A nomogram model for challenging cases: differentiating fat-poor angiomyolipoma from clear cell renal cell carcinoma in uncertain or misdiagnosed tumors

Yue Xiao; Mei Yang; Zilong Jiang; Yihan Sheng; Haifeng Fan; Xupeng Ye; Ying Lei; Huchao Mao; Yan Zhang; Quanqian Zhang

doi:10.1007/s00261-025-05362-4

A nomogram model for challenging cases: differentiating fat-poor angiomyolipoma from clear cell renal cell carcinoma in uncertain or misdiagnosed tumors

Abdom Radiol (NY). 2026 Jan 12. doi: 10.1007/s00261-025-05362-4. Online ahead of print.

Authors

Yue Xiao¹, Mei Yang¹, Zilong Jiang¹, Yihan Sheng¹, Haifeng Fan¹, Xupeng Ye¹, Ying Lei¹, Huchao Mao¹, Yan Zhang¹, Quanqian Zhang²

Affiliations

¹ The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, China.
² The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, China. Zhangquanqian@wmu.edu.cn.

PMID: 41524985
DOI: 10.1007/s00261-025-05362-4

Abstract

Purpose: To develop and validate a nomogram model that can accurately differentiate fat-poor angiomyolipoma (fp-AML) from clear cell renal cell carcinoma (ccRCC) in clinically challenging scenarios where conventional imaging diagnosis is uncertain or erroneous.

Methods: This retrospective study included 404 patients with renal tumors. Unlike conventional cohort partitioning, we designed a clinically relevant test cohort specifically comprising two types of "difficult" cases: 74 patients with an indeterminate imaging diagnosis and 47 patients whose initial imaging diagnosis contradicted the pathological report. The training cohort (n = 283) consisted of cases with consistent imaging and pathological diagnoses. From preoperative computed tomography (CT) images, 1864 radiomic features were extracted. After a rigorous feature selection process using the Least Absolute Shrinkage, Selection Operator (LASSO) regression, and Max-Relevance and Min-Redundancy (mRMR), 25 robust features were retained. Seven machine learning algorithms were evaluated to build the prediction model.

Results: The ExtraTrees model demonstrated the best generalizability on the independent test cohort comprised of challenging cases, achieving an area under the curve (AUC) of 0.777. By integrating the radiomic model with clinical features, we constructed a nomogram. This combined model showed significantly improved performance, with an AUC of 0.815 (95% CI: 0.739-0.891) in the test cohort, outperforming the clinical model (AUC = 0.712) and the radiomics-only model (AUC = 0.777). Decision curve analysis confirmed the substantial clinical net benefit of the nomogram.

Conclusion: This model provides a reliable tool to assist radiologists and urologists in making precise preoperative diagnoses for the most challenging renal tumors, thereby potentially avoiding unnecessary surgeries.

Keywords: CT; Clear cell renal cell carcinoma; Fat-poor angiomyolipoma; Machine learning; Nomogram; Radiomics.

Abstract

Grants and funding