Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cascaded Convolutional Neural Networks with Perceptual Loss for Low Dose CT Denoising

Sepehr Ataei, Javad Alirezaie, Paul Babyn

TL;DR

This work tackles low-dose CT denoising by addressing the loss of fine structural detail with standard pixel-wise losses. It proposes a cascaded architecture where the first network reconstructs normal-dose CT from low-dose input by optimizing a perceptual loss $L_p$ based on VGG-16 features, followed by a second network that predicts the MSE-based difference to refine the reconstruction, effectively preserving low-contrast details. On the AAPM Grand Challenge dataset, the cascaded perceptual-loss approach outperforms single models and purely difference-image cascades, delivering higher perceptual fidelity and improved contrast with lower computational cost. The method offers a practical, scanner-independent solution that enhances diagnostic quality for LDCT while maintaining efficiency for training and testing.

Abstract

Low Dose CT Denoising research aims to reduce the risks of radiation exposure to patients. Recently researchers have used deep learning to denoise low dose CT images with promising results. However, approaches that use mean-squared-error (MSE) tend to over smooth the image resulting in loss of fine structural details in low contrast regions of the image. These regions are often crucial for diagnosis and must be preserved in order for Low dose CT to be used effectively in practice. In this work we use a cascade of two neural networks, the first of which aims to reconstruct normal dose CT from low dose CT by minimizing perceptual loss, and the second which predicts the difference between the ground truth and prediction from the perceptual loss network. We show that our method outperforms related works and more effectively reconstructs fine structural details in low contrast regions of the image.

Cascaded Convolutional Neural Networks with Perceptual Loss for Low Dose CT Denoising

TL;DR

This work tackles low-dose CT denoising by addressing the loss of fine structural detail with standard pixel-wise losses. It proposes a cascaded architecture where the first network reconstructs normal-dose CT from low-dose input by optimizing a perceptual loss based on VGG-16 features, followed by a second network that predicts the MSE-based difference to refine the reconstruction, effectively preserving low-contrast details. On the AAPM Grand Challenge dataset, the cascaded perceptual-loss approach outperforms single models and purely difference-image cascades, delivering higher perceptual fidelity and improved contrast with lower computational cost. The method offers a practical, scanner-independent solution that enhances diagnostic quality for LDCT while maintaining efficiency for training and testing.

Abstract

Low Dose CT Denoising research aims to reduce the risks of radiation exposure to patients. Recently researchers have used deep learning to denoise low dose CT images with promising results. However, approaches that use mean-squared-error (MSE) tend to over smooth the image resulting in loss of fine structural details in low contrast regions of the image. These regions are often crucial for diagnosis and must be preserved in order for Low dose CT to be used effectively in practice. In this work we use a cascade of two neural networks, the first of which aims to reconstruct normal dose CT from low dose CT by minimizing perceptual loss, and the second which predicts the difference between the ground truth and prediction from the perceptual loss network. We show that our method outperforms related works and more effectively reconstructs fine structural details in low contrast regions of the image.

Paper Structure

This paper contains 11 sections, 3 equations, 6 figures.

Table of Contents

  1. Introduction
  2. Background
  3. Perceptual Loss
  4. Methods
  5. Denoising Model
  6. Dataset Preparation
  7. Objective Functions
  8. MSE Based Difference Image Prediction
  9. Training Details
  10. Experiments and Results
  11. Discussion and Conclusion

Figures (6)

  • Figure 1: DRL Model Architecture
  • Figure 2: Denoising result on lung test image. A) MSE-based difference image network B) DRL Network with Perceptual Loss, C) Cascade of A and B, LDCT (Low Dose CT), NDCT (Normal Dose CT), D) Two-level cascade of A
  • Figure 3: Proposed Denoising System Architecture
  • Figure 4: CNN Model Architecture Building Block
  • Figure 5: Comparing DRL vs CNN10 Architecture
  • ...and 1 more figures