Empirical Evaluation of Pre-Trained Deep Learning Networks for Pneumonia Detection

Abstract

Pneumonia is a significant global health issue, characterized by a substantial mortality risk, impacting a vast number of individuals on a global scale. The quick and precise identification of pneumonia is crucial for the optimal treatment and management of this condition. This research work aims to answer the pressing need for precise diagnostic methods by using two advanced deep learning models, namely VGG19 and ResNet50, for the purpose of pneumonia detection in chest X-ray pictures. Furthermore, the present area of research is on the use of deep learning methodologies in the domain of medical image analysis, namely in the identification of pneumonia cases via the examination of chest X-ray images. The study challenge pertains to the pressing need for accurate and automated pneumonia diagnosis to assist healthcare professionals in making timely and educated judgements. The VGG19 and ResNet50 models were trained and assessed using the comprehensive RSNA Pneumonia dataset. In order to enhance their performance in the classification of chest X-ray pictures as either normal or pneumonia-affected, the models underwent rigorous training and meticulous fine-tuning. Based on the results obtained from our investigation, it was seen that the VGG19 model exhibited a notable accuracy rate of 93\%, surpassing the ResNet50 model's accuracy of 84\%. Furthermore, it is worth noting that both models demonstrated a notable level of precision, recall, and f1-scores in the identification of normal and pneumonia patients. This indicates their potential for accurately classifying such instances. Furthermore, our research findings indicate that deep learning models, namely VGG19, have a high level of efficacy in reliably detecting pneumonia via the analysis of chest X-ray pictures. These models has the capacity to function as helpful tools for expediting and ensuring the precise identification of pneumonia by healthcare practitioners.