Iterative peak combination: a robust technique for identifying relevant features in medical image histograms

Abstract

Histogram-based methods can be used to analyse and transform medical images. Histogram specification is one such method which has been widely used to transform the histograms of cone beam CT (CBCT) images to match those of corresponding CT images. However, when the derived transformation is applied to the CBCT image pixels, significant artefacts can be produced. We propose the iterative peak combination algorithm, a novel and robust method for automatically identifying relevant features in medical image histograms. The procedure is conceptually simple and can be applied equally well to both CT and CBCT image histograms. We also demonstrate how iterative peak combination can be used to transform CBCT images in such as way as to improve the Hounsfield Unit (HU) calibration of CBCT image pixel values, without introducing additional artefacts. We analyse 36 pelvis CBCT images and show that the average difference in fat tissue pixel values between CT images and CBCT images processed using the iterative peak combination algorithm is 23.7 HU. Compared to 136.7 HU in unprocessed CBCT images and 50.9 in CBCT images processed using histogram specification.