Texture Matching GAN for CT Image Enhancement

Madhuri Nagare,Gregery T. Buzzard,Charles A. Bouman

2023-12-21

Abstract:Deep neural networks (DNN) are commonly used to denoise and sharpen X-ray computed tomography (CT) images with the goal of reducing patient X-ray dosage while maintaining reconstruction quality. However, naive application of DNN-based methods can result in image texture that is undesirable in clinical applications. Alternatively, generative adversarial network (GAN) based methods can produce appropriate texture, but naive application of GANs can introduce inaccurate or even unreal image detail. In this paper, we propose a texture matching generative adversarial network (TMGAN) that enhances CT images while generating an image texture that can be matched to a target texture. We use parallel generators to separate anatomical features from the generated texture, which allows the GAN to be trained to match the desired texture without directly affecting the underlying CT image. We demonstrate that TMGAN generates enhanced image quality while also producing image texture that is desirable for clinical application.

Image and Video Processing,Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

### Problems Addressed by the Paper

This paper aims to address the issue of texture matching in low-dose CT image enhancement. Specifically:

1. **Noise Reduction and Image Enhancement**:
- Deep Neural Networks (DNN) are typically used to reduce noise in X-ray Computed Tomography (CT) images and sharpen the images, thereby reducing the X-ray dose received by patients while maintaining image quality. However, simply applying DNN methods may result in undesirable image textures in clinical applications.

2. **Generating Appropriate Textures**:
- Generative Adversarial Networks (GAN) can generate suitable textures, but simply applying GAN may introduce inaccurate or even unrealistic image details (referred to as "hallucinations"). To address this, the authors propose a Texture Matching Generative Adversarial Network (TMGAN) that can enhance CT images while generating image textures that match the target textures.

3. **Separating Anatomical Features and Textures**:
- By separating anatomical features and generated textures through parallel generators, the GAN can be trained to match the required textures without affecting the underlying CT image. Experimental results show that TMGAN improves image quality while generating textures suitable for clinical applications.

### Main Contributions

- Proposed a new TMGAN architecture that can:
- Generate desired target textures while denoising or sharpening CT images;
- Separate the underlying clean CT image by subtracting noise textures from two conditionally independent estimates;
- Use a novel bias reduction method to reduce bias in the estimated images.