Detection of COVID-19 and its pulmonary stage using Bayesian hyperparameter optimization and deep feature selection methods

Abstract

Since the first case of COVID-19 was reported in December 2019, many studies have been carried out on artificial intelligence for the rapid diagnosis of the disease to support health services. Therefore, in this study, we present a powerful approach to detect COVID-19 and COVID-19 findings from computed tomography images using pre-trained models using two different datasets. COVID-19, influenza A (H1N1) pneumonia, bacterial pneumonia and healthy lung image classes were used in the first dataset. Consolidation, crazy-paving pattern, ground-glass opacity, ground-glass opacity and consolidation, ground-glass opacity and nodule classes were used in the second dataset. The study consists of four steps. In the first two steps, distinctive features were extracted from the final layers of the pre-trained ShuffleNet, GoogLeNet and MobileNetV2 models trained with the datasets. In the next steps, the most relevant features were selected from the models using the Sine-Cosine optimization algorithm. Then, the hyperparameters of the Support Vector Machines were optimized with the Bayesian optimization algorithm and used to reclassify the feature subset that achieved the highest accuracy in the third step. The overall accuracy obtained for the first and second datasets is 99.46% and 99.82%, respectively. Finally, the performance of the results visualized with Occlusion Sensitivity Maps was compared with Gradient-weighted class activation mapping. The approach proposed in this paper outperformed other methods in detecting COVID-19 from multiclass viral pneumonia. Moreover, detecting the stages of COVID-19 in the lungs was an innovative and successful approach.

Keywords: Bayesian optimization; H1N1 viral pneumonia; bacterial pneumonia; chest CT findings; feature selection; occlusion sensitivity maps; sine–cosine optimization.

FIGURE 1

Percentage distributions of the images. (a) CPBH dataset, and (b) CovCT‐findings dataset

FIGURE 2

Computed tomography (CT) images from the CPBH dataset used in the experimental analysis of this study. (a) COVID‐19 pneumonia, (b) H1N1 viral pneumonia, (c) bacterial pneumonia, (d) healthy lung images

FIGURE 3

Computed tomography (CT) images of COVID‐19 pulmonary stages from the CovCT‐findings dataset used in the experimental analysis of this study. (a) Crazy‐paving‐pattern (b) consolidation, (c) ground‐glass opacities, (d) ground‐glass opacities and nodule, (e) ground‐glass opacities and consolidation

FIGURE 4

Effects of sine and cosine in Equations (7) and (8) on the next position

FIGURE 5

Representation of three classes of linear separable multi‐class support vector machines

FIGURE 6

Overall block diagram of SCA‐BayeSVMNet multi‐class approach

FIGURE 7

Training and validation accuracy graphs of the CPBH dataset with convolutional neural network (CNN) models. (a) GoogLeNet, (b) MobileNetV2 and (c) ShuffleNet

FIGURE 8

Confusion matrix of the pre‐trained models using Softmax on CPBH dataset. (a) GoogLeNet, (b) MobileNetV2 and (c) ShuffleNet

FIGURE 9

Confusion matrix of the pre‐trained models using support vector machine (SVM) on CPBH dataset. (a) GoogLeNet, (b) MobileNetV2 and (c) ShuffleNet

FIGURE 10

Confusion matrix of the pre‐trained models using support vector machine (SVM) and sine‐cosine algorithm (SCA) method on CPBH dataset. (a) GoogLeNet, (b) MobileNetV2, (c) ShuffleNet, (d) combined 515 features and (e) 5‐fold cross validation for combined 515 features

FIGURE 11

(a) Min classification error graph using Bayesian optimization algorithm with combined 515 features, (b) best point hyperparameter optimization confusion matrix of 515 combined features

FIGURE 12

Training and validation accuracy graphs of the CovCT‐findings dataset with convolutional neural network (CNN) models. (a) GoogLeNet, (b) MobileNetV2 and (c) ShuffleNet

FIGURE 13

Confusion matrix of the pre‐trained models using Softmax on CovCT‐findings dataset. (a) GoogLeNet, (b) MobileNetV2 and (c) ShuffleNet

FIGURE 14

Confusion matrix of the pre‐trained models using support vector machine (SVM) on CovCT‐findings dataset. (a) GoogLeNet, (b) MobileNetV2 and (c) ShuffleNet

FIGURE 15

Confusion matrices from classification of CovCT‐findings dataset using the sine‐cosine algorithm (SCA) method. (a) GoogLeNet, (b) MobileNetV2, (c) ShuffleNet, (d) combined 683 features, (e) 5‐fold cross validation for combined 683 features

FIGURE 16

(a) Min classification error graph using Bayesian optimization algorithm with combined 683 features, (b) best point hyperparameter optimization confusion matrix of 683 combined features

FIGURE 17

Comparison of occlusion sensitivity maps and grad‐CAM in differentiating multifocal lesions in COVID‐19 and CovCT findings classes. (a) Ground‐glass opacities, (b) ground‐glass opacities and nodule (c) crazy‐paving‐pattern (d) consolidation, (e) bacterial pneumonia (f) H1N1 viral pneumonia, (g) healthy lung