Angiomyolipoma with minimal fat and non-clear cell renal cell carcinoma: differentiation on MDCT using classification and regression tree analysis-based algorithm. (December 2014)

• Record Type: Journal Article
• Authors: Woo, Sungmin; Cho, Jeong Yeon; Kim, Seung Hyup; Kim, Sang Youn
• Is Part Of: Acta radiologica. Volume 55:Number 10(2014:Dec.)
• Page Start: 1258
• Page End: 1269
• Publication Date: 2014-12
• Subjects: Kidney -- CT -- neoplasms – primary -- angiomyolipoma with minimal fat -- non-clear cell renal cell carcinoma -- classification and regression tree analysis; Radiology, Medical -- Periodicals; Radiography, Medical -- Periodicals; Radiotherapy -- Periodicals; 616.0757
• DOI: 10.1177/0284185113513887
• Languages: English
• ISSNs: 0284-1851

Abstract

Background: Differentiation between angiomyolipoma with minimal fat (AMLmf) and non-clear cell renal cell carcinoma (nccRCC) may be difficult owing to lack of macroscopic fat in AMLmf. However, the differential points between AMLmf and nccRCC has not been well established in the literature. Purpose: To evaluate quantitative triphasic multidetector computed tomography (MDCT) features that differentiate between small AMLmf and nccRCC, and to integrate them to develop a simple and easy diagnostic algorithm. Material and Methods: This study was approved by the Institutional Review Board; informed consent was waived. Triphasic MDCT images of pathologically-proven AMLmfs ( n = 24) and nccRCCs ( n = 55) of 79 patients were retrospectively evaluated. Age, sex, size, long-to-short axis ratio (LSR), attenuation and enhancement degree in all phases, unenhanced tumor-kidney attenuation difference (UTKAD) in Hounsfield units (HU) were compared with Chi-square analysis, independent-samples t-test, and receiver-operating characteristic (ROC) curves. A criterion was formulated with classification and regression tree analysis (CART). Thereafter, CART-based algorithm was tested with additional interpretations from two radiologists. Intra- and inter-observer variability was analyzed with Bland-Altman analysis. Results: LSR was greater in AMLmf than nccRCC ( P < 0.001). AMLmf showed higher attenuation (all phases), CMP enhancement, and wash-out than nccRCC ( P ≤ 0.001). UTKAD was greater in AMLmf than nccRCC ( P < 0.001). ROC curve analysis yielded area under the curves of 0.936, 0.888, and 0.853 using UTKAD, unenhanced attenuation, and LSR. CART-based algorithm (UTKAD >7.5 HU, LSR > 1.23) predicted AMLmf with sensitivity, specificity, PPV, and NPV of 87.5%, 96.4%, 91.3%, and 94.6%. Mean intra- and inter-observer difference was –0.1/0.03 HU and −1.0/0.09 HU for UTKAD/LSR, respectively. These interpretations changed the final diagnosis in 1.3% (1/79) and 5.1% (4/79) patients for radiologists 1 and 2. Conclusion: Triphasic MDCT was useful for differentiating AMLmf and nccRCC. CART-based algorithm using UTKAD > 7.5 and LSR > 1.23 was simple and accurate in predicting AMLmf.