Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CP-UNet: Contour-based Probabilistic Model for Medical Ultrasound Images Segmentation

Ruiguo Yu, Yiyang Zhang, Yuan Tian, Zhiqiang Liu, Xuewei Li, Jie Gao

TL;DR

A contour-based probabilistic segmentation model CP-UNet is proposed, which guides the segmentation network to enhance its focus on contour during decoding, and designs a novel down-sampling module to enable the contour probability distribution modeling and encoding stages to acquire global-local features.

Abstract

Deep learning-based segmentation methods are widely utilized for detecting lesions in ultrasound images. Throughout the imaging procedure, the attenuation and scattering of ultrasound waves cause contour blurring and the formation of artifacts, limiting the clarity of the acquired ultrasound images. To overcome this challenge, we propose a contour-based probabilistic segmentation model CP-UNet, which guides the segmentation network to enhance its focus on contour during decoding. We design a novel down-sampling module to enable the contour probability distribution modeling and encoding stages to acquire global-local features. Furthermore, the Gaussian Mixture Model utilizes optimized features to model the contour distribution, capturing the uncertainty of lesion boundaries. Extensive experiments with several state-of-the-art deep learning segmentation methods on three ultrasound image datasets show that our method performs better on breast and thyroid lesions segmentation.

CP-UNet: Contour-based Probabilistic Model for Medical Ultrasound Images Segmentation

TL;DR

A contour-based probabilistic segmentation model CP-UNet is proposed, which guides the segmentation network to enhance its focus on contour during decoding, and designs a novel down-sampling module to enable the contour probability distribution modeling and encoding stages to acquire global-local features.

Abstract

Deep learning-based segmentation methods are widely utilized for detecting lesions in ultrasound images. Throughout the imaging procedure, the attenuation and scattering of ultrasound waves cause contour blurring and the formation of artifacts, limiting the clarity of the acquired ultrasound images. To overcome this challenge, we propose a contour-based probabilistic segmentation model CP-UNet, which guides the segmentation network to enhance its focus on contour during decoding. We design a novel down-sampling module to enable the contour probability distribution modeling and encoding stages to acquire global-local features. Furthermore, the Gaussian Mixture Model utilizes optimized features to model the contour distribution, capturing the uncertainty of lesion boundaries. Extensive experiments with several state-of-the-art deep learning segmentation methods on three ultrasound image datasets show that our method performs better on breast and thyroid lesions segmentation.

Paper Structure

This paper contains 10 sections, 7 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Multi-group channel shifted downsampling
  4. Contour Probabilistic Modeling
  5. Gating-based feature filtering module
  6. Experiments
  7. Datasets and Experimental Settings
  8. Performance Comparison
  9. Ablation Studies
  10. Conclusion

Figures (4)

  • Figure 1: Enlarged images of nodules in six ultrasound images: (a) Clear contours; (b)-(d) Blurred contour edges; (e)-(f) Irregular shapes. The green contour line is physician-labeled
  • Figure 2: Framework of CP-UNet. CMP, Contour Probabilistic Modeling. MgCSD, Multi-group channel shifted downsampling. GF, Gating-based feature filtering module
  • Figure 3: Visualization of the results on two ultrasound images. The first three rows show the segmentation results of the breast ultrasound image, and the last row of the thyroid ultrasound
  • Figure 4: Visualization of contour segmentation performance by TransUNet and CP-UNet. The red boxes are poorly outlined areas, and the green line is the output of the model.