Classification of Lung Diseases on Chest CT Images Using Convolutional Neural Networks

Abstract

Lung diseases are amongst the most prevalent causes of death around the world. As early indicators of lung disease are difficult to foresee, Computed Tomography (CT) scans are generally used to diagnose lung ailments because, they provide a complete picture of the body's numerous lung abnormalities. Though CT is favoured over other techniques, the visual interpretation of CT scan images is a potentially error-prone process that delays disease identification. It is also a challenge to use technology to evaluate images for disease identification. Hence to overcome these difficulties, a Convolutional Neural Network (CNN) model for classifying lung diseases is presented in this paper. The tests were conducted on an open dataset that included CT images of normal, malignant, pneumonia, and pulmonary embolism collected from different patients. In these experiments, CNNs are utilized for feature extraction and classification. Image augmentation techniques have been used to improve the classification accuracy during training. The dataset is used to train pre-trained models ResNet50, AlexNet, and VGG-16. Finally features attained from the last fully connected layer of CNN are given as input to Support Vector Machine (SVM) machine learning model, to attain the best classification performance. The combination of ResNet50 model and SVM classifier provided an accuracy of 99.90%.