Convolutional neural networks applied in the detection of pneumonia by X-ray images. (16th September 2022)

• Record Type: Journal Article
• Authors: Silva, Luan; Araújo, Leandro; Ferreira, Victor; Neto, Raimundo; Santos, Adam
• Is Part Of: International journal of innovative computing and applications. Volume 13:Number 4(2022)
• Page Start: 187
• Page End: 197
• Publication Date: 2022-09-16
• Subjects: deep learning -- pattern recognition -- convolutional neural networks -- CNNs -- pneumonia -- X-ray; Evolutionary computation -- Periodicals; Neural networks (Computer science) -- Periodicals; Genetic programming (Computer science) -- Periodicals; Biologically-inspired computing -- Periodicals; Swarm intelligence -- Periodicals; Quantum computers -- Periodicals; 006.3
• Languages: English
• ISSNs: 1751-648X

Abstract

According to the World Health Organization (WHO), pneumonia kills about 2 million children under the age of 5 and is constantly estimated as the leading cause of child mortality, killing more children than AIDS, malaria, and measles together. The application of deep learning techniques for medical image classification has grown considerably in recent years. This research presents three implementations of convolutional neural networks (CNNs): ResNet50, VGG-16, and InceptionV3. These CNNs are applied to solve the classification problem of medical radiographs from people with pneumonia, as a manner to assist in the disease diagnosis. The three architectures used in this research obtained satisfactory results. The ResNet50 outperformed InceptionV3 and VGG-16, achieving the highest percentage of training and testing precision, as well as superior recall and f1-score. For the normal class, the f1-score related to ResNet50 was 88.42%, compared to 81.54% for InceptionV3 and 81.42% for VGG-16. For the pneumonia class, this metric was 95.10% against 92.82% for InceptionV3 and 92.54% for VGG-16.