Multiscale image analysis of prostatic carcinoma and benign hyperplasia using confocal laser scanning microscope.

摘要

Nuclear shape analysis has proved more objective than Gleason scores for prognosis assessment in clinically localized prostatic carcinoma (CaP). However, it relies upon manual nuclear contour digitization that makes the method tedious and slow. Currently available image analysis systems using light-absorption microscopy provide images with fuzzy and overlapping nuclear boundaries that make it difficult to automatically segment nuclei from background.;A semi-automated prostatic carcinoma grading system has been developed in this study. Optically sectioned, high resolution, and non-overlapping nuclear boundaries were digitally sampled by confocal laser scanning microscopy. Nuclear boundaries and centers of mass were automatically detected by a multiscale segmentation algorithm based upon multiscale Laplacian of Gaussian filtering and a course-to-fine recombination scheme. Artifacts were discriminated by criteria derived from mathematical morphology and in an interactive way. Shape, texture, and architecture features were extracted from semi-automatically segmented nuclear objects.;As a preliminary test of our system, we studied prostatectomy specimens from 8 patients with clinically localized prostatic carcinoma and 7 patients with benign prostatic hyperplasia (BPH). Nuclear shape was studied by a manual tracing and light microscope based (Dynacell) system and the semi-automated system. Both methods distinguished CaP from BPH by nuclear roundness factor, ellipticity by ferret diameter ratio, nuclear area, and perimeter. The semi-automated system provides higher values with greater range for nuclear shape than standard analysis. Nuclear texture analysis showed CaP differed from BPH in number, area, heterogeneity, margination, and condensation of chromatin clumps (nucleoli and highly condensed chromatin). Analysis of glandular nuclear architecture showed that mean and variance of internuclear distance distinguished CaP from BPH.;The ease of this semi-automated specimen analysis will allow evaluation of large CaP patient sets of known outcome to determine whether nuclear shape, chromatin texture, and architecture analysis can be extended from research to clinical usage.