An Efficient Ensemble Deep Learning Model for Binary and Multi-Class Classification of Chest-Xray Images with Web Interface

Abstract Chest diseases like pneumonia, tuberculosis, COVID-19 are serious health related problems and dangerous if they are not detected earlier stage. Using Chest X-ray images, manually to check patients takes time and sometimes leads to errors, so an automatic and reliable system is necessary. In this work, three pre-trained models are used to create an ensemble model that is used to classify the chest X-ray images into binary and multi-class categories. The binary model, which used to predict pneumonia and normal cases, achieved accuracy of 96.63% also training accuracy of 98.04% and validation accuracy of 97.44%. The training and validation losses 6.2% and 7.7% shows that the model is stable and does not overfit. It also achieved evaluation results such as 98.36% recall, 97.45% precision, and 97.91% F1-score. The multi-class model classifies six different chest classes and achieved accuracy of 91.48% also training accuracy of 92.65% and validation accuracy of 90.50%. The training loss 18.67%and validation loss 25.17% show good learning, generalization, and evaluation results shows 91.48% recall, 91.52% precision, and 91.47% F1-score. Overall, the proposed system shows good performance and helps to obtain more accurate diagnosis.