This study detected the Coronavirus (COVID-19) disease by implementing artificial learning methods. Coronavirus disease occurs in the lungs and can cause death. The detection process was performed on chest Computed Tomography (CT) images. The training process was implemented by using 32x32 patches that were obtained from CT images. This study includes three phases: The first phase classifies patches by the SVM algorithm without implementing the feature extraction methods. The second phase extracts features on patches by using Grey Level Co-occurrence Matrix (GLCM), Grey Level Run Length Matrix (GLRLM), Grey-Level Size Zone Matrix (GLSZM), Discrete Wavelet Transform (DWT), Fast Fourier Transform (FFT), and Discrete Cosine Transform (DCT) methods and classifies the features extracted. The third phase uses Convolutional Neural Networks (CNN) method to classify the patches. 10-fold cross-validation is implemented in the classification process. The sensitivity, specificity, accuracy, precision, and F-score metrics measure the classification performance. The highest classification accuracy was achieved as 99.15% by the CNN method during the training process. The classification structure, which has the highest classification accuracy, was used during the test performance and had 80.21% mean sensitivity rate, which is the COVID detection performance, on 727 test images.
COVID-19 disease occurred at the end of 2019 at Wuhan region of China. COVID-19 disease shows fever, cough, fatigue, and myalgias in the human body during the early phases [1]. The patients have abnormal situations in their CT chest images. The respiratory problems, heart damages, and secondary infection situations were observed as complications of the disease. The findings show that the COVID-19 virus spreads from person to person. The infected person needs to be treated in the intensive care unit. Infected people have serious respiratory problems. The CT images of the infected people show that COVID-19 disease has its characteristics. Therefore, clinical experts know the characteristics and need lung CT images to diagnose the COVID-19 in the early phase. The serial CT examinations help clinical experts to understand the occurrence, development, and prognosis of the disease. CT imaging can be sorted into four stages: early stage, progressive stage, severe stage, and dissipative stage [2]. Chest CT imaging modality is one of the key elements during the diagnose of suspected patients [3]. A total of 91,636,996 cases have been diagnosed with COVID-19 infection; 65,524,142 patients have recovered, and 1,961,037 patients have died by January 12, 2021.
The development of computer vision systems supports medical applications such as increasing the image quality, organ segmentation, and organ texture classification. The analysis of time series and tumor characteristics [4], the segmentation and detection [5] of tumor modules are some of the machine learning applications in biomedical image processing field.
In the literature, there are not a detailed
study and dataset on coronavirus disease. Xu et
al. [6] classified CT images of COVID-19 into
three classes as COVID-19, Influenza-A viral
pneumonia, and healthy cases. They obtained
images from the hospitals in Zhejiang region of
China. The dataset consisted of a total of 618
images, which includes 219 images from 110
patients with COVID-19, 224 images of 224
patients with Influenza-A viral pneumonia, and
175 images of 175 healthy people. Their study
classified the images with a 3D-dimensional
deep learning model and achieved an 87.6%
overall classification accuracy. Shan et al. [7]
implemented a deep learning-based system for
segmenting and quantification of the infected
regions as well as the entire lung on chest CT
images. They used 249 COVID-19 patients and
300 new COVID-19 patients for validation in
their study. The study obtained the Dice
similarity coefficient as 91.6%. The regular
delineation system often takes 1 to 5 hours;
however, their proposed system reduced the
delineation time to four minutes. Wang et al. [
This study created patches from 202 CT images and the samples were labeled as infected/non-infected. The patches were sampled on lung and infected areas. Three different phases were used during the classification of the coronavirus images. The findings showed that the highest classification accuracy was assessed by automatic feature extraction on the images.
This paper is organized as follows. Section 2 analyses the images statistically and visually. Section 3 briefly explains the feature extraction and classification methods. Section 4 presents the classification results. Section 5 concludes the results.
The dataset was taken from [
This study classifies the patches into two classes. The first class was created by the patches that were sampled in non-infected areas. The second class was created by the patches that were sampled in infected areas. Figure 2 shows the sample images of infected areas.
