Radiology:Volume 260: Number 1 -July 2011
Matthew S. Davenport, MD Amy M. Neville, MD James H. Ellis, MD Richard H. Cohan, MD Humaira S. Chaudhry, MD Richard A. Leder, MD
Purpose: To retrospectively determine the optimal Hounsfield unit threshold and region of interest (ROI) size required to accurately diagnose renal angiomyolipoma (AML) and differentiate it from renal cell carcinoma (RCC).
Materials and Methods: This retrospective study was institutional review board approved and HIPAA compliant, and the requirement for written informed patient consent was waived. The radiologic reports on 4502 dual-phase abdominal computed tomography (CT) examinations (nonenhanced and nephrographic phases, 5-mm collimation, 120-140 kVp, variable milliampere-second settings) performed in 2872 patients from June 2002 through October 2007 were reviewed. Solid-component masses reported as suspicious for RCC or AML were correlated with histologic and/or follow-up imaging findings. ROIs of three different sizes-tiny (8-13 mm2), small (19-24 mm2), and medium (30-35 mm2)-were drawn in the lowest-attenuation focus on images obtained during both phases. The test characteristics (sensitivity, specificity, positive predictive value, negative predictive value, false-positive rate, false-negative rate) of multiple attenuation thresholds at each combination of ROI size and contrast enhancement phase were calculated, and receiver operating characteristic (ROC) curves were derived. Areas under the ROC curve were calculated.
Results: There were 217 RCCs and 65 AMLs. With an attenuation threshold of -10 HU or lower at nonenhanced CT, RCC would be misdiagnosed as AML in 11 (5.1%) cases, one (0.5%) case, and one (0.5%) case with use of the tiny, small, and medium ROIs, respectively. With use of the tiny, small, and medium ROIs, misdiagnosis rates would be 2.3%, 0.5%, and 0.5%, respectively, at a threshold of -15 HU or lower and 1.8%, 0%, and 0%, respectively, at a threshold of -20 HU or lower. Areas under the ROC curve for the nonenhanced phase images (range, 0.874-0.889) were superior to those for the nephrographic phase images (range, 0.790-0.826).
Conclusion: Nonenhanced CT images were superior to nephrographic phase CT images for the diagnosis of AML. An attenuation threshold of -10 HU or lower with an ROI of at least 19-24 mm2 is optimal for the diagnosis of AML. This threshold is not accurate with use of smaller (8-13-mm2) ROIs.