Liver 4D-MRI: An Image Mutual Information based Retrospective Self-sorting Method

Four-dimensional MRI (4D-MRI) is an emerging technique for soft-tissue motion management in radiotherapy treatment planning. The purpose of this study was to develop a novel image mutual information (MI) based retrospective 4DMRI sorting method that is fully automatic and free of external surrogates. Two novel components of the image-based 4D-MRI method were developed: automatic EOE/EOI phase sorting and inter-slice phase propagation. Image MI was first used to find EOE/EOI pair for each slice and subsequently used to form EOE/EOI chain across slices. After EOE and EOI phase determination, the MI values between intra-slice frames and EOE phase were employed as surrogates for phase sorting. In addition, MI-based inter-slice phase propagation was utilized to maximize the similarity between matching phases of neighboring slices so that the issue of image discontinuity can be mitigated. The first component was examined on a liver cancer patient and the second component on a 4D-XCAT digital human phantom infused with twelve real patient data. Our results showed that the fully automatic EOE/EOI phase sorting matched well with the manual sorting method. The inter-slice phase propagation method worked successfully on the XCAT digital phantom with less than 1% of voxels being mismatched. In conclusion, unlike some existing image-based 4D-MRI methods, the proposed MI-based 4D-MRI sorting method is fully automatic and potentially less sensitive to anatomy discontinuity caused by breathing irregularity. However, a future cohort study with a larger pool of human subjects is warranted to further assess the robustness of the proposed method.