Date Published: August 14, 2019
Publisher: Public Library of Science
Author(s): Susannah Maxwell, Richard Fox, Donald McRobbie, Max Bulsara, Jenny Doust, Peter O’Leary, John Slavotinek, John Stubbs, Rachael Moorin, Ganesh Jagetia.
Organ radiation dose from a CT scan, calculated by CT dosimetry software, can be combined with cancer risk data to estimate cancer incidence resulting from CT exposure. We aim to determine to what extent the use of improved anatomical representation of the adult human body “phantom” in CT dosimetry software impacts estimates of radiation dose and cancer incidence, to inform comparison of past and future research.
We collected 20 adult cases for each of three CT protocols (abdomen/pelvis, chest and head) from each of five public hospitals (random sample) (January-April inclusive 2010) and three private clinics (self-report). Organ equivalent and effective dose were calculated using both ImPACT (mathematical phantom) and NCICT (voxelised phantom) software. Bland-Altman plots demonstrate agreement and Passing-Bablok regression reports systematic, proportional or random differences between results. We modelled the estimated lifetime attributable risk of cancer from a single exposure for each protocol, using age-sex specific risk-coefficients from the Biologic Effects of Ionizing Radiation VII report.
For the majority of organs used in epidemiological studies of cancer incidence, the NCICT software (voxelised) provided higher dose estimates. Across the lifespan NCICT resulted in cancer estimates 2.9%-6.6% and 14.8%-16.3% higher in males and females (abdomen/pelvis) and 7.6%-19.7% and 12.9%-26.5% higher in males and females respectively (chest protocol). For the head protocol overall cancer estimates were lower for NCICT, but with greatest disparity, >30% at times.
When the results of previous studies estimating CT dose and cancer incidence are compared to more recent, or future, studies the dosimetry software must be considered. Any change in radiation dose or cancer risk may be attributable to the software and phantom used, rather than—or in addition to—changes in scanning practice. Studies using dosimetry software to estimate radiation dose should describe software comprehensively to facilitate comparison with past and future research.
Computed Tomography (CT) scanning provides an essential tool for protecting and improving health. The technology is widely used to diagnose disease, define its extent, assess response to therapy and aid in the planning and conduct of medical procedures and interventions. However, every CT scan delivers a small radiation dose to the body that is potentially carcinogenic. Concerns about the adverse impact of this radiation dose, and a world-wide trend towards increasing collective radiation dose[2, 3] have led to guidelines advising on the indications for CT scanning and radiation dose,[4, 5] as well as epidemiological research on the potential incidence and mortality of cancers resulting from exposure to CT scans within a population.[2, 6–12]
Scanning data were obtained for 160 cases for each of the abdomen/pelvis and head protocols, and 158 cases for the chest protocol. Cases with a tube voltage of 120kV accounted for all of the abdomen/pelvis cases, 155/158 of the chest cases and 124/160 head scan cases. Median effective dose and median organ equivalent dose as calculated by ImPACT and NCICT dosimetry software are shown in Table 1. The relationship between the effective dose as calculated by the ImPACT and NCICT software on a case by case basis are shown in Fig 2, showing consistently larger effective doses calculated by NCICT than ImPACT for the abdomen/pelvis and chest protocols, and lower effective doses for the head protocol. Fig 3 shows the relationship between the ImPACT and NCICT medians for each protocol by male and female. Along the horizontal line is the magnitude of the NCICT median (the reference result), while the vertical line shows the percentage difference between the two estimates relative to the NCICT median. The organs represented below the line of equivalence (i.e. zero difference) are those for which the NCICT median is lower than the ImPACT median. The male ImPACT and NCICT calculations of organ equivalent doses for the abdomen/pelvis protocol show a general decrease in the percentage difference in the estimated organ dose as the magnitude of the dose increases. The greatest percentage difference is shown for the prostate (90%) and the thyroid (84%) dose estimates. However, the thyroid organ receives very little equivalent dose comparatively to the other organs, and in terms of the difference in magnitude, the NCICT dose is only 0.22 mSv higher (Table 1). This is the same for the female thyroid estimates (Table 1). The female results also show large variation for breast (56%) with smallest variation for the stomach (11%) and liver (14%) (Fig 3). The chest protocol (Fig 3B) shows much higher mean variation in results, with the NCICT results that are consistently higher than the ImPACT results, with the exception of leukaemia (i.e. bone marrow) and lung dose estimates in males. The head protocol results (Fig 3C) also show higher NCICT than ImPACT results for most organs.
For the majority, but not all, of the organs used in epidemiological studies of cancer incidence, the NCICT software (voxelised phantom) provides higher estimates of organ dose among three different scanning protocols compared to the ImPACT software (mathematical phantom). The smallest percentage differences were generally seen among those organs that are included in their entirety in the scan region. These are stomach, liver, bladder, colon, uterus and ovaries for the abdomen/pelvis protocol and lung and breast for the chest protocol, with larger percentage differences seen for organs further away, or on the boundary of the scan region. The head protocol, for which most of the organs considered were outside the scan region, the percentage difference between the software results are highest. This is a general observation, and some deviation does occur. Within the chest protocol among females, the uterus and ovaries (outside of the scan region) had percentage differences equal to or smaller than the within-scan organs, breast and lung. These results showing greater percentage differences depending on the proximity of the organ to the scan region are consistent with previous studies comparing software using mathematical and voxelised phantoms.[22, 27]
Previous studies estimating cancer incidence using mathematical phantoms remain valid. However when the results of these studies are compared to more recent, or future, studies using voxelised and/or hybrid phantoms, the role of different software in the calculation of results must be considered. Definite comparisons should only be made where dosimetry and cancer estimates can be recalculated using the same software. Where this is not feasible, care must be taken not to overlook the potential role of the change in software on outcomes. It is clear that any change, or lack of change, in radiation dose or cancer risk may be attributable to the type of software and phantom used, rather than—or in addition to—changes in scanning practice. In line with the recommendations of the ICRP, new studies should aim to use software based on the more anatomically realistic voxelised phantoms. Studies using dosimetry software to estimate radiation dose should describe the software comprehensively to facilitate comparison with both past and future research.