Discrimination between benign and malignant prostate biopsies using three-dimensional chromatin texture analysis by confocal laser scanning microscopy.

Abstract

Objective: To evaluate the clinical usefulness of computing three-dimensional (3-D) nuclear texture features on prostate biopsy specimens to discriminate among benign, prostatic intraepithelial neoplasia (PIN), and malignant specimens.

Study design: Twelve prostate cancer biopsy specimens were selected, diagnosed as either benign (N = 4), PIN (N = 4), or malignant (N = 4). Sections 14 microm thick were stained. 3-D image stacks of selected benign and malignant areas were obtained by confocal laser scanning microscopy (CLSM) and analyzed off-line using in-house-developed software for 3-D semiautomated segmentation and calculation of texture features. The power of the 3-D texture features to discriminate among the pooled benign (N = 1,507), PIN (N = 673), and malignant nuclei (N = 1,251) was established by multivariate linear discriminant analysis.

Results: A total of 68.8% of the benign nuclei, 77.2% of the PIN nuclei, and 78.5% of the malignant nuclei could be classified correctly after cross validation.

Conclusion: Quantification of changes in the distribution of nuclear chromatin by means of 3-D texture feature computation on CLSM images allows discriminating most benign and malignant prostate nuclei, which could be useful in cases that are difficult to diagnose morphologically.