Hybrid deep autoencoder network based adaptive cross guided bilateral filter for motion artifacts correction and denoising from MRI

Removal of motion artifacts from medical images is essential, because the respiration by patient causes degradation in artifact removal. Therefore, getting an MRI without any artifacts and noise is challenging. Previous techniques failed to preserve the image quality during denoising, thereby affect...

Full description

Saved in:
Bibliographic Details
Published in:The imaging science journal Vol. ahead-of-print; no. ahead-of-print; pp. 1 - 16
Main Authors: Samuel, Shiju, Ochawar, Rohini S., Rukmini, M.S.S.
Format: Journal Article
Language:English
Published: Taylor & Francis 02.01.2024
Subjects:
autoencoder
bald eagle search optimization
cross-guided bilateral filtering
Deep learning
kernel selection
Motion artifact removal
Peak Signal to Noise Ratio (PSNR)
Salp swarm optimization
ISSN:1368-2199, 1743-131X
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Removal of motion artifacts from medical images is essential, because the respiration by patient causes degradation in artifact removal. Therefore, getting an MRI without any artifacts and noise is challenging. Previous techniques failed to preserve the image quality during denoising, thereby affecting the classification accuracy. To overcome such challenges, two modules are considered in this work: classification and denoising. The classification process is successfully performed by Hybrid Deep autoencoder-convolutional neural network (DAE-CNN). It classifies the original images into motion artifacts, available and normal images. Then, the available artifacts are removed using cross guided bilateral filter (CGBF). The optimal parameters required to improve the CGBF performance are selected using the hybrid optimization algorithm. The filter's performance relies on the optimal spatial and range kernels selection. Finally, the classification and denoising-based results are evaluated using the BRATS 2015 dataset and achieved 95.84% accuracy and PSNR - 46.4, which is better than existing methods.
ISSN:1368-2199
1743-131X
DOI:10.1080/13682199.2023.2196494