Vector Quantized Convolutional Autoencoder Network for LDCT Image Reconstruction with Hybrid Loss

Medical image reconstruction is the process of creating high-quality and accurate images. During acquisition, these devices capture raw measurements or signals that represent the internal structures of the human body. However, these raw measurements are often noisy or incomplete. Low-dose CT is a me...

Full description

Saved in:
Bibliographic Details
Published in:SN computer science Vol. 5; no. 1; p. 2
Main Authors: Ramanathan, Shalini, Ramasundaram, Mohan
Format: Journal Article
Language:English
Published: Singapore Springer Nature Singapore 01.01.2024
Springer Nature B.V
Subjects:
Accuracy
Algorithms
Artificial intelligence
Codes
Computed tomography
Computer Imaging
Computer Science
Computer Systems Organization and Communication Networks
COVID-19
Data compression
Data Structures and Information Theory
Datasets
Deep learning
Dictionaries
Image acquisition
Image enhancement
Image quality
Image reconstruction
Imaging techniques
Information Systems and Communication Service
Machine learning
Medical imaging
Neural networks
Noise reduction
Original Research
Pattern Recognition and Graphics
Quality assessment
Radiation
Radiation dosage
Representations
Research Trends in Computational Intelligence
Software Engineering/Programming and Operating Systems
Vision
Deep learning
Quantization
Convolutional neural network medical image reconstruction
Autoencoders
Low-dose computed tomography
ISSN:2661-8907, 2662-995X, 2661-8907
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Medical image reconstruction is the process of creating high-quality and accurate images. During acquisition, these devices capture raw measurements or signals that represent the internal structures of the human body. However, these raw measurements are often noisy or incomplete. Low-dose CT is a medical imaging technique that uses a reduced amount of radiation to obtain detailed cross-sectional images of the body. Deep learning for low-dose CT is an emerging field that utilizes advanced artificial intelligence techniques to enhance the image quality and diagnostic accuracy of CT scans acquired with reduced radiation doses. A convolutional autoencoder for low-dose CT is a specific type of deep-learning architecture designed to address the challenges of reducing radiation doses while maintaining image quality. They are effectively trained on large datasets of both low-dose and standard-dose CT images to learn patterns and features that can compensate for the noise and artifacts present in low-dose scans. This research offers a new vector quantization-based convolutional autoencoder network strategy for CT image reconstruction. In vector quantization, discrete data elements are mapped to a set of representative vectors known as codebook entries. Each data element is associated with the nearest codebook entry based on a defined distance metric. This mapping allows for the efficient representation of discrete data by replacing each element with its corresponding codebook entry. Discrete data representation is essential for the efficient storage and transmission of information during the image reconstruction task. The results’ quality is assessed based on the perceptual and bias-reducing loss functions. On the LoDoPaB-CT benchmark dataset, experimental evaluations are done. Its findings demonstrated that, in terms of quantitative and visual evaluation, respectively, the proposed network obtained better performance metric values and better noise reduction results.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2661-8907
2662-995X
2661-8907
DOI:10.1007/s42979-023-02295-x