Deep Learning-Based Pneumonia Detection Using X-Ray Images: Leveraging MongoDB for Efficient Storage and Management of the MIMIC-IV CXR Dataset

Abstract

Introduction: Pneumonia is a significant global health hazard that is very important in resource-poor countries, where the conventional methods of diagnosis (subjective interpretation of chest X-rays) could delay the diagnosis or give inaccurate interpretations.

Objectives: This research proposes a new methodology that integrates deep learning models with MongoDB, an elastic NoSQL database, to augment pneumonia diagnosis from chest X-ray images.

Methods: The system leverages the MongoDB GridFS storage scheme and optimizes the storage management of large-scale medical image datasets for seamless integration with clinical data in the imaging domain, which maximizes efficiency in diagnosis. Initial experiments with ResNet, MobileNet, and DenseNet reported disappointing results of 56%, 51%, and 56%, respectively, with the most notable reasons being overfitting and insufficient training. PneumoNet, a particular CNN, was developed to meet the issues above and has achieved 92% accuracy after fine-tuning architecture with additional epochs and output channels.

Results: This deep learning system also allows enhanced diagnostic accuracy and radiological workflow management in integration with MongoDB using PyMongo, thus enabling real-time prediction while monitoring a patient.

Conclusions: In summary this part calls for a marriage between high-end deep learning platforms and adaptive storage solutions like MongoDB to improve health outcomes and enable big-scale medical image analytics.