Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FRCNet Frequency and Region Consistency for Semi-supervised Medical Image Segmentation

Along He, Tao Li, Yanlin Wu, Ke Zou, Huazhu Fu

TL;DR

Medical image segmentation is hampered by scarce pixel-wise labels. The authors propose FRCNet, a semi-supervised framework that jointly leverages frequency-domain and region-level cues through two consistency regularizers: Frequency Domain Consistency (FDC) and Multi-granularity Region Similarity Consistency (MRSC), built on a Mean Teacher backbone. FDC transforms RGB features with a Discrete Cosine Transform and uses a Frequency Enhancement Module to encode low- and high-frequency lesion cues, while MRSC enforces multi-scale region relationships via region similarity matrices across granularities. The training objective combines supervised loss with $\mathcal{L}_{fdc}$, $\mathcal{L}_{mrsc}$, and $\mathcal{L}_{pix}$ as $L = L_{sup} + \lambda( L_{fdc} + L_{mrsc} + L_{pix} )$, and FDC/MRSC can be removed at inference to keep complexity low; on Kvasir-SEG and ISIC 2016, FRCNet achieves state-of-the-art Dice and IoU with as little as 10% labeled data, while reducing annotation cost by almost 80%.

Abstract

Limited labeled data hinder the application of deep learning in medical domain. In clinical practice, there are sufficient unlabeled data that are not effectively used, and semi-supervised learning (SSL) is a promising way for leveraging these unlabeled data. However, existing SSL methods ignore frequency domain and region-level information and it is important for lesion regions located at low frequencies and with significant scale changes. In this paper, we introduce two consistency regularization strategies for semi-supervised medical image segmentation, including frequency domain consistency (FDC) to assist the feature learning in frequency domain and multi-granularity region similarity consistency (MRSC) to perform multi-scale region-level local context information feature learning. With the help of the proposed FDC and MRSC, we can leverage the powerful feature representation capability of them in an effective and efficient way. We perform comprehensive experiments on two datasets, and the results show that our method achieves large performance gains and exceeds other state-of-the-art methods.

FRCNet Frequency and Region Consistency for Semi-supervised Medical Image Segmentation

TL;DR

Medical image segmentation is hampered by scarce pixel-wise labels. The authors propose FRCNet, a semi-supervised framework that jointly leverages frequency-domain and region-level cues through two consistency regularizers: Frequency Domain Consistency (FDC) and Multi-granularity Region Similarity Consistency (MRSC), built on a Mean Teacher backbone. FDC transforms RGB features with a Discrete Cosine Transform and uses a Frequency Enhancement Module to encode low- and high-frequency lesion cues, while MRSC enforces multi-scale region relationships via region similarity matrices across granularities. The training objective combines supervised loss with , , and as , and FDC/MRSC can be removed at inference to keep complexity low; on Kvasir-SEG and ISIC 2016, FRCNet achieves state-of-the-art Dice and IoU with as little as 10% labeled data, while reducing annotation cost by almost 80%.

Abstract

Limited labeled data hinder the application of deep learning in medical domain. In clinical practice, there are sufficient unlabeled data that are not effectively used, and semi-supervised learning (SSL) is a promising way for leveraging these unlabeled data. However, existing SSL methods ignore frequency domain and region-level information and it is important for lesion regions located at low frequencies and with significant scale changes. In this paper, we introduce two consistency regularization strategies for semi-supervised medical image segmentation, including frequency domain consistency (FDC) to assist the feature learning in frequency domain and multi-granularity region similarity consistency (MRSC) to perform multi-scale region-level local context information feature learning. With the help of the proposed FDC and MRSC, we can leverage the powerful feature representation capability of them in an effective and efficient way. We perform comprehensive experiments on two datasets, and the results show that our method achieves large performance gains and exceeds other state-of-the-art methods.
Paper Structure (12 sections, 7 equations, 3 figures, 3 tables)

This paper contains 12 sections, 7 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Notations and Overview.
  4. Frequency Domain Consistency.
  5. Multi-granularity Region Similarity Consistency.
  6. Experiments
  7. Dataset and Evaluation Metrics.
  8. Implementation Details.
  9. Comparisons with SOTA Methods.
  10. Ablation Study.
  11. Conclusion
  12. Acknowledgements

Figures (3)

  • Figure 1: Overall framework of the proposed FRCNet, which is based on MT structure. For labeled data, they are used to train the student network directly with the labels, and unlabeled data is used to train the student network through the proposed consistency regularization.
  • Figure 2: The segmentation results of skin (first row) and polyp segmentation (second row) tasks using 10% labeled data.
  • Figure 3: The feature response with and without frequency domain consistency regularization. We can see that after frequency domain feature learning, the feature response of lesion region can be highlighted, and thus it can provide a more accurate decision boundary for the final segmentation.