key: cord-1032896-yz3w8eu6 authors: Sun, Junding; Li, Xiang; Tang, Chaosheng; Wang, Shui-Hua; Zhang, Yu-Dong title: MFBCNNC: Momentum factor biogeography convolutional neural network for COVID-19 detection via chest X-ray images [Image: see text] date: 2021-09-15 journal: Knowl Based Syst DOI: 10.1016/j.knosys.2021.107494 sha: b35c3fb63ae758d8e3dd97ea10cb8143f67acbc5 doc_id: 1032896 cord_uid: yz3w8eu6 AIM: By October 6, 2020, Coronavirus disease 2019 (COVID-19) was diagnosed worldwide, reaching 3,355,7427 people and 1,037,862 deaths. Detection of COVID-19 and pneumonia by the chest X-ray images is of great significance to control the development of the epidemic situation. The current COVID-19 and pneumonia detection system may suffer from two shortcomings: the selection of hyperparameters in the models is not appropriate, and the generalization ability of the model is poor. METHOD: To solve the above problems, our team proposed an improved intelligent global optimization algorithm, which is based on the biogeography-based optimization to automatically optimize the hyperparameters value of the models according to different detection objectives. In the optimization progress, after selecting the immigration of suitable index vector and the emigration of suitable index vector, we proposed adding a comparison operation to compare the value of them. According to the different numerical relationships between them, the corresponding operations are performed to improve the migration operation of biogeography-based optimization. The improved algorithm (momentum factor biogeography-based optimization) can better perform the automatic optimization operation. In addition, our team also proposed two frameworks: biogeography convolutional neural network and momentum factor biogeography convolutional neural network. And two methods for detection COVID-19 based on the proposed frameworks. RESULTS: Our method used three convolutional neural networks (LeNet-5, VGG-16, and ResNet-18) as the basic classification models for chest X-ray images detection of COVID-19, Normal, and Pneumonia. The accuracy of LeNet-5, VGG-16, and ResNet-18 is improved by 1.56%, 1.48%, and 0.73% after using biogeography-based optimization to optimize the hyperparameters of the models. The accuracy of LeNet-5, VGG-16, and ResNet-18 is improved by 2.87%, 6.31%, and 1.46% after using the momentum factor biogeography-based optimization to optimize the hyperparameters of the models. CONCLUSION: Under the same experimental conditions, the performance of the momentum factor biogeography-based optimization is superior to the biogeography-based optimization in optimizing the hyperparameters of the convolutional neural networks. Experimental results show that the momentum factor biogeography-based optimization can improve the detection performance of the state-of-the-art approaches in terms of overall accuracy. In future research, we will continue to use and improve other global optimization algorithms to enhance the application ability of deep learning in medical pathological image detection. with an accuracy of 78%, a sensitivity of 80%, a specificity of 53%, and AUC of 71%. In the above 84 literature, the accuracy of CNN models is lower than that of using the more complex network models. Although the detection accuracy of ResNet-50 and UNet++ is relatively high, they are only verified on 86 the binary classification problem. The binary classification problem is relatively simple. It has practical 87 significance only for the detection of COVID-19 and Normal by the chest X-ray images. It has less 88 reference value for detection COVID-19, Normal, and Pneumonia by the chest X-ray images. that the accuracy of VGG-16 can be increased by 7.2% by using Bayesian inference. Feng et al. [13] 103 used VB-Net to segment the chest images and applied a random forest method based on the size of the 104 infected area to the segmented image to classify the COVID-19 images. Zheng et al. [14] proposed a 105 model using a 3D deep neural network to detecting CT images of COVID-19. Although many 106 optimization methods are used in the models mentioned above, naive Bayes needs to satisfy the 107 assumption that the distribution is independent. When the random forest is faced with the problems that 108 more decision trees are needed, the time complexity and the space complexity are relatively large. If the 109 samples have large noise, the