Towards abdominal MRI-based treatment planning using population-based Hounsfield units for bulk density assignment

This study investigates the dosimetric impact of using population-based Hounsfield units (HUs) and ICRU-based HUs as a function of the number of tissue segments for bulk density assignment toward MRI-based treatment planning in the abdomen. To avoid potential geometric differences between CT and MR images, CT images rather than MR images were chosen to simulate an MRI-only planning scenario. A retrospective study was performed utilizing 18 patients that had previously undergone stereotactic body radiation therapy for liver or pancreas cancer. HU values in the CT datasets were collected for various tissue types, and compared with the HUs derived from ICRU report 46. Doses were recalculated using the fluence obtained from clinical plans and with (1) homogeneous assignment, (2) ICRU-based HU assignment and (3) population-based HU assignment using three, four, five, nine or ten tissue segments. Dose-volume metrics for targets and organs-at-risk for all scenarios were compared with those obtained using the clinical CT. For the planning target volume (PTV) D99.9%, the mean differences from clinical CT plans were -2.1% ± 3.9%, -0.6% ± 0.3% and -0.1% ± 0.3% for homogeneity, ICRU-HUs and population-HUs using ten tissue segments, respectively. The population-HU method resulted in better dosimetric accuracy compared to the ICRU-HU method (p-value < 0.05). The dosimetric accuracy of homogeneity plans was comparable to that of both ICRU-HU and population-HU plans when targets were far from the lungs but deteriorated when targets were close to the lungs. As the number of tissue segments decreased, the dosimetric accuracy for PTV D99.9% reduced for the population-HU method, from -0.1% for ten tissue segments to -0.4% for three tissue segments, while no such dependence was observed for the ICRU-HU method. Hence, to generate a clinically acceptable plan when using MRI to synthesize CT in the abdomen for treatment planning, it might be sufficient for electron density assignment with either the population-HU or ICRU-HU method to only use three tissue segments.