Purpose: The purposes of this study are (1) to utilize multivariable logistic regression in order to evaluate which image feature combination is most predictive in the diagnosis of cholecystitis for computed tomography (CT) and ultrasound (US) in adult ED patients and (2) to use these results to compare the accuracy of CT and US.
Methods: For RUQ pain patients undergoing US and CT at the same visit, multiple image features were evaluated independently by 2 radiologists blinded to additional data. Inter-reader variability was measured with the Kappa statistic. Sonographic Murphy's sign (SMS) information was obtained from original reports. Multivariable logistic regression was utilized to develop optimal predictive models for each modality. For US, models with/without SMS were compared to establish its relative value.
Results: 446 patients met inclusion criteria. For CT, the combination of cholelithiasis, short-axis gallbladder diameter > 3 cm, pericholecystic fluid or inflammation, and mural thickening > 3 mm provided the optimal model for both readers. For US, the optimal model included cholelithiasis, short-axis diameter > 3 cm, mural heterogeneity/striation, and sludge/debris for both readers. Kappa = 0.79-0.96 for included image features. For both readers, CT and US models had equivalent diagnostic performances; the SMS did not contribute significantly to US models.
Conclusion: For a diagnosis of cholecystitis in the ED, (1) the optimal image feature combination for CT is cholelithiasis, short-axis diameter > 3 cm, pericholecystic fluid or inflammation, mural thickening > 3 mm; and cholelithiasis, short-axis diameter > 3 cm, mural heterogeneity/striation, sludge/debris for US; (2) CT and US have equivalent diagnostic performance; (3) inter-reader reliability is substantial to excellent for utilized image features; (4) the SMS does not affect US model accuracy.
Keywords: Cholecystitis; Computed tomography; Sonographic Murphy’s sign; Ultrasound.
© 2021. The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.