network.
  • Figure 3: Comparison between the original image, ground truth, and the outputs of Model 1 and Model 2. As shown, Model 2 improves upon Model 1 by reducing noise and generating smoother boundaries, which can be seen by comparing the lower two images.