A Deep Ensemble Transfer Learning for Pneumonia Detection from Chest X-Ray Images to Improve Health Outcome

Abstract

Pneumonia is a serious respiratory issue that requires timely and accurate diagnosis to reduce the mortality rate. Chest X-ray images are a common diagnostic tool for pneumonia assessment. Early detection of pneumonia is crucial for prompt treatment and improved patient outcomes. The conventional diagnosis of pneumonia is highly challenging because of limited medical expertise or poor radiograph (X-ray) image quality. The recent advancement of deep learning for image analysis might be resolved the problem. The proposed study ensemble has two pre-trained Convolutional Neural Networks (CNNs), including EfficientNetB3 and DenseNet121. Pre-processing is the crucial step for this investigation, where lung region cropping, Contrast Limited Adaptive Histogram Equalization (CLAHE), and augmentation are employed to enhance the image visibility. The primary aim is more focused on important areas during training, leading to improved classification performance. Transfer learning is utilized, where EfficientNetB3 is trained using a warm-up training followed by fine-tuning. The DenseNet121 is trained for end-to-end processing. Predictions from both the models are combined using a soft average ensemble technique to enhance the efficiency of the system. The experimental results achieve an accuracy of 98.21% and an Area under the Curve (AUC) score of 0.9983.The outcomes indicate that the proposed system can serve as an effective computer-aided screening tool for early pneumonia diagnosis.