Batch Similarity Based Triplet Loss Assembled into Light-Weighted Convolutional Neural Networks for Medical Image Classification.

Zhiwen Huang,Quan Zhou,Xuming Zhang,Xingxing Zhu

doi:10.3390/s21030764

Abstract

In many medical image classification tasks, there is insufficient image data for deep convolutional neural networks (CNNs) to overcome the over-fitting problem. The light-weighted CNNs are easy to train but they usually have relatively poor classification performance. To improve the classification ability of light-weighted CNN models, we have proposed a novel batch similarity-based triplet loss to guide the CNNs to learn the weights. The proposed loss utilizes the similarity among multiple samples in the input batches to evaluate the distribution of training data. Reducing the proposed loss can increase the similarity among images of the same category and reduce the similarity among images of different categories. Besides this, it can be easily assembled into regular CNNs. To appreciate the performance of the proposed loss, some experiments have been done on chest X-ray images and skin rash images to compare it with several losses based on such popular light-weighted CNN models as EfficientNet, MobileNet, ShuffleNet and PeleeNet. The results demonstrate the applicability and effectiveness of our method in terms of classification accuracy, sensitivity and specificity.

Highlights

Received: 29 December 2020Medical image classification is one of the more basic and important tasks for computeraided diagnosis (CAD)
The visual geometry group (VGG) network [5] is utilized to identity papillary thyroid carcinomas in cytological images [6] and discover COVID-19 cases based on X-ray images [7]
The results show that the BSTriplet loss can guide the show that the BSTriplet can guide the convolutional neural networks (CNNs)

Summary

Introduction

Medical image classification is one of the more basic and important tasks for computeraided diagnosis (CAD). Some researchers have developed and applied lots of heavy-weighted CNNs for medical image classification [1]. The residual network (ResNet) [11] is applied to HEp-2 cell classification [12] and the quality assessment of retinal OCT images [13]. Even though these mentioned heavy-weighted models can achieve better performance in some specific applications, they have limited capabilities in many medical applications in the case of small samples. The reason lies in the fact that the effectiveness of these networks depends on the quality and quantity of training data, while there are usually not enough annotated image data to train very

Methods

Results

Conclusion