论文部分内容阅读
Purpose:To effectively manage tumor motion in pancreas treatment using MR guided radiotherapy,automated segmentation needs to be performed.A low rank decomposition combining with Hessian matrix enhance methods(LRDE)was developed for this purpose.Methods and material:T2 weighted HASTE and T1 weighted VIBE images were acquired on 8 patients and 2 healthy volunteers for a total of 26 imaging volumes.3D imaging slice for each imaging series was decomposed into low rank component and sparse component.Hessian matrix filter was used to enhance searching for pancreas's geometrical structures which can be regarded as tubular.In order to obtain pancreas we need user-provided strokes that mark the object for auto-segmentation based on the location of pancreas.The segmentation results were refined using a morphology filter including corrosion and expansion operators.LDRE was compared to state of the art segmentation method graph cut(GC)using manual contours as the reference.Results:LDRE resulted in more accurate pancreas segmentation(DI=0.73)than GC(DI=0.71).DIs of LDRE based on HASTE and VIBE were 0.69 and 0.77,respectively.Conclusion:A low rank decomposition enhancement method was developed and applied for automated segmentation of healthy and pancreatic cancer patient MRI images.LDRE resulted in high pancreas segmentation accuracy that is potentially useful for MR guided pancreas radiotherapy.This method is an unsupervised segmentation that needn't training and learning process.VIBE images appear to be better suited for automated segmentation than HASTE and HASTE of healthy has great improvement compared with the existing segmentation method.Key words:Pancreas,MRI,Automated Segmentation,Low Rank Decomposition,Image Guided Radiotherapy.