An engineering approach for detection of pneumonia disease: Development of a deep learning based decision support software

Pneumonia is a relatively frequent lung disease that can cause death. Early detection of this disease is critical for lowering the number of deaths. X-ray imaging is one of the most widely utilized diagnostic methods. Diseases can be detected at a high rate using X-rays taken from the chest area of the human body. Many studies are being conducted to improve diagnosis accuracy, particularly those involving deep learning methodologies. The aim of this thesis study is to present a high-accuracy deep learning-based approach for detecting pneumonia and develop decision-support software that will benefit field experts. The experimental studies used two publicly available datasets in the Kaggle repository. First, experimental studies were conducted with DenseNet-121, DensenNet-201, EfficientNet-B0, ResNet-50, ResNet-101, Inception-V3, and InceptionResNet-V2 pre-trained models. Then, hard voting and soft voting ensemble learning approaches, which included the five most successful models, were implemented. According to the findings, the soft voting approach outperformed others, with accuracies of 98,55% and 97,26% in two-class and three-class datasets, respectively. In this context, software containing this approach was developed to assist field experts with decision-making.