Automated Endoscopic Image Classification via Deep Neural Network With Class Imbalance Loss

Recently, many computer-aided diagnosis (CAD) methods have been proposed to help physicians automatically classify endoscopic images. However, most existing methods often result in poor performance, especially for the minority classes, when the dataset is imbalanced. In this article, we propose a ne...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on instrumentation and measurement 2023, Vol.72, p.1-11
Main Authors: Yue, Guanghui, Wei, Peishan, Liu, Yun, Luo, Yu, Du, Jingfeng, Wang, Tianfu
Format: Article
Language:English
Subjects:
Artificial neural networks
Automation
Classification
Datasets
Deep neural network (DNN)
Endoscopes
endoscopic image classification
Endoscopy
Feature extraction
Image classification
Image color analysis
Image recognition
imbalanced learning
Lesions
loss function
Neural networks
Task analysis
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recently, many computer-aided diagnosis (CAD) methods have been proposed to help physicians automatically classify endoscopic images. However, most existing methods often result in poor performance, especially for the minority classes, when the dataset is imbalanced. In this article, we propose a new CAD method for automated endoscopic image classification by introducing a novel class imbalance (CI) loss to the classical deep neural network (DNN). Specifically, we use the DNN to extract rich feature representations. Given that the majority class usually dominates the prediction error and influences the gradient of the network, the proposed CI loss considers both the class frequency and prediction probability of the ground-truth class to assign the weight of each sample and helps the minority classes contribute more to descending the gradient in the training process than the majority classes. Thanks to the CI loss, the network pays more attention to minority classes and hard samples. To verify the effectiveness of our proposed method, we conduct comprehensive experiments in the binary-class classification task on our collected polyp recognition dataset (22935 images) and in the multiclass classification task on the public Hyper-Kvasir dataset (10662 images). Experimental results show that our method is competent for the imbalanced endoscopic image classification task with good performance.
ISSN:0018-9456
1557-9662
DOI:10.1109/TIM.2023.3264047