Toward a non-invasive screening tool for differentiation of pancreatic lesions based on intra-voxel incoherent motion derived parameters.

Early recognition of and differential diagnosis between pancreatic cancer and chronic pancreatitis is an important step in successful therapy. Parameters of the IVIM (intra-voxel incoherent motion) theory can be used to differentiate between those lesions. The objective of this work is to evaluate the effects of rigid image registration on IVIM derived parameters for differentiation of pancreatic lesions such as pancreatic cancer and solid mass forming pancreatitis. The effects of linear image registration methods on reproducibility and accuracy of IVIM derived parameters were quantified on MR images of ten volunteers. For this purpose, they were evaluated statistically by comparison of registered and unregistered parameter data. Further, the perfusion fraction f was used to differentiate pancreatic lesions on eleven previously diagnosed patient data sets. Its diagnostic power with and without rigid registration was evaluated using receiver operating curves (ROC) analysis. The pancreas was segmented manually on MR data sets of healthy volunteers as well as the patients showing solid pancreatic lesions. Diffusion weighted imaging was performed in 10 blocks of breath-hold phases. Linear registration of the weighted image stack leads to a 3.7% decrease in variability of the IVIM derived parameter f due to an improved anatomical overlap of 5%. Consequently, after registration the area under the curve in the ROC-analysis for the differentiation approach increased by 2.7%. In conclusion, rigid registration improves the differentiation process based on f-values.