A statistical method for comparing chest radiograph images in MTB patients

Abstract

Digital images of chest radiograph taken at different time points may be compared to investigate the effect of treatment on mycobacterium tuberculosis (MTB) patients. One method of comparison is that of visually locating “snow-flakes� which should decrease in area or size with each subsequent image. This paper propose a more objective method; the comparison of image histograms whereby a leftward shift of the histogram indicates a positive effect of treatment. The comparison of two histograms is equivalent to either comparing the corresponding box-plots or the corresponding set of percentiles. However, before the comparison is made the images need to be registered and resized. The results of this study show that the proportion of percentiles (from histogram) can be used as an indicator of treatment effect (or patient’s progress). Further the correlations, RF2, is shown to be the best similarity measure to indicate the quality of image registration. Finally, this study also shows that a combination of registration and resizing can improve the pair-wise comparison.