Processing math: 0%
Enhancing Low-Dose CT Imaging Reconstruction through Fusion of MLEM and Deep Convolutional Neural Network Priors | IEEE Conference Publication | IEEE Xplore

Enhancing Low-Dose CT Imaging Reconstruction through Fusion of MLEM and Deep Convolutional Neural Network Priors


Abstract:

The process of reconstructing low-dose CT imaging involves dealing with noisy data, which is characterized as an ill-posed inverse problem. Traditional methods like itera...Show More

Abstract:

The process of reconstructing low-dose CT imaging involves dealing with noisy data, which is characterized as an ill-posed inverse problem. Traditional methods like iterative reconstruction rely on manually designed priors. Recently, deep learning-based CT image reconstruction techniques have gained importance in filed of CT image reconstruction, but deep learning techniques face limitations due to the substantial data requirements for training and testing purposes. Our research exploit the efficiency of deep Convolutional Neural Networks (Deep CNNs) in identifying meaningful patterns amidst noise. We propose an iterative reconstruction approach for low-dose CT using the Maximum Likelihood Expectation Maximization (MLEM) algorithm, enhanced by a Deep Neural Network acting as a prior. The proposed technique demonstrates improved performance when compared to traditional methods i.e. Maximum Likelihood Expectation Maximization (MLEM), Simultaneous Algebraic Reconstruction Technique (SART), and SART+TV (Total Variation).
Date of Conference: 22-24 November 2023
Date Added to IEEE Xplore: 28 May 2024
ISBN Information:

ISSN Information:

Conference Location: Solan, India

I. Introduction

Computed tomography (CT) imaging is a diagnostic modality that offers significant clinical insights into the body’s anatomical structures, including organs, tissues, bones, and blood arteries. Mathematically CT reconstruction problem can be represented as \begin{equation*}D=A e^*+\eta \tag{1}\end{equation*}

References

References is not available for this document.