2020 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) © IEEE 2021. This article is free to access and download, along with rights for full text and data mining, re-use and analysis. 2257 Automatic diagnosis of COVID-19 and pneumonia using FBD method Pradeep Kumar Chaudhary and Ram Bilas Pachori Discipline of Electrical Engineering Indian Institute of Technology Indore Indore - 453552, India Email: phd1801202005, pachori@iiti.ac.in Abstract—Novel coronavirus (COVID-19) is spreading rapidly and has taken millions of lives worldwide. A medical study has shown that COVID-19 affects the lungs of patients and shows the symptoms of pneumonia. X-ray images with artificial intelligence (AI) can be useful for a fast and accurate diagnosis of COVID- 19. It can also solve the problem of less testing kits and fewer doctors. In this paper, we have introduced the Fourier-Bessel series expansion-based dyadic decomposition (FBD) method for image decomposition. This FBD is used to decompose an X-ray image into subband images. Obtained subband images are then fed to ResNet50 pre-trained convolution neural network (CNN) individually. Deep features from each CNN are ensembled using operations, namely; maxima (max), minima (min), average (avg), and fusion (fus). Ensemble CNN features are then fed to the softmax classifier. In the study, a total of 750 X-ray images are collected. Out of 750 X-ray images, 250 images are of pneumonia patients, 250 of COVID-19 patients, and 250 healthy subjects. The proposed model has provided an overall accuracy of 98.6% using fus ensemble ResNet-50 CNN model. Keywords— Fourier-Bessel series expansion (FBSE), Im- age decomposition, Corona virus, Pneumonia, X-ray image. I. I NTRODUCTION Novel coronavirus (COVID-19) is the result of severe acute respiratory syndrome coronavirus 2 (SARS-COV-2). The symptoms of COVID-19 ranges from dry cough, sore throat, loss of taste, and fever to organ failure, pneumonia, and acute respiratory distress syndrome [1], [2]. Another challenge countries are facing along with COVID-19 is the lack of the testing kits. So there is a requirement of finding new ways by which the diagnosis of COVID-19 becomes fast and accurate. The studies found that X-ray and computer tomography (CT) scan images can be tools that can diagnose pneumonia caused by COVID-19 [3]. The idea is to use image processing techniques and artificial intelligence (AI) to get contact-less testing. Some research works have been done for diagnosing COVID-19 with X-ray and CT images using AI [4], [5], [6]. Xu et al. [7] extracted infected region from CT scan image using a pre-trained 3D-convolution neural network (CNN). These regions are fed to CNN for three classes (COVID- 19, Influenza-A-viral-pneumonia, and healthy) classification. The CT scans have been utilized in [8] to detect COVID- 19 cases, where all slices of CT scans are fed to the CNN model separately. The output from each model is aggregated by applying a max-pooling process. Wang and Wong [9], use a pre-train CNN model which was first trained with ImageNet dataset [10], which are then fine-tuned with X-ray images to classify subjects as COVID-19, normal, bacterial infection, and none-COVID-19 viral. Similar work was done by Sethy and Behera [11], where several CNN models are trained on X- ray images, and a support vector machine (SVM) classifier used to detect COVID-19. Research has found that a CT scan can be the better tool for COVID-19 diagnosis compared to X-ray images [3]. But the drawback of using a CT scan is that it takes more time than X-ray imaging. High-quality CT scanners are usually not available in rural or underdeveloped regions, making this time-consuming process [12]. This paper uses a new multi-resolution analysis tech- nique for image decomposition, called Fourier-Bessel se- ries expansion (FBSE) based dyadic decomposition (FBD). The method is inspired by multi-frequency scale 2D-FBSE- empirical wavelet transform (EWT) (2D-FBSE-EWT) [13], where multi-frequency scale (dyadic frequency scale) FBSE spectrum is used for boundaries detection in EWT method. FBD is used to decompose X-ray images to get subband im- ages. Each subband is fed to the ResNet-50 CNN individually. Deep features are then extracted from the last fully connected layer of each ResNet-50 CNN. These features are then ensem- bled using the operations, namely maximum (max), minimum (min), average (avg), and fusion (fus) [14]. Ensemble features are then fed to a softmax classifier to classify pneumonia caused by COVID-19 and other pneumonia. The rest of the paper is organized as follows: Section 2 presents the paper’s database and briefly introduces the FBD. The method proposed for the automated diagnosis of COVID- 19 and pneumonia is explained briefly in Section 3. Section 4 provides experimental results and discussion. Conclusion has been provided in Section 5. II. DATABASE AND PROPOSED FBD A. Database In this paper, 750 images are collected from two databases. 250 X-ray images are downloaded from URL: https://github.com/ieee8023/covid-chestxray-dataset. Images are of pneumonia caused by COVID-19 [15]. For 250 healthy and 250 viral pneumonia X-ray images, the kaggle repository database called ”Chest X-Ray Images” [16] is used. 2020 IEEE International Conference on Bioinformatics and Biomedicine (BIBM) 978-1-7281-6215-7/20/$31.00 ©2020 IEEE DOI: 10.1109/BIBM49941.2020.9313252