An Ensemble Model based on Deep Learning for Lung Disease Prediction using Chest XRay Images

The main objective of the framework is to improve the rapid identification and classification of critical respiratory conditions such as pneumonia, tuberculosis, asthma, Chronic Obstructive pulmonary diseases (COPD) and COVID-19 by utilizing the chest X -Rays and Computerized Tomography (CT) scan images. Automated analysis of lung disorders from high dimensional data is done by utilizing the modern deep learning technology to enhance the diagnostic capabilities. The chest Xrays (CXR) and CT scan images are used to extract the clinical information, and the data is processed using convolutional neural networks (CNNs) along with a pre-trained model to maximize the performance of the model. The performance of CNN degrades if the images contain certain degree of rotation or orientations. Therefore, a robust ensemble deep learning framework is proposed by adding different image variations through the process of data augmentation and also by fusing the capabilities of ResNet50, EfficientNet, and Inceptionv3. The proposed model extracts pertinent information and the features from these three pretrained models are concatenated to generate a composite feature representation to recognize intricate patterns from these images. The developed model outperforms standalone models with an accuracy of 93.63% and loss of 0.24. The proposed framework assists the physicians to diagnose the severity of the lung disorders and to take necessary decisions pertaining to the treatment of the disorders.