Prostate cancer is a common malignancy among maturing men and the second leading cause of cancer death in USA.
Histopathological grading of prostate cancer is based on tissue structural abnormalities. Gleason grading system is the
gold standard and is based on the organization features of prostatic glands. Although Gleason score has contributed on
cancer prognosis and on treatment planning, its accuracy is about 58%, with this percentage to be lower in GG2, GG3
and GG5 grading. On the other hand it is strongly affected by "inter- and intra observer variations", making the whole
process very subjective. Therefore, there is need for the development of grading tools based on imaging and computer
vision techniques for a more accurate prostate cancer prognosis.
The aim of this paper is the development of a novel method for objective grading of biopsy specimen in order to support
histopathological prognosis of the tumor. This new method is based on texture analysis techniques, and particularly on
Gray Level Co-occurrence Matrix (GLCM) that estimates image properties related to second order statistics.
Histopathological images of prostate cancer, from Gleason grade2 to Gleason grade 5, were acquired and subjected to
image texture analysis. Thirteen texture characteristics were calculated from this matrix as they were proposed by
Haralick. Using stepwise variable selection, a subset of four characteristics were selected and used for the description
and classification of each image field. The selected characteristics profile was used for grading the specimen with the
multiparameter statistical method of multiple logistic discrimination analysis. The subset of these characteristics
provided 87% correct grading of the specimens. The addition of any of the remaining characteristics did not improve
significantly the diagnostic ability of the method. This study demonstrated that texture analysis techniques could provide
valuable grading decision support to the pathologists, concerning prostate cancer prognosis.