COVID-19 Identification System Using Transfer Learning Technique With Mobile-NetV2 and Chest X-Ray Images

• Main Authors: Mahmoud Ragab (Author), Samah Alshehri (Author), Gamil Abdel Azim (Author), Hibah M. Aldawsari (Author), Adeeb Noor (Author), Jaber Alyami (Author), S. Abdel-khalek (Author)
• Format: Book
• Published: Frontiers Media S.A., 2022-03-01T00:00:00Z.
• Subjects: article
• Online Access: Connect to this object online.
• Notes: 2296-2565; 10.3389/fpubh.2022.819156
• Keywords: machine learning; convolution neural networks; transfer learning; MobileNetV2; COVID-19; Public aspects of medicine; RA1-1270
• Source: Frontiers in Public Health, Vol 10 (2022)

Diagnosis is a crucial precautionary step in research studies of the coronavirus disease, which shows indications similar to those of various pneumonia types. The COVID-19 pandemic has caused a significant outbreak in more than 150 nations and has significantly affected the wellness and lives of many individuals globally. Particularly, discovering the patients infected with COVID-19 early and providing them with treatment is an important way of fighting the pandemic. Radiography and radiology could be the fastest techniques for recognizing infected individuals. Artificial intelligence strategies have the potential to overcome this difficulty. Particularly, transfer learning MobileNetV2 is a convolutional neural network architecture that can perform well on mobile devices. In this study, we used MobileNetV2 with transfer learning and augmentation data techniques as a classifier to recognize the coronavirus disease. Two datasets were used: the first consisted of 309 chest X-ray images (102 with COVID-19 and 207 were normal), and the second consisted of 516 chest X-ray images (102 with COVID-19 and 414 were normal). We assessed the model based on its sensitivity rate, specificity rate, confusion matrix, and F1-measure. Additionally, we present a receiver operating characteristic curve. The numerical simulation reveals that the model accuracy is 95.8% and 100% at dropouts of 0.3 and 0.4, respectively. The model was implemented using Keras and Python programming.