Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Diagonal Hierarchical Consistency Learning for Semi-supervised Medical Image Segmentation

Heejoon Koo

TL;DR

The paper tackles the high annotation cost in medical image segmentation by introducing DiHC-Net, a semi-supervised framework built from three identical multi-scale V-Net sub-models with distinct sub-layers. It combines deep supervision on labeled data with two consistency mechanisms: mutual consistency via sharpened pseudo labels and a novel diagonal hierarchical consistency that links pseudo labels to other models' intermediate and final predictions, optimized through a warming-up loss schedule. Empirical results on LA and BraTS 2019 demonstrate that DiHC-Net outperforms several strong baselines, with ablations confirming the contribution of multi-scale diversification and the DiHC loss to performance gains. This approach offers a robust, relatively simple strategy to leverage unlabeled data in MIS, tightening segmentation accuracy and contour fidelity while reducing annotation demands.

Abstract

Medical image segmentation, which is essential for many clinical applications, has achieved almost human-level performance via data-driven deep learning technologies. Nevertheless, its performance is predicated upon the costly process of manually annotating a vast amount of medical images. To this end, we propose a novel framework for robust semi-supervised medical image segmentation using diagonal hierarchical consistency learning (DiHC-Net). First, it is composed of multiple sub-models with identical multi-scale architecture but with distinct sub-layers, such as up-sampling and normalisation layers. Second, with mutual consistency, a novel consistency regularisation is enforced between one model's intermediate and final prediction and soft pseudo labels from other models in a diagonal hierarchical fashion. A series of experiments verifies the efficacy of our simple framework, outperforming all previous approaches on public benchmark dataset covering organ and tumour.

Diagonal Hierarchical Consistency Learning for Semi-supervised Medical Image Segmentation

TL;DR

The paper tackles the high annotation cost in medical image segmentation by introducing DiHC-Net, a semi-supervised framework built from three identical multi-scale V-Net sub-models with distinct sub-layers. It combines deep supervision on labeled data with two consistency mechanisms: mutual consistency via sharpened pseudo labels and a novel diagonal hierarchical consistency that links pseudo labels to other models' intermediate and final predictions, optimized through a warming-up loss schedule. Empirical results on LA and BraTS 2019 demonstrate that DiHC-Net outperforms several strong baselines, with ablations confirming the contribution of multi-scale diversification and the DiHC loss to performance gains. This approach offers a robust, relatively simple strategy to leverage unlabeled data in MIS, tightening segmentation accuracy and contour fidelity while reducing annotation demands.

Abstract

Medical image segmentation, which is essential for many clinical applications, has achieved almost human-level performance via data-driven deep learning technologies. Nevertheless, its performance is predicated upon the costly process of manually annotating a vast amount of medical images. To this end, we propose a novel framework for robust semi-supervised medical image segmentation using diagonal hierarchical consistency learning (DiHC-Net). First, it is composed of multiple sub-models with identical multi-scale architecture but with distinct sub-layers, such as up-sampling and normalisation layers. Second, with mutual consistency, a novel consistency regularisation is enforced between one model's intermediate and final prediction and soft pseudo labels from other models in a diagonal hierarchical fashion. A series of experiments verifies the efficacy of our simple framework, outperforming all previous approaches on public benchmark dataset covering organ and tumour.
Paper Structure (12 sections, 5 equations, 2 figures, 2 tables)

This paper contains 12 sections, 5 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodologies
  3. Task Formulation
  4. Diversifying Multi-scale Sub-models and Deep Supervision
  5. Diagonal Hierarchical Consistency Learning
  6. Experimental Setting
  7. Dataset
  8. Implementation Details
  9. Experimental Results
  10. Performance Comparison
  11. Ablation Studies
  12. Conclusion

Figures (2)

  • Figure 1: An Overview of Our Proposed Framework, DiHC-Net. The left provides a high-level conceptualisation of the proposed framework, whilst the right presents a visualisation on both deep-supervision and diagonal hierarchical consistency.
  • Figure 2: 2D Visualisation of Ground Truth and Predictions of Our DiHC-Net and Other Baselines on LA and BraTS Dataset. The first row is acquired when trained with 10% labelled data from LA dataset and the second row with 20% labelled data from BraTS dataset.