Intelligent Prediction of WHO/ISUP Grade in Clear-Cell Renal Cell Carcinoma Using CT-Based Fusion Delta Features: A Multicenter Study with Biological Interpretability Exploration.
Authors
Affiliations (8)
Affiliations (8)
- Department of Radiology, Yantai Yuhuangding Hospital, Qingdao University, Yantai, Shandong, China (L.W., Z.W., J.X., G.L., R.X., J.W., F.L., H.M., H.G.).
- Department of Radiology, Yantai Yuhuangding Hospital, Qingdao University, Yantai, Shandong, China (L.W., Z.W., J.X., G.L., R.X., J.W., F.L., H.M., H.G.); School of Medical Imaging, Shandong Second Medical University, Weifang, Shandong, China (R.X.).
- Department of Radiology, Yantai Yuhuangding Hospital, Qingdao University, Yantai, Shandong, China (L.W., Z.W., J.X., G.L., R.X., J.W., F.L., H.M., H.G.); School of Medical Imaging, Shandong Medical and Pharmaceutical University, Yantai, Shandong, China (J.W.).
- School of Computer Science and Technology, Shandong Jianzhu University, Jinan, Shandong, China (X.X.).
- School of Software, Shandong University, Jinan, Shandong, China (X.K.).
- Department of Urology, Yantai Yuhuangding Hospital, Qingdao University, Yantai, Shandong, China (J.W.). Electronic address: [email protected].
- Department of Radiology, Yantai Yuhuangding Hospital, Qingdao University, Yantai, Shandong, China (L.W., Z.W., J.X., G.L., R.X., J.W., F.L., H.M., H.G.). Electronic address: [email protected].
- Department of Radiology, Yantai Yuhuangding Hospital, Qingdao University, Yantai, Shandong, China (L.W., Z.W., J.X., G.L., R.X., J.W., F.L., H.M., H.G.). Electronic address: [email protected].
Abstract
Our study aimed to develop a delta deep learning radiomics model for predicting World Health Organization/International Society of Urological Pathology (WHO/ISUP) grade in clear cell renal cell carcinoma (ccRCC) and to find relevant biomarkers and explore the potential biological mechanisms. This multicenter cohort of 754 patients with ccRCC from five hospitals was split into training and internal validation cohorts in a 7:3 ratio, along with an external validation cohort (n=221) from three other hospitals. Radiomics and deep learning features were extracted from the tumor regions on multiphase contrast-enhanced computed tomography (CT). Delta features were calculated as the absolute differences between features from different imaging phases. The selected features were used to develop delta deep learning radiomics signatures using different classifiers. The diagnostic performance in each model was assessed. Furthermore, gene expression profiling was conducted to for exploratory analysis of biological correlates associated with the model's grade predictions. The delta model achieved AUCs of 0.736 (95% CI 0.671-0.801) and 0.701 (95% CI 0.635-0.768) in the internal and external validation cohorts, respectively, which were numerically higher than those of the original model based on traditional deep learning radiomics features and demographic baseline clinical model. Additionally, the delta model demonstrated good calibration and a higher net benefit. Bioinformatics and multiomics analysis revealed that high-grade risk-related differentially expressed genes (DEGs) were associated with inflammation-related pathways, providing preliminary exploratory biological insights into the model's predictions. The delta model demonstrated promising performance in WHO/ISUP grade. DEGs in predicted high-grade ccRCC were associated with inflammation-related pathways.