Accuracy of computed tomography attenuation values in the characterization of pleural fluid: an ROC study.

ABSTRACT

RATIONALE AND OBJECTIVES: To assess the accuracy of computed tomography (CT) in characterizing pleural fluid based on attenuation values. MATERIALS AND METHODS: Protocol was approved by the local institutional review board and informed consent was waived. We retrospectively analyzed 145 pleural effusions of 145 patients (mean/standard deviation age 60.7/15.9 years; 69 females) who underwent CT of the thorax and diagnostic thoracentesis within 7 days of each other. Effusions were classified as transudates or exudates using laboratory markers based on Light's criteria. The mean Hounsfield units (HU) of an effusion was determined by a region of interest on the three slices with the greatest anteroposterior diameter. A receiver operating characteristic curve was constructed to determine threshold values for classification on the basis of mean HU and to examine overall accuracy, using the area under the curve (A(z)). RESULTS: Of the 101 exudates and 44 transudates, the mean attenuation of exudates (17.1 HU/standard deviation 4.4) was significantly higher than transudates (12.5 HU/6.3), (P < .001). There was a modest but significant positive relationship between mean HU and laboratory markers, with the strongest relationship with pleural/serum protein (r = 0.57, P < .001) and total pleural protein (r = 0.56, P < .001). The overall accuracy of attenuation values for identifying exudates was moderate, Az = 0.775, standard error = 0.039, with the largest limitation being the overlap with transudates in the 10-20 HU range, which constituted 66% (90/145) of the total effusions measured. CONCLUSION: Although the mean attenuation of exudates is significantly higher than transudates, the clinical use of CT numbers to characterize pleural fluid is not recommended, as their accuracy is only moderate. Moreover, there is a notable overlap in attenuation values between transudates and exudates for a majority of effusions.