Prediction of Pneumonia from Chest X-Ray Images Using Pre-trained Convolutional Neural Networks

Abstract

Machine learning plays a major role in the realm of the medical field to improve the delivery of healthcare services from medication disclosure to clinical practices. Wide usage of machine learning in the medical field has cut down treatment costs and facilitates the opportunity for new treatment procedures, etc. In recent trends, machine learning algorithms used to predict chest diseases such as tuberculosis, pneumonia, asthma, pulmonary, lung cancer diseases, etc., using chest X-ray (CXR) images. In this proposed work, CXR images are used to detect pneumonia in the chest. In this work, authors have proposed four handcrafted convolutional neural networks (CNN) and evaluated the performance of these models on three benchmark datasets. To realize the training time limitation issue of CNN, the new model is derived using a transfer-learning-based pre-trained neural networks. In this work, the customized classifier is designed and developed using VGG16, and compared its performance with handcraft CNN on those three benchmark datasets. From the experimentation, it is clear that the proposed approach outperforms with an accuracy of 82, 86, and 97% on Montgomery County, Shenzhen, and Kaggle’s chest X-ray datasets, respectively.