Development of Population-Based Pediatric Computational Phantoms for Radiotherapy Applications.

Abstract

Age-specific pediatric computational phantoms are used in radiotherapy (RT) for quality assurance and for reconstruction of historical RT doses (within others). Phantoms are typically developed from healthy patients and may not effectively represent those with cancer due to pathology and/or treatment effects. This study evaluated a set of population-based pediatric computational phantoms developed in-house in terms of anatomical plausibility. Planning CTs and contours from historical craniospinal irradiation (CSI) patients (n = 74, median age 7y, range: 1-17y) were used to generate and evaluate a set of in-house age-specific population-based RT phantoms (RT-P). The RT-P were generated by combining a sub-set of clinical CTs and contours through groupwise deformable image registration, generating average models of CSI sub-populations (n = 74, median age 7y, range: 3-14y). Models were then compared against clinical data and two libraries of phantoms representing healthy populations: the International Commission on Radiological Protection (ICRP) pediatric reference computational phantoms (n = 8, median age 8y, range: 1-15y) and a variety of default 4D extended cardiac torso (XCAT) phantoms (n = 75, median age 9y, range: 1-18y). Variation between organ volumes for the different datasets was assessed through a linear fit of organ volume with age, reporting the slope (∑) of each fit [y-1]. Average difference between the volume datapoints and the linear fit for clinical data (Δ) [%] were also reported. This allowed for comparisons of the RT-P to clinical and reference data in terms of organ volumes across developmental stages. The table shows 9 of the 19 investigated organs. The ∑ reported for RT-P models were of similar magnitude as the clinical data and other phantoms, effectively modelling changes with age. The greatest and least ∑ were reported from lungs and thyroid respectively, in agreement with expected relative sizes between organs. Larger values for Δ were likely due to differences in organ filling and segmentation strategy between datasets, limitations of RT-P methodology, and/or anatomical differences between healthy and cancer populations. The RT-P models show promise in representing the RT cohort that may benefit from specialized anatomical phantoms. Further work is needed to address the limitations of the current methodology and its applicability to other RT cohorts.