Objective: To develop procedures for the segmentation of cribriform prostatic glands.
Study design: A knowledge-guided procedure following a model-based reasoning process was developed in the context of a set of interacting expert systems for machine vision in histometry.
Results: With 78 entities in the knowledge file, fully automated, correct segmentation of approximately 70-80% of cribriform glands was attained--i.e., outlining of histologic components agreed with visual assessment. Measurement of gland size, shape, lumen area, number of lumina per gland, epithelial layer thickness, degree of cribriformity and determination of completeness of lining of a gland by a basal cell layer could be taken from the correctly segmented images.
Conclusion: The automated procedure allows a histometric characterization of premalignant and malignant prostatic lesions. Extension of system capabilities to the utilization of spectral information is expected to allow an increase in the correct segmentation rate.