Chest X-Ray Image Analysis to Augment the Decision Making in Diagnosing Pneumonia using Convolutional Neural Networks Algorithm

Abstract

Pneumonia is a rarely diagnosed disease amongst severe pre-eclampsia diseases, which are the frequent reasons for pregnant woman's morbidity and mortality in Indonesia. According to the findings of “CheXNet”, which runs Convolutional Neural Networks (CNN) to analyze Chest X-Ray (CXR) image automatically while achieving the state of the art success in detecting numerous chest's disease. The ChestX-ray14 dataset from the National Institutes of Health (NIH) Clinical Center which is located in the United States. Originally, the data consisted of 112,120 frontal CXR images with 15 disease labels from 30,805 unique patients. Transforming is the first process conducted in this research, followed by dividing the dataset into two groups, positive and negative. After transforming, observed, and found out an unbalanced dataset. The proportion of positive labels was 2,516 images or 2.24% of total data. The research method consists of training data using cross-validation and testing data using the holdout method. AUROC analysis on the training results of the best re-sampling dataset is obtained at modification 16 at 5% dataset - 80–20 holdout. The mean of AUROC value is 0.9086, with a sensitivity value of 83.514%, and a specificity value of 17.273%.