This study consists of three phases. In the
first phase, the patches were transformed into
vectors and the SVM classified the vectors. In
the second phase, six different feature
extraction methods as Grey Level
Cooccurrence Matrix (GLCM) [
CNN is a biologically-inspired multi-layer perceptron (MLP) structure, which was originated by Hubel and Wiesel who worked on the mammalian visual cortex [27]. Fukushima [28] introduced CNN in 1980, then LeCun et al. [29] progressed in 1989 with a learning deep network study.
CNN has a convolution layer and automatic learning capabilities, which provide widely usage in image classification, object detection, and visual tracking applications. The progress in hardware and parallelization increases this intensive use of CNN [30]. Before the progress, the training process has taken months. After the progress, the training takes several days. The CNN structure includes one or more convolution layers, pooling layers, ReLUs, and softmax regression. The convolutional layer reduces the computational cost by reducing the number of image parameters. The pooling layer is often used to reduce matrix size without losing important features. The ReLU is an activation function and generally used following convolution or pooling layers. Finally, the softmax layer is used for the output value regression. Figure 4 shows the CNN structure used in this study.
This study presents a coronavirus classification in three phases. Phase 1 classified the patches without feature extraction. Phase 2 implemented the feature extraction process on all patches and classified the features extracted. Phase 3 used the CNN method to classify the patches. Five different evaluation metrics. These metrics are sensitivity (SEN), specificity (SPE), accuracy (ACC), precision (PRE), and F-score.
All experiments were trained on a computer with Intel Core i7-8700K CPU (3.7 GHz), 32 GB DDR4 RAM, and NVIDIA GeForce GTX 1080 graphic card.
The dataset for training consists of 202 CT images. The test set consists of 727 images. 59354 infected and 34513 non-infected patches were obtained on 202 training images. These patches were classified by using three phases. Table 2 presents the classification results obtained.
Phase Phase 1 Phase 2
Feature Extraction
x Manual
Feature Extraction
Classifier Structure
SVM FOS-SVM GLCM-SVM GLRLM-SVM GLSZM-SVM DWT-FOS-SVM FFT-FOS-SVM DCT-FOS-SVM
Table 2 shows that the classification accuracy was obtained as 56.88% in Phase 1. The best classification performance in Phase 2 was obtained with the FOS-SVM method as 97.09% with six features. The lowest performances were obtained with the GLSZMSVM structure, which detected infected patches with too low performance as a 2.92% sensitivity rate. The results show that extracting features increase the classification performance. In Phase 3, a CNN structure, which consists of a convolution layer, ReLu layer, pooling layer, and the softmax layer, was used. Table 3 shows the highest results for Phase 3.
This study used the trained structure, which has high classification accuracy, to detect the infected areas in images that were not used during the training stage. Figure 5 presents the test stage implemented.
The trained CNN structure classified the 32x32 divided patches, which were not overlapped. During the test stage, a threshold value was determined as “1”. If one patch is classified as infected, the image was classified as infected. If there were not any infected patches in the image, the image was classified as non-infected. When the threshold value was determined as “2”, the classification performance reduced. The reason is that some of images have only one patch size of infected area. When the threshold value was taken bigger than “1”, even these areas were classified as infected, the image was classified as non-infected. This caused classification performance to reduce. 727 test images were classified during test process. Table 4 presents the mean results of test process.
Table 4 shows that the proposed method detects the infected images with 80.2% mean sensitivity rate on 727 images. The noninfected patches are classified with a 99.47% mean specificity rate. The CNN structure has 99.15% classification performance during training; however, it has 80.21% performance on test images, which were not used for training.
