Differential Privacy Practice on Diagnosis of COVID-19 Radiology Imaging Using EfficientNet Zümrüt Müftüoğlu Department of Big Data and Artificial Intelligence Applications The Presidency of the Republic of Turkey, The Digital Transformation Office Ankara, Turkey zumrut.muftuoglu@cbddo.gov.tr https://orcid.org/0000-0003-3754-7491 M. Ayyüce Kızrak Department of Big Data and Artificial Intelligence Applications The Presidency of the Republic of Turkey, The Digital Transformation Office Ankara, Turkey ayyuce.kizrak@cbddo.gov.tr https://orcid.org/0000-0001-8545-4586 Tülay Yıldırım Department of Electronics and Communication Engineering Yıldız Technical University Istanbul, Turkey tulay@yildiz.edu.tr https://orcid.org/0000-0001-9993-5583 Abstract— Medical sciences are an important application area of artificial intelligence. Healthcare requires meticulousness in the whole process from collecting data to processing. It should also be handled in terms of data quality, data size, and data privacy. Various data are used within the scope of the COVID-19 outbreak struggle. Medical and location data collected from mobile phones and wearable devices are used to prevent the spread of the epidemic. In addition to this, artificial intelligence approaches are presented by using medical images in order to identify COVID-19 infected people. However, studies should be carried out by taking care not to endanger the security of the data, people, and countries needed for these useful applications. Therefore, differential privacy (DP) application, which was an interesting research subject, has been included in this study. CXR images have been collected from COVID-19 infected 139 and a total of 373 public data sources were used for a diagnostic concept. It has been trained with EfficientNet-B0, a recent and robust deep learning model, and proposal the possibility of infected with an accuracy of 94.7%. Other evaluation parameters were also discussed in detail. Despite the data constraint, this performance showed that it can be improved by augmenting the dataset. The most important aspect of the study was the proposal of differential privacy practice for such applications to be reliable in real-life use cases. With this view, experiments were repeated with DP applied images and the results obtained were presented. Here, Private Aggregation of Teacher Ensembles (PATE) approach was used to ensure privacy assurance. Keywords—: COVID-19, deep learning, EfficientNet, X-Ray, radiology imaging, PATE, differential privacy. I. INTRODUCTION Coronavirus, which was uncovered for the first time in Wuhan, China in December 2019, was proclaimed as a new coronavirus by the World Health Organization (WHO-WHO) on 11 February 2020 and named as COVID-19 (2019-nCov) [1]. Governments take various prevention to reduce the extent of the epidemic. Some of them are to close the borders and to recommend social distance limits. However, the number of individuals affected by coronavirus continues to increase in most countries. Clinical trials, medicines, and vaccines need to be developed and implemented to ensure that the importance and challenges are also medically functional. Based on the data published in the process, while waiting for the Polymerase Chain Reaction (PCR) test time required for © IEEE 2020. This article is free to access and download, along with rights for full text and data mining, re-use and analysis diagnosis, pioneering estimates help healthcare professionals, accelerate diagnosis by data science and artificial intelligence studies. New studies are added to the literature for the diagnosis of COVID-19 using medical images. A study examines the clinical features of pneumonia patients infected with coronavirus and influenza virus and emphasizes that it is possible to record the stage of the disease based on chest CT and CXR images and other test results. CT and CXR shots are used as an auxiliary diagnostic parameter recommended by radiologists and other experts [2], [3]. Wang et al. Deep learning model uses 1119 CT images in their studies. However, these images also contain images of patients diagnosed with viral pneumonia. The total accuracy rate is calculated as 79.3%, specificity for test dataset 83%, sensitivity 67%. In this research where modified-Inception was used as the deep learning model, the accuracy of giving correct diagnosis to COVID-19 positive patients was 85.2% [4]. Zhao et al. in their study, use a convolutional neural network-based model by making use of 275 CT images. In this binary classification study, the general accuracy rate is 84.7%, while it reaches 85.3% in F1-score [5]. This study uses artificial intelligence for a successful and rapid diagnostic recommendation using CXR prior to the waiting time for tests due to the density in hospitals. Nevertheless, the focus of the study is on the privacy approaches that should be taken into consideration while producing artificial intelligence solutions by making use of medical data. This paper covers an application to draw attention to data privacy while reaching fast solutions. Differential privacy ensures that data-driven work develops in a reliable environment. It allows governments or technology companies to collect and share information about individuals while protecting the privacy of individual users. The organization of the paper is as follows. In the second section, details about the relevant COVID-19 medical imaging and dataset used are inscribed. Besides, details about the implemented deep learning and explainability model and experimental results are presented in common with the evaluation metrics. The proposed differential privacy application was introduced in section 3. In section 4, the Authorized licensed use limited to: T.C. Cumhurbaskanligi Kutuphanesi. Downloaded on July 29,2021 at 07:40:42 UTC from IEEE Xplore. Restrictions apply.