Detection and Analysis of Covid-19 using Chest X-Ray Images

Detection and Analysis of Covid-19 using Chest X-Ray Images

1 2 3

4 5 6

K. KrishnamRaju , T. Chandini , S. Vamsi , S. Tejaswarrao , T. Devika , Ch. Madhu

1

2, 3,4,5,6

Assistant Professor, Department of Electronics and Communication Engineering

Student, Department of Electronics and Communication Engineering Aditya Institute of Technology and Management,

Tekkali, Srikakulam-AP

Abstract Covid-19 is a rapidly spreading viral disease that infects not only humans, but also animals are infected because of this disease. The daily life of human beings and their health and the economy of the country are affected due to this viral disease. X-ray imaging is a tool which is easily accessible and that can be an excellent alternative in COVID- 19 diagnosis. The accuracy, sensitivity, specificity for thisanalysis was 97%, 96.3%, 97.6% respectively. This would be extremely useful in this pandemic and need for preventive measures are at odds with available resources. The aim of this paper is to propose automatic detection of covid-19 from digital chest X-ray images by applying deep-learning algorithms.

1. INTRODUCTION

   Covid-19 is a severe disease where a large number of people lose their lives every day. This disease affects not only a single country, and even the whole world suffered because of this disease. In the past decade, several kinds of viruses (like SARS, MERS, Flu,etc.) came into the picture, but they stand for only a few days. Many scientists are working on these kinds of viruses, and few are diagnosed due to the availability of vaccines prepared by them (i.e., Scientists or researchers). In the present time, the whole world is affected by Covid-19 disease, and the most important thing is no single country scientists can prepare a vaccine for the same.

   In March 2020, X-ray images of healthy people andCovid- 19 infected peoples were available in different repositories such as Github, Kaggle for analysis. Covid-19 is a disease that threatens humans at a global level and turned into a pandemic. To diagnose covid-19 patients with healthy patients is a critical task. The dialysis of Covid-19 infected patients needs more precaution and must be cured under strict procedures to reduce the risk of patients unaffected with covid-19.

   As per literature review, after analyzing the potentials of computer tomography (CT) scans and X- ray images in detecting COVID-19. There have been several studies [5]- [7] trying to develop COVID-19 classification models, with an overall accuracy of 86.7%. Consequently, There are two or three regions which are randomly selected images from the dataset. Finally, features are extracted from CNN and fed to ensemble of classifiers for COVID- 19 prediction of accuracy 88%. CT scans are also used in reference [8] to identify positive COVID-19 cases, where all images are separately fed to the model and outputs are aggregated using operation of sensitivity of 90%.

   Deep convolutional neural networks perform better with a larger dataset than a smaller one. Transfer learning can be used in the training of deep CNN.

   Large number of COVID-19 patients infected worldwide, the images of chest X-ray available in onlineare small and scattered. So, In this work we have reported a comparatively large dataset of COVID-19 positive chest X-ray images and unaffected person chest X-Ray images are used for this study. The collected dataset consists of 3000 total chest X-ray images.

2. PROPOSED METHODOLOGY

Data-Base Description

In this study, posterior-to-anterior (AP)/anterior-to- posterior (PA) images of chest X-ray were used by radiologist in clinical diagnosis. Different sub- databases are used to create one database. Normal and Pneumonia databases were collected from different websites. The methodology was explained by using block diagram as shown in fig.2.b.

COVID-19 positive Chest X-ray images fromdifferent articles

More than 1000 covid affected images were collected from different website and kept into a database. The radiography images were compared to other publicly available datasets to avoid duplication.

Pneumonia Detection

In this database, Normal chest X-ray with no lung infection and non-COVID person images as shown in fig.2.a were available.

Fig.2.a

1. Covid affected person X-ray image

2. Normal person X-ray image

2.1.3. Block diagram

cnn models 1.VGG16

2.MOBILENETV

Fig.2.b

3.. ARCHITECTURE OF VGG16

It was proposed by simmonyan and Zisserman in 2014 and the block diagram of VGG16 architecture is shown in fig.3.a. This framework has been trained on Image Net dataset that consists of 14million images and attains accuracy of 92.7%.

Most of the parameters are used in fully connectedlayers. It has total memory of 96 MB image for only forward propagations.

CONV BLOCK1

CONVOLUTION 2D

CONV BLOCK2

CONVOLUTION 2D MAXPOOLING 2D

RESULTS

COVID NORMAL

1. MOBILENETV2 ARCHITECTRE MobileNetV2 is issued as a component of tensor flow slim image classification library. MobileNetV2 is a convolutional neural network builds upon the ideas from MobileNetV1. The architecture of MobileNetV2 consists of 54 layers that takes as input. The main feature of this architecture is that it utilizes depth wise seperable convolution.

