Deep Learning Based Pneumonia Classification

Abstract

The timely and accurate diagnosis of respiratory diseases from chest X-ray (CXR) images is critical for effective patient treatment. While cloud-based Artificial Intelligence (AI) has shown remarkable success in medical image analysis, it often suffers from limitations such as high latency, dependency on internet connectivity, and data privacy concerns. This paper proposes an AI AI framework to overcome these challenges by deploying deep learning models directly on resource-constrained AI devices for on-site CXR classification. We evaluate the performance of two prominent deep convolutional neural network (CNN) architectures, VGG-16 and ResNet-101, for identifying anomalies in CXR images. The models were trained on a large public dataset and subsequently optimized for AI deployment. Our experimental results show that the VGG-16 model achieved a classification accuracy of 93.0%, while the ResNet-101 model achieved 91.2%. The study demonstrates the viability of deploying sophisticated deep learning models on AI devices for rapid, private, and offline medical diagnostics, highlighting a crucial trade-off between model accuracy and computational efficiency in real-world applications.