Diagnosis of angiomyolipoma using computed tomography—region of interest ≤−10 HU or 4 adjacent pixels ≤−10 HU are recommended as the diagnostic thresholds

Abstract

To study and compare the diagnostic accuracy of region of interest (ROI) density measurement and pixel mapping [computed tomography (CT) density of individual pixels] for the diagnosis of renal angiomyolipoma (AML) using CT. A study group of histologically proven AMLs was compared with a control group of histologically proven renal cell cancers, normal renal parenchyma, and simple renal cysts. The mean tissue density (ROI circle) and a pixel density map were recorded. The diagnostic accuracy of various thresholds of ROI and pixel mapping values were compared using receiver operating characteristic curves. Twenty-two AMLs, 16 renal cell carcinomas (RCCs), 30 simple cysts, and 30 sites of renal parenchyma were evaluated. The mean (+/-1 SD) density of the AMLs was significantly lower [-15.2(20.8) units] than the three control groups [+36.0(8.1) units, +5.4(3.4) units and +22.2(46.5) units for RCC, renal cyst and parenchyma respectively; p < 0.001 (analysis of variance)]. The sensitivities and specificities of the ROI diagnostic thresholds of < or =0 units, < or =-10 units and < or =-20 units were 77 and 97%, 73 and 100% and 50 and 100%, respectively. Using pixel mapping [diagnostic thresholds of either a line of 4 pixels < or =-10 units or a square of 4 pixels < or =-10 units] the sensitivity improves to 86% with a specificity of 97%. Although a ROI threshold value of < or =-10 units has a very high specificity (100% in the present study) the sensitivity is modest at only 73%. Pixel mapping is more sensitive for recognizing small clusters of fat. In practice, both methods can be recommended for the analysis of suspected AMLs. ROI density measurement is convenient when analysing large areas of suspected fat and < or =-10 units should be used as the diagnostic threshold. When faced with small lucent areas or indeterminate values after ROI analysis, pixel mapping is recommended using a line of 4 pixels < or =-10 units or a square of 4 pixels < or =-10 units as the discriminating thresholds.