Fractal Behavior Of Gleason And Srigley Grading Systems

Abstract

Introduction/ Background

Prostate cancer remains one of the major malignancies of modern society. The need of grading this malignancy is still in dispute. Two major grading systems have emerged and are world-wide adapted: Gleason grading system [1] and Srigley grading system [2]. Both systems use optical subjective descriptions of different architec- tural growth patterns of prostate adenocarcinoma. The fractal dimension (FD) is used in the medical field as an objective feature for describing a given image rather than showing a precise value for a known fractal. The FD can be an objective measurement for different patterns description.

 

Aims

The aim of our study is to assess the fractal behavior of images labeled according to Gleason and Srigley grading systems both in terms of in-class and inter-class variation.

 

Methods

299 Gömöri stained microscopic digital images of prostate adenocarcinoma were labeled independently according to Gleason and Srigley patterns. Each image was firstly transformed to grayscale then a maximum cropped square of the image was resized to a standard 256x256 pixel image. For the resulted images the fractal dimension was approximated with two different algorithms: a standard box-counting algorithm (applied to the binary image obtained with Roberts’s method for edge detection) and a novel algorithm that is applied to the grayscale version of the  image  consisting  in the ratio between image’s volume and area (R-VA) at different scales [3]. In-class variation was assessed as the average standard deviation (SD).Lower SDmeans better discrimination. For the inter-class variation assessment each class was compared with all other classes using a two-tail, Student’s t-test. The resulted value was defined as the ratio between the statistically different means and the total number of comparisons. The maximum possible value for Gleason grading system was 28, be- cause there were no images labeled as Gleason pattern 1, while for the Srigley grading system the maximum possible value was 6.

 

Results

In-class variation was 0.045 using the box-counting algorithm and 0.048 using the R-VA algorithm for Gleason grading system and 0.161 using the box-counting algorithm and 0.178 using the R-VA algorithm for Srigley grading system. Inter-class variation was, for Gleason grading system 13/28 using the box-counting algorithm and 20/28 using the R-VA algorithm while for the Srigley grading system was 3/6 using the box-counting algorithm and 5/6 using the R-VA algorithm respectively. Srigley grading system seems to perform better than Gleason’s on inter-class variation, but has lower performance on in-class variation. Nevertheless, we must note that there is a large difference between the two systems regarding the number of classes. The FD computed with the R-VA algorithm has better discrimination results than the one computed with the box-counting algorithm in both grading systems, thus proving once again the R-VA’s performance [3].