Development of a Hybrid Deep Learning Model Combining U-Net and DenseNet for Enhanced Pneumonia Prediction Using Chest X-Ray Images

Abstract

In recent years, deep learning has transformed medical image analysis, offering enhanced accuracy and efficiency in disease diagnosis. Pneumonia, a critical respiratory illness, necessitates prompt and precise detection to improve patient outcomes. This study aimed to develop and evaluate a hybrid deep learning model, combining U-Net and DenseNet architectures, to advance the automated prediction of pneumonia using chest X-ray images. A dataset of 5,863 images from Kaggle, categorized into Pneumonia and Normal classes, was employed to train and validate the model. The proposed hybrid model strategically integrated U-Net's spatial localization capabilities with DenseNet's efficient feature propagation, capitalizing on the strengths of both architectures. This integration facilitated enhanced feature extraction and spatial precision, crucial for accurate classification. The model was compared to traditional architectures, including VGGNet and ResNet, and demonstrated superior performance. Evaluated through key metrics, the hybrid model achieved an accuracy of 94% and an AUC of 0.98, underscoring its clinical applicability. Future research should address these limitations by exploring architectural innovations and integrating multi-modal data to further enhance diagnostic precision and expand applicability in varied clinical settings. In conclusion, this research significantly contributes to medical image analysis by providing insights into effective model design and deployment for respiratory disease diagnosis, highlighting the hybrid model's potential to improve clinical outcomes through reliable automated diagnosis.