Histopathological prostate tissue glands segmentation for automated diagnosis

In this work, we propose a methodology for segmenting glands automatically in digitized images of histopathological prostate tissue for grade classification. Gleason grading describes the abnormality of cancer cells and their degree of aggressiveness by using numerical scale from grade 1 that represents benign tissues through grade 5 for tissues characterized as advanced stage cancer. The special characteristics of glands in prostate tissue for each grade play a significant role in discriminating Gleason grades. Therefore, lumen objects and tissue glands were segmented as the major regions of interest for tissue grading. Lumen objects were segmented by an empirical thresholding technique. Since we are mainly concerned with the inner regions of the glands consisting of the lumen, cytoplasm and the inner boundary of the cell nuclei, a k-means clustering approach was employed to the a* color channel of the L*a*b* color model for each of the tissue images. This was followed by statistical and morphological features extraction for the segmented lumen objects and glands. Finally, a naive Bayes classifier was used to classify tissue images to the correct grade. The efficiency of the automated segmentation method was evaluated, and classification results achieved accuracy, sensitivity, and specificity of 91.66%, 96.66%, and 95.00%, respectively. These results indicate that our automated methodology could serve as an adjunct to histopathologists and would have a positive impact when integrated with conventional histopathological diagnosis procedures.