ResNet50 Pretrained Model Based Pneumonia Detection System

Abstract

Pneumonia among other pulmonary diseases is considered to one of the major causes of death around the globe. For rapid intervention and better treatment outcomes, it is crucial that this disorder is timely and accurately detected. The X-Ray based approach to detection is tedious and result interpretation are characterized by differences in interpretation and opinions. Clinical workflow and integration of AI-based technologies for pneumonia and lung disease identification are in their infancy. This study is focused on the application of convolutional neural networks (CNN) in the classification and detection of pneumonia. The goal is to create user friendly end-to-end closed system, interfaces and workflows with an underpinning machine learning technique, making it easier for both physicians and patients to use. Django framework is employed in the development of the overall system and integration. The efficiency of the proposed system and techniques has been tested on a large data set generated from various publicly available datasets of diverse image sizes and quality. The test conducted shows that the system achieved good results and generalization. Accuracy scores of 98.9%, Receiver Operating characteristic (ROC) of 0.99, and F1 Score of 0.95 were achieved after we trained the model with ResNet50 using deep learning approaches.