Convolutional Neural Networks Model for Medical Radiographic Image Recognition COVID-19 Cases of Madagascar

Abstract

The symptoms related to COVID-19 are diverse depending on the severity of the disease. COVID-19 is responsible for a clinical picture called the coronavirus, named SARS-CoV-2 by the who, which involves multiple organ systems, including the lungs. To determine if the lungs are affected, the doctor relies on radiographic images and its interpretation requires a specialist physician. Our research work proposes an artificial intelligence-based system to replace the specialist doctor in order to provide an interpretation of the obtained image and address the problems of a shortage of qualified doctors (radiologists). Indeed, a convolutional neural network has been proposed to train data from real images for cases of patients diagnosed with COVID or not, based on real data COVID-19 in Madagascar. Various parameters of the network were adjusted to obtain an efficient neural network model. Due to a shortage of image data and the limited computing resources (CPU and memory) of our machine, and in order to achieve sufficient performance, we used the transfer learning technic, which involves reusing a pretrained model capable to classify and adapte images to our own model. Our validation shows that the obtained model provides better classification.