Automated classification of emphysema using data augmentation and effective pixel location estimation with multi-scale residual network

In modern medical diagnosis also, the emphysema is still recognized by the computed tomography (CT) scans with a set of defined patterns as a classification problem in computer vision. There were as many algorithms developed in the past that attempt to classify the underlying patterns and their rele...

Full description

Saved in:
Bibliographic Details
Published in:Neural computing & applications Vol. 34; no. 23; pp. 20899 - 20914
Main Authors: Manikandan, T., Maheswari, S.
Format: Journal Article
Language:English
Published: London Springer London 01.12.2022
Springer Nature B.V
Subjects:
Accuracy
Algorithms
Artificial Intelligence
Automation
Classification
Computational Biology/Bioinformatics
Computational Science and Engineering
Computed tomography
Computer Science
Computer vision
Data augmentation
Data Mining and Knowledge Discovery
Emphysema
Generative adversarial networks
Image classification
Image processing
Image Processing and Computer Vision
Medical imaging
Modelling
Original Article
Pixels
Position measurement
Probability and Statistics in Computer Science
Deep learning
Relative location estimation
Emphysema classification
Absolute location estimation
Position encoding network
MS-ResNet
GAN
Lung diseases
Data augmentation
ISSN:0941-0643, 1433-3058
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In modern medical diagnosis also, the emphysema is still recognized by the computed tomography (CT) scans with a set of defined patterns as a classification problem in computer vision. There were as many algorithms developed in the past that attempt to classify the underlying patterns and their relevant associated clusters by modeling an automated system. And this classification modeling approach is responsible for the benchmarking classification and quantification of various emphysematous tissues from lung CT images on different scales in the literature. Hence, with the same motivation and intents, this article put forth a multiscale residual network with data augmentation model (MS-ResNet-DA). First, a generative adversarial network (GAN) is employed to augment the training samples and avoid the overfitting problem. These images are again augmented based on different image processing methods. Then, the obtained images are learned by MS-ResNet to categorize the emphysema. Still, the accuracies of categorizing the centrilobular emphysema (CLE) and panlobular emphysema (PLE) are not satisfactory because they do not have spatial dependence. So, an enhanced MS-ResNet-DA (EMS-ResNet-DA) model is proposed, which applies an effective position estimation algorithm to measure relative and absolute location data of emphysema pixels in the images. The relative location data give the current location of the emphysema pixel by extracting the relative dislocation measures from CT images. Also, the absolute location estimation model is based on the position encoding network to match the diseased image with the reference emphysema images and validate whether location data are implicitly learned when trained on categorical labels. Moreover, these location data of all pixels in the images are learned by the MS-ResNet for emphysema classification. Finally, the experimental results demonstrated that the EMS-ResNet-DA achieves an overall classification accuracy of 94.6% that outclasses the conventional models.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0941-0643
1433-3058
DOI:10.1007/s00521-022-07566-x