A new CT protocol to scan the head -- as well as the abdomen and thorax -- that reduces patient radiation dose by 24% to 36% and maintains diagnostic efficacy is being used in Malta, following the publication of a study this month.
Balancing image quality with radiation dose is always a struggle. Image quality should be able to deliver enough information to the radiologist, but at the same time follow the ALARA (as low as reasonably achievable) principle.
Use VGC and ordinal regression analysis, advises Dr. Francis Zarb."Hence, there is a need to have a keen understanding of image quality evaluation tools and methods of data analysis to identify the required level of image quality required for diagnosis in the development of optimized scan protocols using the lowest possible radiation dose," wrote Dr. Francis Zarb and colleagues from the department of radiography at the University of Malta Faculty of Health Sciences in Msida, and colleagues (Insights into Imaging, June 2015, Vol. 6:3, pp. 393-401).
Observer performance tests on images obtained using CT scanning protocols should be carried out testing visualization of anatomical structures or known pathologies, they wrote. Established methods for analysis include image criteria (IC) studies, visual grading analysis (VGA), and receiver-operating characteristic (ROC) analysis.
The researchers focused on VGA, which involves grading of the visibility of anatomical structures on the images. With relative VGA, the visibility of anatomical structures is compared and graded against the visibility of the same structures within a reference image. The observers grade the visibility of the structure with an arbitrary ordinal scale. A grade of 0 implies a visibility equal to the structure within the reference image, while negative or positive values imply inferior or superior visibility.
With absolute VGA, the visibilities of anatomical structures within the images are graded against each other.
"The scales are ordinary and are usually given a description facilitating interpretation and improving the agreement between observers," the study authors wrote. "A VGA score calculated from the results of such analysis allows statistical analysis of the differences."
VGC treats the scale steps as ordinal with no assumptions on the distribution of the data, they continued. The resemblance between VGC and ROC analysis leads to the possibility of using the well-established ROC evaluation methods in analyzing VGC data. The variation in the visual grading of the reviewers of two imaging techniques can be used to describe the variation between the two techniques in the same way as in an ROC study.
VGC can be considered as a repeated image criteria scoring where reviewers change their threshold to fulfill each criterion in a similar way to the scale steps in an ROC study. So the reviewers state their confidence about the fulfillment of a criterion obtaining an ordinal scale. The different ratings don't necessarily correspond to the same numerical intervals on the decision scale nor do all reviewers use the ratings with the same meaning, because the ordinal scale is just used to test the probability distribution for each imaging technique.
Zarb and colleagues evaluated image quality using VGC during optimization of CT head exams. The researchers analyzed image quality scores from visual grading assessment using VGC and ordinal regression analysis. Ordinal logistic regression analysis is a statistical technique able to process data on an ordinal scale that handles situations involving several factors that could potentially influence the outcome.
The researchers included 66 patient images obtained using current and optimized imaging protocols from two CT suites: a 16-slice scanner (GE Healthcare, BrightSpeed Elite) at the national Maltese center for trauma and a 64-slice scanner (Philips Healthcare, Brilliance) in a private center. Six local resident radiologists performed VGA followed by VGC and ordinal regression analysis.
They found VGC alone indicated that optimized protocols had similar image quality as current protocols. Ordinal logistic regression analysis provided an in-depth evaluation, criterion by criterion allowing the selective implementation of the protocols. The local radiology review panel supported the implementation of optimized protocols for brain CT exams (including trauma) in one center. Radiation dose reductions ranged from 24% to 36%. In the second center, a 29% reduction in radiation dose was achieved for follow-up cases.
Things changed in Zarb's department as a result of the research, he noted in an email to AuntMinnieEurope.com. "Based on the results of our research, we are now using the optimized head protocols with a reduction in patient dose whilst maintaining diagnostic efficacy," he wrote. "The same methodology was also used in the optimization of general CT abdomen and thorax examinations."
Furthermore, the same methodology was also adopted in comparing oral contrast preparation protocols for oncology patients before undergoing CT abdomen and pelvis for staging purposes.
"I think the most important aspect for the success of such research is the cooperation between all stakeholders, who should be involved from the very beginning of the study," Zarb wrote. "So liaison with radiologists is crucial, especially to get them to review the image datasets and to get them to accept changes in protocols."
Why visual grading?
The advantages of using visual grading studies in the evaluation of clinical images are that they can be carried out with clinically available images and there is no need for a gold standard during the evaluation.
"However, the use of appropriate data analysis methods should be emphasized," the study authors wrote.
The terminology is somewhat confusing, because some authors use VGA to denote standard statistical tests with assumptions that may not be appropriate, they added.
"Analyzing visual grading analysis methods with parametrical statistical tests such as t-tests and analysis of variance (ANOVA) incorrectly assumes the grading data are an interval variable; VGC and ordinal regression analysis correctly treat visual grading data as ordinal and categorical," they wrote.
An added advantage of the logistic regression model is it can simultaneously consider multiple factors influencing the quality of the image, which means more complete, detailed information can be obtained and a specific clinical decision can be made.
"Based on the results of both VGC and ordinal regression analysis performed in this study, the optimized protocols were implemented for all patient presentations (inclusive of trauma) for general brain on the GE BrightSpeed scanner as the negative odds did not affect any of the two most important criteria [visually sharp reproduction of the border between white and gray matter and also visually sharp reproduction of the cerebrospinal fluid space over the brain]," they wrote.
However, the optimized protocol was limited to follow-up cases on the Philips Brilliance, because visually sharp reproduction of the border between white- and gray-matter findings were affected by the optimization process, and this was considered important for initial diagnosis by the radiology experts.
Study limitations
A limitation of the study is that the image quality evaluation was based primarily on morphologically normal anatomical structures; radiologists were not asked to evaluate or comment on any pathology present in the image datasets, the researchers wrote.
"So the question still remains as to whether this low radiation dose is applicable for specific brain pathologies, which frequently present as low-contrast differences in comparison to normal brain tissue," they added. "The inclusion of pathologies together with an ROC analysis to investigate the applicability of the optimized protocols in the diagnosis of subtle to obvious pathologies is recommended."
Zarb and colleagues also suggested different weighting levels for anatomical criteria to be incorporated in future image quality research.
"Currently, research incorporating both VGC and ordinal regression analysis in review of clinical images and anatomical criteria is limited and, therefore, further research involving these statistical test tools is recommended," they concluded.
To read the full text of the study, visit the Insights into Imaging website.
