Machine learning-based pathomics signature could act as a novel prognostic marker for patients with clear cell renal cell carcinoma

Background Traditional histopathology performed by pathologists through naked eyes is insufficient for accurate survival prediction of clear cell renal cell carcinoma (ccRCC). Methods A total of 483 whole slide images (WSIs) data from three patient cohorts were retrospectively analyzed. We performed machine learning algorithm to identify optimal digital pathological features and constructed machine learning-based pathomics signature (MLPS) for ccRCC patients. Prognostic performance of the prognostic model was also verified in two independent validation cohorts. Results MLPS could significantly distinguish ccRCC patients with high survival risk, with hazard ratio of 15.05, 4.49 and 1.65 in three independent cohorts, respectively. Cox regression analysis revealed that the MLPS could act as an independent prognostic factor for ccRCC patients. Integration nomogram based on MLPS, tumour stage system and tumour grade system improved the current survival prediction accuracy for ccRCC patients, with area under curve value of 89.5%, 90.0%, 88.5% and 85.9% for 1-, 3-, 5- and 10-year disease-free survival prediction. Discussion The machine learning-based pathomics signature could act as a novel prognostic marker for patients with ccRCC. Nevertheless, prospective studies with multicentric patient cohorts are still needed for further verifications.

Automated summary

The machine learning-based pathomics signature can significantly distinguish ccRCC patients with high survival risk and may serve as a novel prognostic marker. The integration of MLPS, tumor stage system, and tumor grade system improves the accuracy of survival prediction for ccRCC patients.