Detection of COVID-19 Disease from Chest X-Ray Images: A Deep Transfer Learning Framework Shadman Sakib 1 , Md. Abu Bakr Siddique 2 , Mohammad Mahmudur Rahman Khan 3 , Nowrin Yasmin 4 , Anas Aziz 5 , Madiha Chowdhury 6 , Ihtyaz Kader Tasawar 7 1 Department of EECS, University of Hyogo, 2167 Shosha, 671-2280, JAPAN 2 Department of EEE, International University of Business Agriculture and Technology, Dhaka-1230, Bangladesh 3 Department of ECE, Vanderbilt University, Nashville, Tennessee-37240, USA 4 Department of CSE, Ahsanullah University of Science and Technology, Dhaka-1208, Bangladesh 5 Department of AE, Military Institute of Science and Technology, Dhaka-1216, Bangladesh 6 Department of URP, Bangladesh University of Engineering and Technology, Dhaka-1000, Bangladesh 7 Department of EEE, BRAC University, Dhaka-1212, Bangladesh sakibshadman15@gmail.com 1 , absiddique@iubat.edu 2 , mohammad.mahmudur.rahman.khan@vanderbilt.edu 3 , nowrin_yasmin@outlook.com 4 , anas.aziz@ae.mist.ac.bd 5 , madihac940@gmail.com 6 , ihtyaztasawar@gmail.com 7 Abstract—World economy as well as public health have been facing a devastating effect caused by the disease termed as Coronavirus (COVID-19). A significant step of COVID-19 affected patient’s treatment is the faster and accurate detection of the disease which is the motivation of this study. In this paper, implementation of a deep transfer learning-based framework using a pre-trained network (ResNet-50) for detecting COVID-19 from the chest X-rays was done. Our dataset consists of 2905 chest X-ray images of three categories: COVID-19 affected (219 cases), Viral Pneumonia affected (1345 cases), and Normal Chest X-rays (1341 cases). The implemented neural network demonstrates significant performance in classifying the cases with an overall accuracy of 96%. Most importantly, the model has shown a significantly good performance over the current research-based methods in detecting the COVID-19 cases in the test dataset (Precision = 1.00, Recall = 1.00, F1-score = 1.00 and Specificity = 1.00). Therefore, our proposed approach can be adapted as a reliable method for faster and accurate COVID-19 affected case detection. Keywords—COVID-19 Detection, Viral Pneumonia, Chest X- ray Image Analysis, Deep Convolutional Neural Network (DCNN), Deep Transfer Learning, ResNet-50, Medical Imaging I. INTRODUCTION COVID-19, the pandemic that has brought the world to a halt, was reported in Wuhan, China in the December of 2019 for the first time, when patients with cases of unidentified pneumonia emerged. The virus responsible for the disease, named SARS-CoV-2, belongs to a family of coronaviruses that are zoonotic in nature. Until SARS-CoV-2 surfaced, six types of coronaviruses were known to be able to harm humans by mainly targeting the respiratory system. Among them, two had caused epidemics in the last two decades named SARS-CoV and MERS-CoV. Despite the mortality rates of these epidemics being much higher than that of COVID-19 (10% for SARS and 30-35% for MERS), the cumulative number of deaths for the latter has surpassed that of both the epidemics combined by many folds [1]. As of 28 June 2020, the total number of global cases and deaths exceed 9.8 million and 4.9 lakh respectively [2]. The common clinical characteristics of COVID-19 include a range of symptoms mutual with other viral diseases such as the common cold. In more severe or progressed cases, pneumonia, development of fluid in the lungs, acute respiratory distress syndrome (ARDS), multiorgan failure, septic shock as well as death may occur. Elderly people or people exhibiting comorbidity having a compromised immune system are highly prone to infection and severity. On the other hand, many individuals do not show any symptoms despite being carriers of the virus. This makes detection and containment of the virus even harder. Along with being a highly contagious disease, COVID-19 has a long incubation period, on average, five to six days between the contact and symptom onset phases. Thus, abiding by preventive measures such as social distancing, hygiene maintenance, and contact tracing and enabling a system that can diagnose the disease earlier and faster is paramount. At present, the eminent standard for diagnosing COVID- 19 is the reverse transcription-polymerase chain reaction (RT-PCR) which identifies the nucleotides of the virus from specimens extracted from a nasal swab or oropharyngeal swab. One of the major drawbacks of this method is the tedium involved and the time required as the fastest turn- around time is at least 24 hours. Added with the rapid spread and hence an increased number of specimens collected, the laboratories very rapidly get overwhelmed. Furthermore, it is laborious, relatively expensive, and has a low sensitivity (60%–70%) [3]. Many countries suffer from false results due to multiple plausible causes such as specimen handling, stage of disease when the specimen is collected, and quality of the specimen [4]. With limited resources i.e. testing kits, hospital beds and ICU beds, ventilators, personal protective equipment (PPE), the healthcare systems around most of the globe are loaded at the havoc and bound to make selective decisions in terms of testing, patient admission, ICU beds as well as the provision of ventilators. Radiography chest images (X-ray and CT scan) analysis is a valuable alternative of the PCR method. They may assist in multiple ways from diagnosing the disease to sorting out the high-risk patients for quarantining and prioritizing while selective testing to identifying the false-negative PCR cases. However, since most viral cases of pneumonia' images are akin and overlap, it is very difficult and time consuming for radiologists to distinguish the fine details by vision. Artificial Intelligence models can be a prompt solution. Very recently, the deep learning (DL) approach have been very popularly and successfully used in medical image classification applications owing to its powerful accuracy. . CC-BY-NC 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted November 12, 2020. ; https://doi.org/10.1101/2020.11.08.20227819 doi: medRxiv preprint NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice.