Shape and Boundary Similarity Features for Accurate HCC Image Recognition

Area of Science:

• Pathology
• Medical Imaging
• Computational Biology

Background:

• Nucleus morphology is crucial for diagnosing cancer, visually distinguishing normal from abnormal cells.
• Current methods may not fully capture the subtle morphological differences relevant for hepatocellular carcinoma (HCC) diagnosis.

Purpose of the Study:

• To propose and evaluate novel nucleus shape and boundary similarity features for improved hepatocellular carcinoma (HCC) nucleus recognition.
• To enhance the accuracy of pathological diagnosis for HCC through advanced morphological feature analysis.

Main Methods:

• Nucleus patches were segmented using the center-proliferation segmentation (CPS) method.
• Nucleus shape similarity was measured using Dice, Jaccard, precision, and recall coefficients against a pathologist-curated library.
• Boundary similarity was assessed using triangles formed by nucleus boundary feature points, followed by random forest (RF) classification.

Main Results:

• The proposed novel morphological features demonstrated superior performance in classifying HCC nuclei.
• Cross-validation facilitated the selection of an optimal feature set, confirming the efficacy of the new features.
• Experimental comparisons showed these features are more beneficial than traditional characteristics for HCC nucleus recognition.

Conclusions:

• Novel nucleus shape and boundary similarity features offer significant advantages for HCC nucleus recognition in pathological diagnosis.
• The proposed method provides a more accurate and robust approach to identifying cancerous nuclei compared to existing techniques.