Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Adversarial Distortion Learning for Medical Image Denoising

Morteza Ghahremani, Mohammad Khateri, Alejandra Sierra, Jussi Tohka

TL;DR

A novel adversarial distortion learning (ADL) for denoising two- and three-dimensional biomedical image data using a proposed auto-encoder called Efficient-Unet that allows generalizing the proposed method to any sort of biomedical data.

Abstract

We present a novel adversarial distortion learning (ADL) for denoising two- and three-dimensional (2D/3D) biomedical image data. The proposed ADL consists of two auto-encoders: a denoiser and a discriminator. The denoiser removes noise from input data and the discriminator compares the denoised result to its noise-free counterpart. This process is repeated until the discriminator cannot differentiate the denoised data from the reference. Both the denoiser and the discriminator are built upon a proposed auto-encoder called Efficient-Unet. Efficient-Unet has a light architecture that uses the residual blocks and a novel pyramidal approach in the backbone to efficiently extract and re-use feature maps. During training, the textural information and contrast are controlled by two novel loss functions. The architecture of Efficient-Unet allows generalizing the proposed method to any sort of biomedical data. The 2D version of our network was trained on ImageNet and tested on biomedical datasets whose distribution is completely different from ImageNet; so, there is no need for re-training. Experimental results carried out on magnetic resonance imaging (MRI), dermatoscopy, electron microscopy and X-ray datasets show that the proposed method achieved the best on each benchmark. Our implementation and pre-trained models are available at https://github.com/mogvision/ADL.

Adversarial Distortion Learning for Medical Image Denoising

TL;DR

A novel adversarial distortion learning (ADL) for denoising two- and three-dimensional biomedical image data using a proposed auto-encoder called Efficient-Unet that allows generalizing the proposed method to any sort of biomedical data.

Abstract

We present a novel adversarial distortion learning (ADL) for denoising two- and three-dimensional (2D/3D) biomedical image data. The proposed ADL consists of two auto-encoders: a denoiser and a discriminator. The denoiser removes noise from input data and the discriminator compares the denoised result to its noise-free counterpart. This process is repeated until the discriminator cannot differentiate the denoised data from the reference. Both the denoiser and the discriminator are built upon a proposed auto-encoder called Efficient-Unet. Efficient-Unet has a light architecture that uses the residual blocks and a novel pyramidal approach in the backbone to efficiently extract and re-use feature maps. During training, the textural information and contrast are controlled by two novel loss functions. The architecture of Efficient-Unet allows generalizing the proposed method to any sort of biomedical data. The 2D version of our network was trained on ImageNet and tested on biomedical datasets whose distribution is completely different from ImageNet; so, there is no need for re-training. Experimental results carried out on magnetic resonance imaging (MRI), dermatoscopy, electron microscopy and X-ray datasets show that the proposed method achieved the best on each benchmark. Our implementation and pre-trained models are available at https://github.com/mogvision/ADL.
Paper Structure (15 sections, 8 equations, 13 figures, 5 tables)

This paper contains 15 sections, 8 equations, 13 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Method
  4. Efficient-UNet
  5. Multiscale loss framework
  6. Denoiser's loss function ($\mathcal{L_G}$)
  7. Discriminator's loss function ($\mathcal{L_D}$)
  8. Training details
  9. Experimental Results
  10. Experiments with 2D data with synthetic noise
  11. Experiments with 3D data with synthetic noise
  12. Experiments with noisy data with no reference
  13. Computational complexity
  14. Ablation study
  15. Conclusion

Figures (13)

  • Figure 1: Framework of the Efficient-UNet for denoiser with the training steps. Efficient-UNet is composed of the encoder, decoder, and Content Enhancer blocks. The output of the decoder at every scale is mapped into the image domain by a Transformer block. We then enforce consistency between the outputs of the decoder and their counterparts $x$. Low-level features further contribute to the denoised image by the Content Enhancer block. When the noise level is low, the filters of this block are activated, improving the convergence with no need for high-level features.
  • Figure 2: Framework of the proposed Efficient-UNet for discriminator with the training steps.
  • Figure 3: ATW decomposition for image data. Since the 2D kernel $h$ is separable, the ATW decomposition for 3D image data is obtained by convolving three 1D cubic kernels in the x, y, and z directions. ATW is comprised of inherently low-pass filters that attenuate the side effects of noise.
  • Figure 4: A sample of each dataset used for evaluation in this study. From top to bottom and left to right: 3D Brain MRI cocosco1997brainweb, 3D knee MRI zbontar2018fastmri, chest X-ray kermany2018identifying, dermatoscopic RGB DBW86T_2018, and EM isbi2021.
  • Figure 5: Color image denoising results of different methods on dermatoscopic RGB DBW86T_2018
  • ...and 8 more figures