A Transfer Learning-Based Approach for Multi-class Classification of Pulmonary Diseases

The prevalence of pulmonary diseases is experiencing a notable escalation. The year 2019 witnessed a global phenomenon characterized by the swift emergence and expansion of COVID-19, which gave rise to a state of heightened concern on a worldwide scale. The timely detection of pulmonary diseases is of utmost importance due to the potential fatality associated with these conditions. The utilization of chest X-ray (CXR) is prevalent in the healthcare industry, as it offers a rapid and accurate diagnosis. This research paper presents a proposed deep learning (DL) architecture to achieve numerous classifications of COVID-19, Viral Pneumonia, as well as Normal cases. To ensure a balanced representation of data, a composite dataset was constructed by merging two publicly accessible datasets. The dataset utilized in our study constitutes a total of 4500 chest X-ray data, with an equal distribution of 1,500 images for each class. In the context of classification, our research has employed four pre-trained models, namely InceptionV3, InceptionResNetV2, ResNet101V2, and Xception. These models have been selected due to their established performance and effectiveness in various classification tasks. By utilizing these pre-trained networks, we aim to leverage their learned features and weights to enhance the accuracy and efficiency of our classification system.