Renal papillary carcinoma classification into subtypes may be reproduced by nuclear morphometry

Area of Science:

• Uropathology
• Computational Pathology
• Renal Cell Carcinoma Research

Background:

• Papillary renal cell carcinoma (PapRCC) is a significant subtype of kidney cancer.
• Accurate subtyping of PapRCC is crucial for prognosis and treatment.
• Objective analysis of nuclear morphology could aid in PapRCC classification.

Purpose of the Study:

• To investigate the relationships between nuclear features and PapRCC subtypes.
• To determine if image analysis can differentiate between PapRCC subtypes.
• To assess the potential of computational pathology in classifying PapRCC.

Main Methods:

• Analysis of nuclear geometric and texture features from 53 PapRCC cases (types 1, 2, and intermediate).
• Segmentation of over 100 nuclei per case from DAPI-stained slide images.
• Application of analysis of variance, expectation-maximization clustering, and neural network classification.

Main Results:

• Significant differences in nuclear parameters were observed between individual cases and PapRCC types.
• Unsupervised clustering identified nuclear categories but did not fully reproduce PapRCC classes.
• A neural network achieved >0.6 sensitivity and >0.75 specificity in nucleus classification, with varying accuracy across PapRCC subtypes (Type 1: 74-91%, Type 2: 58-80%, Intermediate: 53-70%).

Conclusions:

• Papillary renal cell carcinoma subtypes possess sufficiently distinct nuclear features.
• Image analysis, particularly using machine learning, shows promise in differentiating PapRCC subtypes.
• Computational methods can aid in the objective classification of PapRCC, supporting diagnostic accuracy.