Patient-individualized dosimetry via semi-automated CT segmentation and voxel-based Monte Carlo transport methods

Objectives Monte Carlo methods using heterogeneous body models (phantoms) are in routine use for performing dose calculations for standardized individuals (e.g. the RADAR ICRP-89 phantom series, using the GEANT4 Monte Carlo transport code). These image-based models are represented in voxel format, with homogeneous activity distributions assumed within source organs. This technology has now been extended to allow input of individual patient anatomic and functional data, to perform 3D dose calculations with heterogeneous activity distributions in organs and tumors.

Methods We have established a method that uses a network of anatomical landmarks to quickly parse 3D CT data and estimate the locations and boundaries of organs present. When important organs are identified and segmented, they are assigned the usual code numbers used in our established GEANT4 transport codes to obtain 3D maps of absorbed dose. To date we have used a single SPECT or PET scan to determine activity distributions, with multiple planar images to establish the kinetics of individual regions, but we intend to ultimately work with multiple SPECT/PET scan data.

Results Using this approach, the segmentation of seven organs and detection of 19 body landmarks can be obtained in under a minute with high accuracy and has been validated against manual segmentation on 80 CT full or partial body scans; errors are typically a few mm. Manual adjustment and segmentation of small objects complements the semi-automated processes. An entire 3D dose work-up is provided by the Monte Carlo code in about 12 hours, with errors mostly around 5%, except for very small organs or tumors, for which more run time can be allotted if needed to improve accuracy.

Conclusions These robust methods provide a tool that can provide accurate, patient-specific 3D dosimetry in times that may be applied in routine clinical practice for radiopharmaceutical therapy