J Comput Assist Tomogr 2008;32:757-763
Vinika V. Chaudhari, MD, Steven S. Raman, MD, Ng L. Vuong, MD, Peter Zimmerman, MD, James Farrell, MD, Howard Reber, MD, James Sayre, PhD, and David S.K. Lu, MD
Objective: To determine the frequency of typical features of pancreatic cystic lesions on high-resolution computed tomography and the combination of features that best influences discrimination.
Methods: Ten computed tomography features of 100 proven pancreatic cystic lesions were retrospectively tabulated by 2 blinded imagers. After final diagnosis was revealed, each lesion was categorized as typical or atypical. Stepwise multivariable logistic regression was used to determine which of 10 imaging and 4 clinical features significantly distinguished between benign and malignant lesions.
Results: There were 38 benign cysts and 62 cystic tumors. Serous lesions presented with greater than 6 cysts (83%) and cysts of less than 2 cm (44%). Mucinous lesions presented with cysts of 2 cm or greater (82%) and less than 6 cysts (64%). Pseudocysts, serous, and mucinous lesions presented typically in 77%, 67%, and 64% of cases, respectively. The significant variables in classifying malignant lesions are pancreatitis history, cyst size, symptoms, and calcification pattern (area under the curve, 0.837).
Conclusions: Four specific imaging and clinical features in combination best predict a malignant lesion.