A Dual-Domain CNN-Based Network for CT Reconstruction

Fengyuan Jiao,Zhiguo Gui,Kunpeng Li,Hong Shangguang,Yanling Wang,Yi Liu,Pengcheng Zhang

DOI: https://doi.org/10.1109/access.2021.3079323

IF: 3.9

2021-01-01

IEEE Access

Abstract:Convolutional neural network (CNN)-based deep learning techniques have enjoyed many successful applications in the field of medical imaging. However, the complicated between-manifold projection from the projection domain to the spatial domain hinders the direct application of CNN techniques in computed tomography (CT) reconstruction. In this work, we proposed a novel CT reconstruction framework based on a CNN, i.e., an intelligent back-projection network (iBP-Net). The proposed iBP-Net method fused a pre-CNN, a back-projection layer, and a post-CNN into an end-to-end network. The pre-CNN adopted CNN techniques to model a filtering operation in the projection domain. In the back-projection layer, a back-projection operation was employed to perform between-manifold projection. Based on CNN techniques, the post-CNN worked together with the pre-CNN to recover reconstructed images from the outputs of the back-projection layer in the spatial domain while maintaining high visual sensitivity. The experimental results demonstrate the feasibility of the proposed iBP-Net framework for CT reconstruction.

computer science, information systems,telecommunications,engineering, electrical & electronic

What problem does this paper attempt to address?

### Problems Addressed by the Paper

This paper aims to address key issues in Computed Tomography (CT) reconstruction. Specifically:

1. **Limitations of Existing Methods**:
- **Analytical Methods**: For example, the Filtered Back-Projection (FBP) algorithm, while computationally efficient, suffers from significant image quality degradation when the projection data is sparse or noisy.
- **Iterative Reconstruction (IR) Methods**: Although these methods can improve image quality by incorporating prior knowledge, they are computationally expensive, especially since projection and back-projection operations need to be performed in each iteration.

2. **Challenges of Applying Deep Learning in CT Reconstruction**:
- When directly applying deep learning techniques based on Convolutional Neural Networks (CNN) to CT reconstruction, there is a complex between-manifold projection problem, i.e., the transformation from the sensor domain to the spatial domain.

3. **Proposed New Method**:
- The paper proposes a method called Intelligent Back-Projection Network (iBP-Net), which combines pre-CNN, back-projection layers, and post-CNN to construct an end-to-end neural network framework. This method avoids the use of fully connected layers, thereby reducing the difficulty of neural network training and enabling the direct learning of all parameters from large datasets.

### Main Contributions
- A new network architecture is proposed, which generates high-quality CT reconstruction images through the collaborative work of pre-CNN and post-CNN.
- The post-CNN not only acts as a denoiser but also further extracts detailed information from the output of the back-projection layer, improving image quality.
- The entire iBP-Net is an end-to-end network, with all parameters learned directly from large datasets, demonstrating excellent end-to-end learning capabilities and outstanding performance in CT reconstruction.

### Experimental Results
- Evaluations were conducted on clinical patient data, and the experimental results show that iBP-Net has good reconstruction capabilities.