• C L Sistrom, C T Reiheld, S B Gay, and K K Wallace.
• Department of Radiology, University of Virginia Health Sciences Center, Charlottesville 22908, USA.
• AJR Am J Roentgenol. 1996 Feb 1; 166 (2): 317-21.

ObjectiveThe purpose of this study was to determine the efficacy of real-time sonography for the detection and estimation of the volume of pneumothorax using receiver operating characteristic (ROC) analysis in 27 patients evaluated by five radiologists using previously described sonographic findings.Materials And MethodsBilateral chest sonograms were obtained in 27 patients after we performed 26 needle biopsies of the lung and two transcostal biopsies of lesions at the dome of the liver. Thirteen unilateral pneumothoraces were found on radiographs done at the time of the sonograms. The sonograms were recorded on videotape and later viewed by five blinded readers who scored each hemithorax independently as to the likelihood and size of pneumothorax. From these interpretations we were able to calculate ROC curves and standard accuracy statistics for each observer and for pooled data. The results were correlated with the findings on chest radiographs, which were used to verify the diagnoses.ResultsThe area under the ROC curves ranged from 0.63 to 0.79 in detecting any pneumothorax. The area under the ROC curve derived by pooling readers and hemithoraces using the jackknife method was 0.73. The average sensitivity of the five observers fro pneumothorax was 73%, the specificity was 68%, and the negative and positive predictive values were 89% and 40%, respectively. No significant correlation was found between readers' estimates of pneumothorax size made on the basis of sonographic findings and actual size as ascertained from chest radiographs.ConclusionReal time sonography is useful to localize known pneumothorax but cannot be used to exclude the diagnosis. Moreover, sonography has a significant false-positive rate and is of no use in estimating the volume of a pneumothorax.

20 keywords...

hide…