ABSTRACT
In the present work, the computer aided classification (CAC) system has been proposed for renal ultrasound (US) images into normal, MRD, and Cyst image classes. The work has been carried out on 35 B-mode renal US images consisting of 11 normal, eight medical renal disease (MRD), and 16 cyst images. The regions of interest (ROIs) have been extracted from the parenchyma region of the kidney in case of normal and MRD classes and from the region inside the lesion for cyst image class. The classification has been carried out using (a) statistical texture description using different Haralick's texture descriptor vectors that includes mean, range, ratio, addition, and a combination of mean and range, (b) texture description in signal processing domain using four different Laws' mask texture descriptor vectors derived from Laws' mask of different resolution i.e., Laws' mask of length three, five, seven and nine, (c) texture description in transform domain using different Gabor texture descriptor vectors evaluated at various scales and orientations. The potential of these texture descriptor vectors for differential diagnosis between different renal ultrasound images have been measured using support vector machine (SVM) classifier. The result of the study indicates that among the five statistical domains Haralick's texture descriptor vectors, the vector consisting of a combination of mean and range features yields the highest overall classification accuracy (OCA) of 85.9% at inter-pixel distance ‘ d ′ = 1 https://s3-euw1-ap-pe-df-pch-content-public-p.s3.eu-west-1.amazonaws.com/9781315154152/905dd614-092a-44c9-95b3-cae749202c91/content/inline-math12_1.tif"/> with individual class accuracy (ICA) values of 86.6%, 83.0% and 91.3% for normal, MRD and cyst classes respectively.
