Predicting Grade and Patient Survival in Renal Cancer Using Machine Learning Analysis of Nucleolar Prominence

Background: Patients with clear cell renal cell carcinoma (ccRCC) often undergo organ resection, with treatment strategies based on recurrence risk. Current metastatic potential assessments rely on the WHO/ISUP grading system, which is subject to interobserver variability. Methods: We developed an artificial intelligence (AI) model to classify cells according to contemporary grading rules and evaluated the prognostic significance of tumor cell profiles, particularly focusing on cells with prominent nucleoli. Results: The model accurately distinguished low (G1/G2) and high (G3/G4) grades, achieving an area under the ROC curve of 0.79. Survival analysis identified four tissue patterns defined by total cell density and the proportion of cells with prominent nucleoli. The relative abundance of such cells had greater prognostic value than their mere presence, correlating with survival times ranging from 2.2 to over 6 years. Additionally, we confirmed that dystrophic changes and focal necrosis are linked to shorter survival. Conclusion: These findings suggest that incorporating refined criteria into the WHO/ISUP system could enhance its prognostic accuracy in future revisions. © 2025 Elsevier B.V., All rights reserved.