Explicit shape descriptors: novel morphologic features for histopathology classification.

Area of Science:

• Digital pathology
• Medical image analysis
• Computational anatomy

Background:

• Object morphology in medical imaging is crucial for disease detection and grading.
• Prostate cancer grading relies on gland morphology, but subtle differences lead to inter-observer variability.
• Current boundary-based descriptors lack discriminability for subtle shape variations.

Purpose of the Study:

• To introduce novel explicit shape descriptors (ESDs) for distinguishing subtle morphological differences in prostate glands.
• To enhance the accuracy of prostate cancer grading, particularly for intermediate Gleason grades (3 and 4).

Main Methods:

• Representing object morphology using explicit shape models (e.g., medial axis).
• Aligning shape models via non-rigid registration with diffeomorphic constraints.
• Applying non-linear dimensionality reduction (e.g., Graph Embedding) to derive ESDs.

Main Results:

• ESDs effectively capture subtle shape differences between prostate glands of varying Gleason grades.
• A Support Vector Machine classifier using ESDs achieved 0.89 maximum accuracy and 0.78 AUC.
• The method correctly distinguished between benign, Gleason grade 3, and grade 4 prostate glands.

Conclusions:

• Explicit shape descriptors offer superior discriminability for subtle prostate gland morphology compared to traditional methods.
• ESDs hold promise for improving the objectivity and accuracy of prostate cancer grading.
• This computational approach can aid pathologists in more precise cancer assessment.