ABSTRACT
Objective To investigate the feasibility of computer-aided pancreas segmentation based on three-dimensional (3D) Dixon MRI at 3.0-T.Methods Five volunteers with health pancreas underwent upper abdominal 3D Dixon MRI at 3.0T with the same scan parameters while holding breath.The automatic segmentation of pancreas is based on the Medical Imaging Interaction ToolKit (MITK) with homemade software.A experienced radiologist performed manual segmentation and computer aided segmentation of the pancreas on the opposed phase and water phase images twice.Similarity was analyzed and compared for the manual and automated segmentations for the two group images.Similarity was presented as Dice coefficients.Results Both of the oppose phase and water phase images of Dixon MRI showed that the signal intensity of pancreas was higher than that of the surrounding tissue,the similarity of water phase was obviously higher than that of oppose phase and the differences were statistically significant.Based on Dixon MRI water phase and oppose phase images,the mean dice coefficients for the manual segmentations were 0.81 ± 0.01 and 0.85 ±0.03,respectively,for both the oppose phase and water phase images (P=0.013).For the automated segmentations,the dice coefficients were 0.69 ± 0.08 and 0.75 ± 0.03 for water phase images and the oppose images,respectively,and the difference was not statistically significant (P =0.155).Conclusions 3.0T oppose phase and water phase of Dixon images could clearly show the pancreas,and automated pancreas segmentation based on water phase Dixon MRI obtained higher similarity,which was feasible.