Pneumonia Detection using Sequential Deep Learning

A full pneumonia detection system, based on a modified version of the VGG16 architecture of convolutional neural networks that has been perfected for chest X-ray image analysis is developed. Model achieves state-of-the-art results with diagnostic accuracy of 96.7, precision of 95.8, and recall of 97.5, surpassing existing state-of-the-art approaches based on DenseNet121, InceptionV3, ResNet50, and MobileNetV2 architectures. The model allows for visibility into decision making and learning via Grad-CAM interpretability visualizations that show pneumonia-affected areas to support explainable and human-centered AI decision support in diagnostic procedures. To provide ease of use for patients or clinicians, this have designed and developed an intuitive web application via Flask that allows users to upload chest X-rays for real-time pneumonia detection, plus educational content about preventative measures and treatments. This application has been clinically validated among healthcare providers, and the findings show significant improvements in providing pneumonia care. The improvements include an overall time to diagnosis reduction of 33, early pneumonia diagnoses rates improved by 22, and an overall cost savings of 14.8. Further, validation of the system across diverse patient groups confirmed the robustness of performance across age groups, with excellent results in pediatric patients. Additionally, the lightweight application provides a way to take pneumonia detection in radiology and provide solutions in low-resource environments with a high level of diagnostic accuracy.