Comparison of Histogram-Based Gaussian Analysis With and Without Noise Correction for the Characterization of Indeterminate Adrenal Nodules.

The purpose of this study is to determine whether gaussian-based histogram analysis without and with noise correction can characterize indeterminate adrenal nodules (those with attenuation greater than 10 HU on unenhanced CT) as lipid-poor adenomas. This retrospective study evaluated adrenal nodules larger than 1 cm on unenhanced CT using gaussian analysis without and with noise correction on intralesional ROIs. Two independent readers who were blinded to the final diagnoses evaluated the nodules. The final diagnosis for each nodule was determined on the basis of pathologic findings or accepted imaging criteria. Interreader agreement was assessed using the intraclass correlation coefficient. Algorithm performance was summarized using sensitivity, specificity, and the AUC. Ninety-four adrenal nodules in 85 patients were analyzed; 36 of these were metastases (34 of which were pathologically confirmed), and 58 were presumed adenomas. Interreader agreement was excellent for nodule size, mean attenuation, SD of attenuation, and the gaussian index. Noise-corrected gaussian analysis had significantly higher specificity (81.9% vs 55.6%; < 0.001) and lower sensitivity (36.2% vs 56.9%; < 0.001) for identifying adenomas than did the uncorrected gaussian analysis. The AUC of corrected gaussian analysis was 0.72, which is significantly greater than that of uncorrected gaussian analysis (0.51; ≤ 0.001) and similar to that of mean attenuation (0.77). Noise correction is necessary when using a gaussian analysis characterization of indeterminate adrenal nodules on modern unenhanced CT examinations. This method may be able to discriminate between adenomas and nonadenomas.