A graphical method of detecting pneumonia using chest radiograph

Abstract

An important ingredient of health care is the correct initial diagnosis of chest ailments using images of chest radiograph. This paper develops a simple graphical method to aid the initial screening and discrimination of pneumonia patients (PNEU) from pulmonary tuberculosis (PTB) patients, lung cancer (LC) patients and normal healthy individuals (NL). Approximate confidence regions using principal component methods on selected texture measures detect and discriminate PNEU from PTB, LC and NL. A brief simulation study indicates that the probability ellipsoid is robust to mild deviation from normality with the presence of two outliers. The main result of this study is that the PNEU-ellipsoid when applied to test data is capable of detecting pneumonia in the sense that 100% of NL, 85% of LC and 65% of PTB will be rejected when using contrast texture measures. Using the combination of twelve features gives similar results. Membership of PNEU-ellipsoid obtained from the chest radiograph image can be used as a useful first stage detection of pneumonia.