Deep ensemble framework for robust classification of pulmonary diseases from radiographs

The purpose of the work is to design and develop a chest X-ray images classification and lungs related disease diagnosis system using deep learning. The study uses the Radiography Database, which is open to the public, to cope with the problem of class imbalance by utilising stratified sampling. The proposed system using an hybrid methods involving three pre-trained convolutional neural network (CNN) architectures: ResNet-50, VGG-16, and EfficientNet-B0 further fine-tuned with customised classification heads to meet the needs of multi-class medical imaging. VGG-16 and EfficientNet-B0 do better than ResNet-50 on a distinct validation set when it comes to precision, recall, and F1-score. The ensemble technique works better when compared with individual methods, with an overall accuracy of 96.47% and a macro-averaged F1-score of 96.53% showing deep learning approaches based on ensembles work well for accurately and reliable clinicians’ judgements.