Related Experiment Video
Updated: Oct 1, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
A blind medical image denoising method with noise generation network
Bo Fu1, Xiangyi Zhang1, Liyan Wang1
1School of Computer and Information Technology, Liaoning Normal University, Dalian, Liaoning, China.
Journal of X-Ray Science and Technology
|March 7, 2022
Summary
This study introduces a novel deep learning framework, the Noisy Generation-Removal Network (NGRNet), to effectively denoise low-dose CT images with unknown noise. NGRNet demonstrates superior performance in preserving image details and enhancing visual quality compared to existing methods.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Image Processing
Background:
- Unknown mixed noise significantly degrades medical image quality during acquisition.
- Existing denoising techniques often struggle with unknown noise distributions.
Purpose of the Study:
- To introduce a novel two-step deep learning framework, Noisy Generation-Removal Network (NGRNet), for denoising low-dose CT (LDCT) images with unknown real noise.
- To develop a method capable of estimating real noise distribution and performing effective denoising on LDCT images.
Main Methods:
- A two-step deep learning approach using a noise generation network and a denoising network.
- Training the noise generation network with pseudo-image pairs derived from L0 Gradient Minimization outputs and real dental CT images to estimate noise distribution.
- Constructing an almost-real noisy dataset by migrating real noise from dental CT to noise-free lung CT images for training the denoising network.
Main Results:
- NGRNet outperforms existing denoising methods on synthetic noise lung CT images, showing improved visual effects and a 0.13dB higher peak signal-to-noise ratio (PSNR).
- The method achieves the best visual denoising effect on real noisy tooth CT images.
Conclusions:
- The proposed NGRNet effectively removes unknown real noise from LDCT images.
- The framework demonstrates impressive denoising performance while retaining crucial image details.