random forest algorithm is prone to overfitting. Besides, the performance 110 of the above models can only be better when dealing with the same datasets. When the category of the 111 data changes, the accuracy of the models will be affected. Therefore, the above models have low 112 robustness and poor generalization ability. To solve the above problems and improve the accuracy of the CNNs on the dataset of the chest X-114 ray images detection for COVID-19, Normal, and Pneumonia. The momentum factor biogeography-115 based optimization is proposed to optimize the hyperparameters of three convolutional neural networks 116 The samples of our dataset are shown in Figure 2 . The first row is 5 chest X-ray images of COVID-143 19. The second row is 5 chest X-ray images of Normal, and the last row is 5 chest X-ray images of It can be seen from Figure 2 To ease the understanding of this paper, Table 13 shows all variables used in our study. Table 14 159 gives the abbreviation and their full names. LeNet-5 is a simple convolutional neural network, and its structure is shown in Figure 3 . LeNet-5 173 is mainly used for handwriting recognition [15] . All the convolution kernel size in LeNet-5 is 5×5; all 174 the convolution kernel stride size is 1. All the size of the pooling kernel is 2×2, and all the stride size of 175 the pooling kernel is 2. Here, the size of the convolution kernels and the stride size of the convolution 176 kernels in the two convolutional layers in Figure 3 are the optimization objectives of using BBO and 177 MF-BBO in the following experiments. As shown in Figure 3 , each convolutional layer in LeNet-5 is followed by a pooling layer. The Figure 6 . Here, the blue curve represents the migration paths of species 231 between habitats. The remaining every single small icon represents a species habitat. To describe the algorithm more accurately, this paper introduces the following term: habitat, which 233 is used to describe the sites of species survival, reproduction, and mutation. When the species number of a habitat is 0, the immigration rate of the habitat is the highest, the 256 emigration rate is 0, and the HSI value is the lowest. When the species number of a habitat reaches the 257 maximum, the immigration rate of the habitat is 0, the emigration rate is the maximum, and the HSI value 258 is the maximum [25] . Therefore, the following formulas can be obtained. Here, represents the number of species. represents the maximum number of species. By taking the derivative of ∆t in (4), the following formula can be obtained. Here, ⃛ represents the probability of species after the derivative. For simplicity, ⃛ can be 295 expressed as the multiplication of matrix A and P shown as below. 298 Figure 9 shows the flow chart of BBO, and Table 1 shows the pseudocode of BBO. As can be seen 299 from Figure 9 and Table 1 336 Feedforward neural network uses backpropagation to optimize the parameters in the network. Assuming that input is (x, y), the loss function obtained after calculation of the feedforward neural 339 network is ℒ( ,̂). To optimize the parameters in the feedforward neural network, the following 340 formulas need to be calculated. ℒ( ,̂) 343 According to formulas (13) and (14), we still need to calculate Here, ( ) represents the identity matrix of the -layer neurons. From formulas (15), (16), and 349 (17), the following formulas can be obtained. 352 Therefore, the execution sequence of BP in feedforward neural network is as follows: first (10) 369 When BP is performed in the pooling layer [31] , the sizes of all matrices in will be restored to 371 the size before pooling. This process is usually called upsampling. Therefore, when the of the 372 pooling layer is known, the following formula should be followed when deriving the −1 of the 373 previous layer. The migration momentum factor is a variable introduced in the MF-BBO proposed in this paper. The main function of the migration momentum factor is to standardize the value of the migration SIV in 398 the migration operation. The following formula is followed for the SIV migration. optimization effects on the convolutional neural networks of this paper. Table 3 and Figure 10 show Iter19 Iter37 Iter55 Iter73 Iter91 Iter109 Iter127 Iter145 Iter163 Iter181 Iter199 Iter217 Iter235 Iter253 Iter271 Iter289 Iter307 Iter325 Iter343 Iter361 Iter379 Iter397 Iter415 Iter433 Iter451 Iter469 Iter487 Number BBO 0