An Optimal Deep Ensemble Model for the Classification of Pneumonia from Chest X-Rays

Pneumonia is a serious respiratory infection that varies in severity affecting the lungs and resulting in the inflammation of the air sacs in one or both lungs. Monitoring and addressing the burden of pneumonia through effective prevention, early detection, and appropriate treatment remain key priorities in global health efforts. With the improvements in medical imaging technology and the dataset's availability, deep learning models can analyze medical images such as chest radiographs (Chest X-Rays) and computed tomography scans (CT scans) to aid in diagnosing and classifying pneumonia. In this work, the three best-suitable convolutional neural networks namely-NASNet-Large, ResNet-152V2, and DenseNet-121 have been used for the classification of pneumonia from the Chest X-Rays. To improve further accuracy, the predictions of the three pre-trained models have been combined to generate an ensemble deep-learning model. The performance of the ensemble model was compared with the different pre-trained models. The ensemble model was also evaluated with different optimizers-ADAM, RMSprop, SGD, and NADAM and results have been compared for the best performance. The deep ensemble model with RMSprop provided an accuracy of 91% and outperformed all the other models considered.