Classification Of Pneumonia, Tuberculosis, And COVID-19 on Computed Tomography Images Using Deep Learning

The accurate diagnosis of pneumonia, tuberculosis, and COVID-19 using computed tomography (CT) images is critical for radiologists. Artificial intelligence (AI) has been introduced as a tool to aid in rapid diagnosis. In this study, we evaluated four deep learning models, including AlexNet, GoogleNet, ResNet, and deep convolutional neural network (DCNN), to classify CT images of tuberculosis, pneumonia, and COVID-19. We collected 2,134 normal images, 943 images of tuberculosis, 2,041 images of pneumonia, and 3,917 images of COVID-19 from online datasets. To assess the efficiency of the models, we measured their image classification performance such as accuracy, F1 score, and area under the curve. Our performance evaluation indicated that ResNet was the highest-performing model, with the best accuracy, F1 score, and area under the curve (0.966, 0.931, 0.954, respectively). The second-best performing model was DCNN, while AlexNet and GoogleNet had the next-best performance, respectively. The deep learning models exhibit a capability that could be viewed as a substitute for predicting lung diseases and could be employed to support radiologists in CT image screening.