The Development of Predictive Models for Pneumonia in Chest X-ray Images Using Statistical Analysis

Abstract

            This experimental research aimed to develop predictive models for pneumonia in chest X-ray images using statistical analysis. A total of 640 images were retrospective collected from October 1, 2021 - September 30, 2022 and were divided into 320 pneumonia images and 320 normal chest images. Subsequently, 192 images were randomly selected for training data and 128 images for testing data. A region of interest (ROI) of 20 x 20 pixels was drawn on the training data images, and statistical data were measured using ImageJ. The statistical analysis 9 values which included mean gray value, standard deviation, modal gray value, minimum gray level, maximum gray level, integrated density, median, skewness, and kurtosis. We compared the difference between pneumonia and normal chest images using Independent Sample t-test or Mann-Whitney U-test. We found that 8 statistical average values of pneumonia chest images were significantly higher than normal chest images (p < 0.001) except skewness. Predictive performances of 8 statistical values showed sensitivity, specificity, and accuracy of 64.06 - 94.53, 71.09 - 96.09, and 67.58 - 95.31, respectively and the total integrated density was the highest predictive performances. The 8 statistical values can predict pneumonia chest x-ray images. However, all chest x-ray images should be approved by radiologists.