Deep Learning Applied to Chest Medical Images Classification

Abstract

Lung disease is common throughout the world. Especially these include pneumonia, covid, tuberculosis, and lung opacity; so an early and precise diagnosis of lung disease is important. For this reason, Computer Aided Diagnostic (CAD) systems using X-Rays medical images are often used to assist healthcare professionals to make more and more accurate decisions, for this purpose, many image processing and machine learning models have been proposed and developed, like deep learning techniques include convolution neural networks (CNN), Visual Geometry Group-based Neural Networks (VGG), and Residual Neural Networks (ResNet). In this work, we used three models based on ResNet50, VGG16, and VGG19 that were pre-trained using ImageNet database. As implementation tools, Jupyter Notebook, Tensorflow, and Keras are used. The three models were applied to the chest X-ray database collected from two different sources on the Kaggle repository using transfer learning. A step of pre-processing was applied to the radiography database using histogram equalization. Finally, the best accuracies results are as follows: for the case based on ResNet50 model, the best validation accuracy values were 94.5% and 93.2%, and in the case based on VGG16 model, the accuracies were 93.2% and 89.7%, and in the case based on VGG19 model the accuracies were 95.2% and 91%.