DETERMINATION OF PATIENT SPECIFIC S-VALUES CORRELATED WITH NONINVASIVE PATIENT ATTRIBUTES

Abstract

Introduction: Nuclear medicine procedures are the second largest source of medical ionizing radiation, behind only computed tomography. The current population has been exposed to medical imaging more than previous generations, therefore knowledge of absorbed dose (ionizing) is not only an important component of patient care, but may have epidemiological implications as well. Unlike other modalities, in nuclear medicine a clinician can often decrease the patient’s administered activity at the expense of extending the acquisition time, administration therefore is not a binary decision . It should be noted radionuclide dosimetry can be accurately calculated if sufficient data is available. However, in most cases crucial information such as uptake, biokinetics and anatomical details cannot or will not be known a priori. Furthermore, the dosimetric process is often arduous, requiring intensive resources with respect to personnel and technology. Convenient models based on MIRD formalism have been developed to overcome the aforementioned challenges. Unfortunately, the models are not patient specific which causes inaccuracies, one such shortcoming centers around spatial variances. In the MIRD schema S values denote the mean absorbed dose to a target volume (organ) per unit activity in a source region (organ). The S values are based on phantom models for specific radionuclides, in the MIRD schema they are separable from the kinetic aspects of nuclear medicine. To improve patient specific nuclear medicine dosimetry S values will be determined for patients directly. The analysis will be conducted for radionuclides of interest and select organs. The range of organ S values will be examined to determine the need of personalization and its possible benefit. The results will be benchmarked against modern phantom models (dosimetric software), with the goal of personalizing phantom results through the utilization of basic patient attributes.

Methods: To evaluate the variances between the S-values of patients and phantoms, one needs diagnostic images. The gold standard modality of anatomical imaging is computed tomography. Fifty-six patient computed tomography exams were made available by the National Institutes of Health (NIH), along with information regarding the patient’s sex, age, height and weight. The patient images were de-identified prior, the clinical images contained the structures of the Chest, Abdomen and Pelvis of persons. The diagnostic images (CT) were segmented by TotalSegmentator which is a deep learning model, resulting in various structures. These structures were used to in part to create a source distribution and a lattice model for Monte Carlo implementation. The source energies and yields regarding emission was obtained from ICRP 107. Radiation transport is one of the areas that has greatly benefited from Monte Carlo methods. The Monte Carlo simulations conducted in this work utilize MCNP 6.2 which is a general-purpose, continuous-energy, generalized-geometry, time-dependent, Monte Carlo radiation transport code. The S values were calculated for a number of target organs (spleen, kidneys, pancreas, stomach, liver and lungs) resulting from activity in the patient’s kidneys for (Tc-99m, F-18, Lu-177 and I-131). The results were compared and benchmarked versus NCINM a modern dosimetry program based in part on phantom models. Results: The variation of S values for adult target organs were found to significant. The variation between the minimum and maximum S values for patients ranged from a two-fold to a five-fold factor and therefore supports the need for personalization. Benchmarking the S values with NCINM revealed in most cases the phantoms were contained within the patient’s distribution if not centered, which adds to the simulation’s credibility. The patient specific values were normalized to their appropriate NCINMphantom models. Significant correlation was found between the patient’s BMI and their normalized S values irrespective of radionuclide. The correlation is believed to be due to internal shielding. Based on a linear fit the NCINM model was extended or personalized. This extension greatly reduced the variation between actual S values and the NCINM dosimetry model. The interquartile range of the distribution of most organ S values were decreased by 50 percent.