Abstract
PURPOSE: Grade-dependent decrease of lipid storage in clear cell renal cell carcinoma (ccRCC) leads to morphology changes in HE sections. This study investigated the role of cytoplasmic features in frozen sections of ccRCC on prognosis using the digital pathology approach. METHODS: We established an automatic pipeline that performed tumor region selection, stain vector normalization, nuclei segmentation, and feature extraction based on the pathologic data from Shanghai General Hospital and The Cancer Genome Atlas database. Extracted features were subjected to survival analysis. RESULTS: Kurtosis of the cytoplasm in the hematoxylin channel was correlated with progression-free survival (HR 0.10, 95% CI: 0.04-0.24, p=6.52 (∗) 10(-7)) and overall survival (HR 0.11, 95% CI: 0.05-0.31, p=1.72 (∗) 10(-5)) in ccRCC, which outperformed other texture features in this analysis. Multivariate Cox regression analysis revealed that low kurtosis of cytoplasm in the hematoxylin channel was an independent predictor for a shorter progression-free survival time (p=0.044) and overall survival time (p = 0.01). Kaplan-Meier survival analysis of progression-free survival and overall survival also showed a significantly worse prognosis in patients with low kurtosis of the cytoplasm in the hematoxylin channel (both p < 0.0001). Lower kurtosis of cytoplasm in the hematoxylin channel was associated with higher pathologic grade, less cholesterol ester, and more mitochondrial DNA content. CONCLUSION: Kurtosis of the cytoplasm in the hematoxylin channel predicts survival in clear cell renal cell carcinoma.