Fuzzy validation of Taguchi-based convolutional fuzzy neural classifier for lung cancer imaging

While deep learning technology is widely used in the field of image classification and recognition, parameter setting for convolutional neural networks is complex, and a high number of parameters make the technology difficult to apply in practice. Therefore, this study proposes a Taguchi-based convo...

Full description

Saved in:
Bibliographic Details
Published in:Multimedia tools and applications Vol. 83; no. 41; pp. 89415 - 89437
Main Authors: Chang, Tsang-Chuan, Lin, Cheng-Jian, Yang, Tang-Yun
Format: Journal Article
Language:English
Published: New York Springer US 01.12.2024
Springer Nature B.V
Subjects:
1237: Advanced Deep Learning for Computer Vision and Multimedia Applications
Accuracy
Artificial neural networks
Computed tomography
Computer Communication Networks
Computer Science
Data Structures and Information Theory
Design of experiments
Extreme values
Fuzzy logic
Fuzzy sets
Image classification
Image quality
Lung cancer
Machine learning
Medical imaging
Multimedia Information Systems
Neural networks
Parameters
Performance evaluation
Special Purpose and Application-Based Systems
Taguchi methods
Fuzzy neural network
Fuzzy validation
Computed tomography image
Lung cancer
Taguchi method
ISSN:1573-7721, 1380-7501, 1573-7721
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:While deep learning technology is widely used in the field of image classification and recognition, parameter setting for convolutional neural networks is complex, and a high number of parameters make the technology difficult to apply in practice. Therefore, this study proposes a Taguchi-based convolutional fuzzy neural classifier (T-CFNC) to classify computed tomography (CT) images. Two layers of convolution and pooling are used to extract features of the input images, and a fuzzy neural network is used to replace fully-connected neural networks to reduce the number of model parameters. To reduce cost and time, the Taguchi experimental design method determines the optimal combination of model parameters with the minimal number of experiments. The SPIE-AAPM lung CT challenge dataset was used to validate the proposed T-CFNC model. Experimental results indicate accuracy of 99.95%, a true positive rate of 99.97%, and a true negative rate of 99.94%. While the confusion matrix is commonly applied to evaluate model performance, its accuracy varies with the quality of model training, and it is easily affected by extreme values. Single and average values also incur the possibility of misjudgment. We therefore further propose fuzzy validation for performance evaluation. Results confirm the superiority of the proposed T-CFNC model in terms of lung cancer image classification over the traditional CFNC model.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1573-7721
1380-7501
1573-7721
DOI:10.1007/s11042-024-20351-3