Computed tomography (CT) provides important medical benefits, but for patient safety it is essential that CT providers use the lowest dose of radiation consistent with achieving the needed diagnostic performance. New algorithmic approaches to image reconstruction will be critical to reducing the dose without compromising image quality;however, the development of novel approaches to image reconstruction is hampered because many image scientists do not have access to CT projection data from patient exams. We propose to develop data sets, metrics, and software tools that will help investigators create and compare new approaches to dose reduction and will guide clinical users in selecting optimized scanning parameters and reconstruction methods.
In Aim 1, we will make reference patient data sets available to researchers in a standardized format after removal of proprietary information. These data will include projection data, statistical noise maps, reconstructed images, and clinical reference information (validation of diagnosis and location, abstracted patient history) for common CT exams, as well as data simulating lower exposure levels. These data sets will greatly expand the pool of researchers that can develop and evaluate algorithms, and will permit comparison of competing approaches. The gold standard for measuring diagnostic performance, observer performance studies, is however very expensive and time consuming.
In Aim 2, we will develop highly automated, interactive, and freely available software tools that will facilitate rapid completion of observer performance studies in order to efficiently and meaningfully compare alternative scanning protocols and reconstruction methods. Still, because of the rapid pace of technical innovation, a substitute for efficient observer performance studies is essential to rapidly translate advances in dose reduction into patient care. Although task-based image quality metrics using model observers are attractive for this purpose, they have not been demonstrated to correlate with radiologist performance in clinical CT imaging.
In Aim 3, we will determine model observers that are substantially correlated with human observer performance in patient data for three common diagnostic tasks and for linear and non-linear image reconstruction techniques. Finally, quantitative methods are needed to assist clinical practices in choosing scanning protocol parameters that will achieve the required level of diagnostic performance using the lowest radiation dose.
In Aim 4, we will develop tools to efficiently and quantitatively optimize CT scanning protocols for specific diagnostic tasks. These tools will calculate quantitative measures of task-based image quality from easily performed phantom scans and will recommend scanning protocol parameters that will deliver the closest match to the desired level of diagnostic performance using the lowest radiation dose. This research is highly innovative and significant in that it will provide the CT community with novel data, methods, and software tools for objective evaluation and efficient optimization of scanning protocol parameters and emerging dose reduction approaches.

Public Health Relevance

While computed tomography (CT) provides important medical benefits, it is essential that CT exams use the lowest dose of radiation that allows accurate diagnosis. In this work, we will develop data sets, methods, and software tools that will help investigators achieve a significant reduction in CT dose without unknowingly compromising diagnostic accuracy.

Agency
National Institute of Health (NIH)
Type
Research Project--Cooperative Agreements (U01)
Project #
5U01EB017185-02
Application #
8719101
Study Section
Special Emphasis Panel (ZEB1)
Program Officer
Sastre, Antonio
Project Start
Project End
Budget Start
Budget End
Support Year
2
Fiscal Year
2014
Total Cost
Indirect Cost
Name
Mayo Clinic, Rochester
Department
Type
DUNS #
City
Rochester
State
MN
Country
United States
Zip Code
55905
Zhang, Yi; Leng, Shuai; Yu, Lifeng et al. (2014) Correlation between human and model observer performance for discrimination task in CT. Phys Med Biol 59:3389-404