COVID-19 was firstly encountered in Wuhan region in China and has been threatening the public health, trade, and world economy. The virus shows partially similar behaviors with other viral pneumonias. Therefore, the spreading rate of the virus made the situation difficult to be under control. CT imaging results of COVID-19 show different findings according to other clinical studies. Some situations, such as the bronchiectasis, lesion swelling symptoms, and different shadowiness in CT images provide to diagnose COVID-19, easily. This study compared manual feature extraction-based SVM and CNN that automatically extracts features, and achieved that CNN has better performance than SVM method. In this study, the coronavirus image set has a different type of images, which were acquired with different CT tools. Therefore, different feature extraction methods and classifiers were implemented to find the method that separates the infected patches. Table 5 presents the literature studies and their classification performances on different coronavirus dataset.
There are different ways to diagnose coronavirus. According to the literature studies in Table 5, CT imaging, X-ray imaging, blood test, and clinical data are used to detect the coronavirus. These datasets were examined by different artificial intelligence methods. This study used a CT dataset and CNN structure. There are not enough CT chest images to train the deep learning methods in the literature. To avoid this problem, the patch classification method was used to overcome the lack of data. If the number of COVID-19 images is increased and a dataset that has data diversity is created, high dimensional deep learning methods may be used, and a classifier structure, which gives higher detection performance, is created.
[1] C. Huang et al., "Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China," vol. 395, no. 10223, pp. 497-506, 2020. [2] M. Li et al., "Coronavirus Disease (COVID-19): Spectrum of CT Findings and Temporal Progression of the Disease," 2020. [3] L. Fan et al., "Progress and prospect on imaging diagnosis of COVID-19," pp. 1-10, 2020. [4] P. Huang et al., "Added value of computer-aided CT image features for early lung cancer diagnosis with small pulmonary nodules: a matched case-control study," vol. 286, no. 1, pp. 286-295, 2018. [5]A. Esteva et al., "Dermatologist-level classification of skin cancer with deep neural networks," vol. 542, no. 7639, pp. 115-118, 2017. [6] X. Xu et al., "Deep learning system to screen coronavirus disease 2019 pneumonia," 2020. [7] F. Shan+ et al., "Lung Infection Quantification of COVID-19 in CT Images with Deep Learning," 2020. [21] A. S. M. Sohail, P. Bhattacharya, S. P. Mudur, and S. Krishnamurthy, "Local relative GLRLMbased texture feature extraction for classifying ultrasound medical images," in 2011 24th Canadian Conference on Electrical and Computer Engineering (CCECE), 2011: IEEE, pp. 001092-001095. [22] G. Thibault, J. Angulo, and F. J. I. T. o. B. E.
Meyer, "Advanced statistical matrices for texture characterization: application to cell classification," vol. 61, no. 3, pp. 630-637, 2013. [23] M. J. J. I. T. o. s. p. Shensa, "The discrete wavelet transform: wedding the a trous and Mallat algorithms," vol. 40, no. 10, pp. 24642482, 1992. [24] H. J. Nussbaumer, "The fast Fourier transform," in Fast Fourier Transform and Convolution Algorithms: Springer, 1981, pp. 80-111. [25] G. J. S. r. Strang, "The discrete cosine transform," vol. 41, no. 1, pp. 135-147, 1999. [26] S. R. Kulkarni and G. J. W. I. R. C. S. Harman, "Statistical learning theory: a tutorial," vol. 3, no. 6, pp. 543-556, 2011. [27] D. H. Hubel and T. N. J. T. J. o. p. Wiesel, "Receptive fields and functional architecture of monkey striate cortex," vol. 195, no. 1, pp. 215243, 1968. [28] K. Fukushima, "Neocognitron: A selforganizing neural network model for a mechanism of pattern recognition unaffected by shift in position," Biological Cybernetics, vol. 36, no. 4, pp. 193-202, 1980/04/01 1980, doi: 10.1007/BF00344251. [29] Y. LeCun et al., "Backpropagation Applied to Handwritten Zip Code Recognition," vol. 1, no. 4, pp. 541-551, 1989, doi: 10.1162/neco.1989.1.4.541. [30] C. Feng et al., "A Novel Triage Tool of Artificial Intelligence Assisted Diagnosis Aid System for Suspected COVID-19 pneumonia In Fever Clinics," 2020.