   The MobileNetV2 consists of two types of blocks. Oneblock is residual block and other block is used for downsizing. Both the block consists of three layers. Those blocks are shown in fig.4.a. Each layer in the mobilenetV2 is represented as series of one or more identical layers that are repeated for n number of times. The Number of output channels is same in all the layersthat have the same series.

   Stride=1 block

   Conv 1X1 relu

   CONV BLOCK3

   CONVOLUTION 2D CONVOLUTION 2D CONVOLUTION 2D MAXPOOLING 2D

   input

   Stride=2 block

   input

   CONV BLOCK4

   Conv 1X1 relu

   Depth wise 3X3 relu

   CONV BLOCK5

   Conv 1X1 linear

   Conv 1X1 linear

   Depth wise 3X3 relu

   FULLY CONNECTED CLASSIFIER

   FLATTEN DENSE DENSE DENSE

   add

   Fig.4.a.convolution of MobileNetV2

   Fig.3.a.Architecture of VGG16

   MobileNetV2 is an extractor for detecting the object and segmentation. It is based on an inverted residual structure where the residual connections are between the layers.

   The whole architecture of MobileNetv2 contains fully convolutional layer with 32 filters follwed by 19 residual layers.

2. DATA PREPROCESSING

   To enhance this process, the images are preprocessed utilizing various preprocessing methods. The main reason for preprocessing the images is to enhance the visual capacity of input image by increasing contrast, by deleting low or high frequencies in the image and by reducing the noise present in original image. In this paper, we applied image resizing and image normalizing techniques for enhancing the quality ofimage.

   Image Resizing

   The dimensions of the image in the dataset are of different ranges. Therefore, to achieve dimensions of all images, original images are resized to the size of pixels to fit to the image in VGG16 and MobileNetV2 pretrained models.

   Where,

   = FP ——-->eq(4)

   FP+TN

   6.RESULT AND CONCLUSION

   We validated our proposed model on COVID-19 normal chest X-Ray images and affected person chest X-ray images using VGG16 and ADAM optimizers.

   It is also significant that these machine learning techniques have been used in the traiing process improves the ability of the work.

   On comparing with other models our work approachesthe better accuracy that is 97.7%. as shown in fig.7.a.

   Image Normalization

   In this paper, we have applied the intensity normalization as preprocessing technique which is widely used in image processing applications. The

   intensity of the image is from 0 to 225 are normalized to the basic normal distribution by utilizing the min- max

   Fig.6.a

   7. REFERENCES

   normalization and is shown in eq(1). [1].H.Lu, C.W. Stratton, Y.W. tang , outbreak of pneumonia of unknown etiology in wuhan, china the

   Where,

   =

   XXmin

   ——-->eq(1)

   mystery and the miracle. J.Med virol. 92(4), 401-402 (2020).

   [2].G. litijens et al ., A survey on deep learning in medical image analysis. Med image anal 42,60-8(2017).

   X=pixel intensity

   X min and X max = minimum and maximum intensity values of input images.

   5.Accuracy

   Accuracy is the measurement tool which is used to determine which model is perfect at identifying patterns between variables in dataset based on input and data. The formula of accuracy is given in eq(2).

   TP+TN

   [3].V.M. Corman et al., Detection of 2019 novel

   coronavirus 25(3), 2000045 (2020)

   [4].S. Chavez, et al. Coronavirus Disease (COVID-19): A primer for emergency physicians. Am. J. Emergency Med. (2020).
   [5]D.P. Kingma, J. Ba. Adam: A method for stochastic http://arxiv.org/abs/arXiv:1412.6980 (2014).
   [6]G. Somua-Gyimah, et al. A computer vision system for

   = TP+FP+TN+F

   N

   6.Specificity

   ———–

   >eq(2)

   terrain recognition and object detection tasks in mining and construction environments, in SME Annual Conference (2019)

3. https://www.kaggle.com. Accessed Sep 4 2020.

   Specificity is the proportion of true negatives that are correctly represented by the model as given in eq(3) and eq(4).

   = TN —————>eq(3)

   FP+TN

4. S.H. Kassani, et al. Classification of histopathological biopsy images using ensemble of deep learning

networks. Arxiv preprint http://arxiv.org/abs/arXiv:1909.11870 (2019).