Manual and automated systems in the analysis of images from prostate tissue microarray cores.

Objective: To compare manual and automated image analysis systems in morphologic analysis of nuclei from benign prostate, high-grade prostatic intraepithelial neoplasia (HGPIN) and prostate cancer (CaP). Morphologic features derived using automated image analysis systems may be more objective and reproducible than manual systems, which require humans to segment nuclei from histologic images.

Study design: Images of hematoxylin-eosin-stained sections of prostate tissue microarray were analyzed independently using the automated and manual systems. Mean optical density (MOD), nuclear area (NA), and nuclear roundness factor (NRF) were the morphologic features studied. The ability to differentiate between tissue types using morphologic features derived from an automated and a manual system was compared.

Results: Nuclei from 17 benign prostate hyperplasia (BPH), 4 HGPIN, and 8 aggressive CaP were analyzed. The manual system distinguished better between BPH and HGPIN (p < 0.0001), whereas the automated system distinguished better between BPH and CaP (p = 0.01) in multivariate models. The manual system distinguished better BPH and HGPIN using NA (p < 0.0001) and MOD (p < 0.0001), whereas the automated system distinguished better BPH and CaP using MOD (p < 0.0001) and NRF (p = 0.004).

Conclusion: The minimal human effort required for automated image analysis makes it superior to the manual system.