Case Study: Support for the diagnosis of renal cell carcinoma

Automatic segmentation and classification of histological subtypes to improve diagnostic accuracy and reduce evaluation time in digital pathology.

Context

Manual classification of renal cell carcinoma subtypes in histological images is a complex process due to tumor heterogeneity and morphological similarities between entities.

Differentiating between the main subtypes—chromophobe renal cell carcinoma (chRCC), papillary renal cell carcinoma (pRCC), and clear cell renal cell carcinoma (ccRCC)—can be particularly challenging in cases with overlapping characteristics.

This complexity can:

  • Increase the evaluation time per case
  • Generate interobserver variability
  • Hinder early clinical decision-making

In high-care settings, these limitations directly impact diagnostic efficiency.

Solution

We developed a deep learning-based artificial intelligence model capable of:

  • Automatically segmenting tumor regions in digitized histological images.
  • Automatically and multi-class classifying the main subtypes of renal carcinoma.
  • Identify differential morphological patterns with high sensitivity.
  • Integrate as a support tool for pathologists within the digital workflow.

The system was trained with images annotated by experts and optimized to ensure robustness against variations in staining, scanning, and image quality.