Utility of whole-lesion ADC histogram metrics for assessing the malignant potential of pancreatic intraductal papillary mucinous neoplasms (IPMNs)

Purpose: To evaluate whole-lesion ADC histogram metrics for assessing the malignant potential of pancreatic intraductal papillary mucinous neoplasms (IPMNs), including in comparison with conventional MRI features. Methods: Eighteen branch-duct IPMNs underwent MRI with DWI prior to resection (n = 16) or FNA (n = 2). A blinded radiologist placed 3D volumes-of-interest on the entire IPMN on the ADC map, from which whole-lesion histogram metrics were generated. The reader also assessed IPMN size, mural nodularity, and adjacent main-duct dilation. Benign (low-to-intermediate grade dysplasia; n = 10) and malignant (high-grade dysplasia or invasive adenocarcinoma; n = 8) IPMNs were compared. Results: Whole-lesion ADC histogram metrics demonstrating significant differences between benign and malignant IPMNs were: entropy (5.1 +/- 0.2 vs. 5.4 +/- 0.2; p = 0.01, AUC = 86%); mean of the bottom 10th percentile (2.2 +/- 0.4 vs. 1.6 +/- 0.7; p = 0.03; AUC = 81%); and mean of the 10-25th percentile (2.8 +/- 0.4 vs. 2.3 +/- 0.6; p = 0.04; AUC = 79%). The overall mean ADC, skewness, and kurtosis were not significantly different between groups (p >= 0.06; AUC = 50-78%). For entropy (highest performing histogram metric), an optimal threshold of > 5.3 achieved a sensitivity of 100%, a specificity of 70%, and an accuracy of 83% for predicting malignancy. No significant difference (p = 0.18-0.64) was observed between benign and malignant IPMNs for cyst size >= 3 cm, adjacent main-duct dilatation, or mural nodule. At multivariable analysis of entropy in combination with all other ADC histogram and conventional MRI features, entropy was the only significant independent predictor of malignancy (p = 0.004). Conclusion: Although requiring larger studies, ADC entropy obtained from 3D whole-lesion histogram analysis may serve as a biomarker for identifying the malignant potential of IPMNs, independent of conventional MRI features.