Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Domain Adaptation of Carotid Ultrasound Images using Generative Adversarial Network

Mohd Usama, Belal Ahmad, Christer Gronlund, Faleh Menawer R Althiyabi

TL;DR

The study tackles domain shift in carotid ultrasound images across different devices by proposing a dual-discriminator GAN that translates source textures and reverberation patterns to match a target domain while preserving anatomical content. The approach introduces a reverberation loss and a content loss to complement adversarial training, and is evaluated against CycleGAN on two datasets, showing improvements in BD and HC metrics and qualitative SSIM analyses. Results indicate robust cross-device adaptation with smoother training and better generalization, reducing the need for device-specific retraining. This method offers practical utility for reliable automated ultrasound analysis across varied imaging systems and parameter settings.

Abstract

Deep learning has been extensively used in medical imaging applications, assuming that the test and training datasets belong to the same probability distribution. However, a common challenge arises when working with medical images generated by different systems or even the same system with different parameter settings. Such images contain diverse textures and reverberation noise that violate the aforementioned assumption. Consequently, models trained on data from one device or setting often struggle to perform effectively with data from other devices or settings. In addition, retraining models for each specific device or setting is labor-intensive and costly. To address these issues in ultrasound images, we propose a novel Generative Adversarial Network (GAN)-based model. We formulated the domain adaptation tasks as an image-to-image translation task, in which we modified the texture patterns and removed reverberation noise in the test data images from the source domain to align with those in the target domain images while keeping the image content unchanged. We applied the proposed method to two datasets containing carotid ultrasound images from three different domains. The experimental results demonstrate that the model successfully translated the texture pattern of images and removed reverberation noise from the ultrasound images. Furthermore, we evaluated the CycleGAN approaches for a comparative study with the proposed model. The experimental findings conclusively demonstrated that the proposed model achieved domain adaptation (histogram correlation (0.960 (0.019), & 0.920 (0.043) and bhattacharya distance (0.040 (0.020), & 0.085 (0.048)), compared to no adaptation (0.916 (0.062) & 0.890 (0.077), 0.090 (0.070) & 0.121 (0.095)) for both datasets.

Domain Adaptation of Carotid Ultrasound Images using Generative Adversarial Network

TL;DR

The study tackles domain shift in carotid ultrasound images across different devices by proposing a dual-discriminator GAN that translates source textures and reverberation patterns to match a target domain while preserving anatomical content. The approach introduces a reverberation loss and a content loss to complement adversarial training, and is evaluated against CycleGAN on two datasets, showing improvements in BD and HC metrics and qualitative SSIM analyses. Results indicate robust cross-device adaptation with smoother training and better generalization, reducing the need for device-specific retraining. This method offers practical utility for reliable automated ultrasound analysis across varied imaging systems and parameter settings.

Abstract

Deep learning has been extensively used in medical imaging applications, assuming that the test and training datasets belong to the same probability distribution. However, a common challenge arises when working with medical images generated by different systems or even the same system with different parameter settings. Such images contain diverse textures and reverberation noise that violate the aforementioned assumption. Consequently, models trained on data from one device or setting often struggle to perform effectively with data from other devices or settings. In addition, retraining models for each specific device or setting is labor-intensive and costly. To address these issues in ultrasound images, we propose a novel Generative Adversarial Network (GAN)-based model. We formulated the domain adaptation tasks as an image-to-image translation task, in which we modified the texture patterns and removed reverberation noise in the test data images from the source domain to align with those in the target domain images while keeping the image content unchanged. We applied the proposed method to two datasets containing carotid ultrasound images from three different domains. The experimental results demonstrate that the model successfully translated the texture pattern of images and removed reverberation noise from the ultrasound images. Furthermore, we evaluated the CycleGAN approaches for a comparative study with the proposed model. The experimental findings conclusively demonstrated that the proposed model achieved domain adaptation (histogram correlation (0.960 (0.019), & 0.920 (0.043) and bhattacharya distance (0.040 (0.020), & 0.085 (0.048)), compared to no adaptation (0.916 (0.062) & 0.890 (0.077), 0.090 (0.070) & 0.121 (0.095)) for both datasets.
Paper Structure (22 sections, 8 equations, 9 figures, 4 tables)

This paper contains 22 sections, 8 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Literature Survey
  3. Method
  4. Proposed GAN
  5. Adversarial losses
  6. Content loss
  7. Reverberation loss
  8. Full Objective Function
  9. Implementation
  10. Generator Architecture
  11. Discriminator Architecture
  12. Training Details
  13. Evaluation Method
  14. Dataset
  15. Results
  16. ...and 7 more sections

Figures (9)

  • Figure 1: Problem Formulation. A: Image set from the target domain. A': Image set from the source domain. B: Background noise from A. B': Background noise from A'. C: Restored A' as clean and target domain image. S: Source domain. T: Target domain
  • Figure 2: Proposed model architecture.
  • Figure 3: Sample images from (a) dataset1 for domain adaptation from A to C. (b) dataset2 for domain adaptation from B to C.
  • Figure 4: Example result images from Dataset1. (a) Sample image from domain C (b) Input Image from domain A (c) Translated images from A to C by CycleGAN (d) Translated images from A to C by ProposedGAN.
  • Figure 5: Illustration of result images from dataset1 with SSIM measure to access the domain adaptation performance. (a) Input sample image. (b) Translated image from domain B to C. (c) SSIM map between (a) and (b). (d) Target domain image from domain C. (e) Translated image from domain B to C. (f) SSIM map between (d) and (e).
  • ...and 4 more figures