COVID-19 Diagnosis from Chest X-ray Images Using Deep Learning Approach

Abstract

Coronavirus (COVID-19) disease is an infectious disease caused by the newly and deadly pneumonia type identified Coronavirus2 (SARS-CoV-2). A real-time Reverse Transcription Polymerase Chain Reaction (RT-PCR) is the main method and has been regarded as the gold standard for diagnosing the COVID-19. Strict requirements and the limited supply of RT-PCR kits for the laboratory environment leads to delay in the accurate diagnosis of patients in addition to the test takes 4-6 hours to obtain the results. To tackle this problem, radiological images such as chest X-rays and CT scan could be the answer to test the COVID-19 infection rapidly and more efficiently. In this paper, an efficient proposed Convolution Neural Network (CNN) architecture model for COVID-19 detection based on chest X-ray images is presented. The proposed model is developed to provide accurate detection for binary classification (Normal vs. COVID-19), three class classification (Normal vs. COVID-19 vs. Pneumonia), and four class classification (Normal vs. COVID-19 vs. Pneumonia vs. Tuberculosis (TB)). Our proposed model produced an overall testing accuracy of 99.7%, 95.02%, and 94.53% for binary, three, and four class classifications, respectively. A comparison is made between this work and others shows the superior of this work over the others.

Explanatory classification of CXR images into COVID-19, Pneumonia and Tuberculosis using deep learning and XAI

“…Similarly, the best performing method among the existing methods has a classification accuracy of 94.53% by Qaqos et al. [30] which is not significantly higher (0.22%) compared to the accuracy of the proposed model. Even though, the proposed model has competitive performance with the model by Qaqos [30] , it is complemented with the XAI framework, which will help the model’s output more trust-able and understandable to the end-user, which is not available with Qaqos et al.…”

“… Ref Cases Classes Method Accuracy (%) XAI Params. (in Millions) [20] TB = 4248 Normal = 453 2 Transfer learning with AlexNet and GoogLeNet 85.68 No AlexNet = 61 GoogleNet = 7 [33] Normal = 8851 Covid19 = 180 Pneumonia = 6054 3 Ensemble of Xception and ResNet50 91.40 No Xception Net = 22 ResNet = 11 [52] Normal = 310 PneumoniaB = 330 PneumoniaV = 327 COVID-19 = 284 4 CNN-based CoroNet 89.60 No CNN = 33.97 [30] Normal = 1583 COVID-19 = 576 Pneumonia = 4273 TB = 155 4 Custom CNN 94.53 No CNN = 34.73 [15] Normal = 310 PneumoniaB = 330 PneumoniaV = 327 COVID-19 = 284 4 Attention based VGG 85.43 No VGG-16 = 18 VGG-19 = 21.2 …”

“…This comparison cohort also includes both kinds of DL methods: Custom-CNN implemented and trained from scratch and DL model implemented with transfer learning. Among these seven existing methods, five models ( [15] , [20] , [32] , [33] , [36] are based on transfer learning and two models ( [30] , [52] are based on custom-designed CNN. For instance, Liu et al.…”

“…Similarly, Qaqos et al. [30] , used stochastic gradient descent to train the DL model with 6587 CXR images. Proceeding with 128 × 128 images and 100 epochs, the model was able to achieve 94.53% accuracy while classifying CXR images into four classes (COVID-19, Pneumonia and TB).…”

Deep COVID-19 Recognition Using Chest X-ray Images: A Comparative Analysis

“…The presented COVIDz framework yielded a classification accuracy of 99.64% and an F-score of 99.20%. Noor and Kareem used a deep CNN to build a COVID-19 diagnosis approach that showed over 94% classification accuracies in three different tasks [14]. Meanwhile, the attention-based deep neural networks also attained benchmark performances in COVID-19 detection from X-ray images [15].…”