Tissue architectural features for the grading of prostatic carcinoma

Area of Science:

• Digital pathology
• Computational oncology
• Biomedical image analysis

Background:

• Objective grading of prostatic carcinoma is crucial for treatment decisions.
• Subjective grading by expert panels can introduce variability.
• Development of automated systems can improve consistency and accuracy.

Purpose of the Study:

• To develop a computer-aided workstation for objective grading of prostatic carcinoma.
• To analyze tissue architectural (histometric) features for cancer differentiation.
• To evaluate automated image segmentation techniques for histometric analysis.

Main Methods:

• Analysis of histometric features from digitized tissue sections of well-, moderately-, and poorly differentiated prostate cancer.
• Comparison of interactive and expert system-guided image segmentation.
• Application of simplicial decomposition and run-length statistics for texture analysis.
• Evaluation of nuclear features such as nuclei per field and nuclei per gland.

Main Results:

• Expert system-guided segmentation achieved over 90% correct nuclear segmentation in 64% of fields.
• Number of nuclei per field showed separation between well-differentiated and other lesions.
• Number of nuclei per gland distinguished well-differentiated from moderately differentiated lesions.
• Combined run-length features and histometric data significantly differentiated all three cancer grades.

Conclusions:

• Histometric features, particularly field cellularity and nuclei per gland, are valuable for discriminating prostate cancer grades.
• Expert system-guided segmentation is effective for automated analysis in digital pathology.
• Insights were gained into challenges like field boundaries for computer-aided grading systems.