COVID-19 Disease Classification by Cough Records Analysis using Machine Learning

The rapid spreading rate of the Coronavirus disease 2019 (COVID-19) has resulted in more than 6.2 million deceased cases. Furthermore, the patients of the latest Omicron variation carry light to almost no symptoms of the disease themselves. Thus, the requirement for a new diagnosis method besides Reverse Transcription-Polymerase Chain Reaction (RT-PCR) becomes the most important step to successfully detect infected cases. In this research, the application of the KNN, Ensemble and SincNet models are implemented as the main models for classification diagnosis based on cough sound records of infected patients. After pre-processing steps for removing silence ranges in the audio scripts, the cough sounds are augmented, subsequently separated into single cough samples, then generated 3 testing scenarios for dealing with the imbalanced problem between the sample classes. Afterward, MelFrequency information and MelSprectrogram are extracted as main features for analysis in order to distinguish patients with COVID-19 disease and healthy cases. The AICV115M dataset consisting of two classes COVID-19 and NonCOVID-19 is implemented for performance evaluation. The recorded highest accuracy on the models KNN, Ensemble and SincNet are 92.49%, 90.1% and 85.15%, respectively. © 2022 IEEE.

Apply Deep Learning on Chest X-Rays Images for COVID-19 Disease Detection by Using Transfer Learning

The outbreak of Coronavirus has caused a million fatal cases recorded globally. The challenge in dealing with the SARS-CoV-2 virus is due to its patients carrying similar symptoms with common viral pneumonia. Therefore, it is essential for doctors to recognize and differentiate the infected patients of this virus in early diagnostic steps, such as using Chest X-Ray images. For that purpose, applying transfer learning with pre-trained models is considered in this work, with the aim to single out the Corona infected images from healthy lungs or other common viral pneumonia. The Curated Dataset for COVID-19 Posterior-Anterior Chest Radiography Images (X-Rays) has been applied to train and evaluate the performance of the implemented models. The dataset consists of 4 classes with a total number of thousands of images, being Normal, COVID-19, Viral - Pneumonia, and Bacterial - Pneumonia, respectively. The high accuracy recorded results from the dataset help to nominate the suitable models for early recognition of Corona infected patients, which allows early intervention and the possibility of being completely cured of the deadly virus. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.