Entropy-based texture analysis of chromatin structure in advanced prostate cancer

Abstract

A new texture operator, gray-level entropy matrix (GLEM), was developed, and nine new textural features were extracted from this matrix. These textural features were applied to light microscopy images of nuclei taken from monolayers of advanced prostate cancer cells representing two different prognostic groups: hormone-sensitive (good prognosis) and hormone-resistant (poor prognosis) tumors. A comparison between the classification results obtained from GLEM features and those obtained from standard textural estimators is also discussed. Single features that gave correct classification rates better than 65% were included in a discriminant analysis in order to find the optimal set of features to discriminate between the two prognostic groups in the training data set. The best combination of features includes three GLEM features together with ENTROPY extracted from gray-level cooccurrence matrix, and this combination gave a correct classification rate of 95% using the leaving-one-out technique. The influences of image sharpness and number of cells were also investigated. The features based on entropy or degree of scatter of minute structures can be used to discriminate between hormone-sensitive and hormone-resistant prostate carcinomas.

Ähnliche Arbeiten