NSF/FDA SIR: Modeling Observer Performance in CT Dose Reduction Assessments

点击翻译按钮获取中文摘要

PI: Abbey, Craig K.Proposal: 1445737Title: Modeling Observer Performance in CT Dose Reduction AssessmentsSignificanceDose reduction in CT imaging has become an important priority in medical imaging that has emerged with the increased use of this imaging modality. The estimated annual rate of CT scanning in large integrated healthcare systems has increased roughly three-fold in the last 15 years, with worrisome estimates of induced cancers as high as 2%. This has motivated intensive investigations into dose reduction, including reduced-dose imaging protocols that use computer-intensive reconstruction techniques to recover information that may otherwise be lost in the in additional noise engendered by these protocols. The FDA has been asked by sponsors to approve claims concerning reduced-dose protocols recently, and expects to see more such applications in the near future. However, demonstrating effective dose reduction is challenging. By definition, such techniques seek to retain diagnostic quality with little or no measureable effect on diagnostic performance. Clinical reader studies using receiver operating characteristic (ROC) methodology are the accepted standard for evaluating effects on diagnostic performance. However, these studies are very expensive and time consuming, and identifying small effects, or non-inferiority, requires prohibitively large sets of readers and cases. As a result, the field is increasingly turning to model-observers as an alternative to standard observer performance studies. To be effective at predicting human observer performance, model observers require extensive validation with human data in psychophysical studies. The goal of this project is performing a validation in the context of x-ray CT imaging at low doses. The aims of the project are to collect a large set of psychophysical data relevant to CT dose reduction and analyze this data using classification image methods translated from vision science to better understand how human observers perform tasks in these noisy images.Technical DescriptionHow human observers perform tasks on the basis of noisy and possibly distorted images is still an open question. The investigation proposed will provide observer performance data and an approach to modelling human observers that are well beyond what has been done to date. Using the classification image methodology with both detection and localization tasks, they will have the opportunity to model how human observers utilize information in noisy CT images, and how the use of image information changes as dose is reduced and/or dose reduction methods (smoothing, etc.) are applied. The results will therefore be of interest to the broader vision science community as well as the fields of medical image perception and CT image reconstruction. The results of our studies will allow better models of observers to be used for assessing dose reduction. By providing a much more tractable model-observer study, which can be conducted in a laboratory setting without the need for a clinical readers and cases, investigators pursuing new CT imaging techniques will be more willing to evaluate and optimize dose in their studies. This is expected to result in a more widespread use of model based approaches to fine tune image dose procedures. It is also expected that the results will have impact in vision science with more insightful models of how human observers perform visual tasks in the presence of noise.