Ensemble Deep Learning Approach for Pneumonia Detection Using DenseNet, MobileNet, and EfficientNet with Transfer Learning

Pneumonia is still a major global health threat that calls for enhanced diagnosis to promote the well-being of those affected. In this study, we present an ensemble model that is a combination of DenseNet, MobileNet, and EfficientNet to maximize the identification of pneumonia from CXR images. This model was trained and tested using the dataset from the Kaggle Chest X-ray Images (Pneumonia) dataset which has over 5,000 images and are classified as normal or pneumonia. The pretrained models are fine-tuned using the transfer learning approach to learn different complementary features that are concatenated to produce a powerful feature representation. A fully connected layer with a sigmoid activation function is used for binary classification. By the proposed model, the test accuracy reached 99.24%, which showed the best performance compared with the separate architectures. This work enhances the development of AI based diagnostic systems for pneumonia with regards to both diagnostic accuracy and transparency.