Trade-offs in human observer performance, image quality metrics, and patient dose

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT Computed tomography (CT) is an excellent diagnostic tool, but it exposes patients to ionizing radiation. Consequently, an intensive, international effort has been made to reduce the radiation dose levels used for CT imaging. Our long-term objective is to develop and validate highly translatable methods that can quantitatively determine, for any specified diagnostic task, CT protocols that deliver the needed diagnostic accuracy at the lowest patient dose. These methods will be, by design, applicable to any scanner model or imaging practice. In our first competitive award period, we demonstrated that differences in scanners and scanning protocols (e.g. doses, reconstruction algorithms) can lead to substantial variations in diagnostic performance. More importantly, our multi-reader, multi-case observer studies demonstrated wide variations in performance among readers (radiologists) and across different cases. These variations were larger than the variations due to dose. Thus, a critical need exists to quantify and reduce these large variations in performance, but little work has been done on this topic. Only after addressing this critical need can the CT community achieve a consistent level of diagnostic performance over a wide range of scanners, cases, and readers and therefore safely adopt lower doses in abdominal imaging – one of the most common CT applications. Thus, we now have a second long-term objective, which is to reduce the variation in diagnostic performance that occurs due to case and reader variation, even when appropriate CT protocols are used, and especially at lower doses. The specific goals of this renewal application are to 1) validate that our methods for establishing lowest- dose protocols (for a targeted level of performance) are indeed applicable to any scanner make or model; 2) characterize case, lesion, and reader factors that lead to low diagnostic performance despite an otherwise acceptable scan protocol; and 3) develop adaptive assessment and learning strategies to improve readers' diagnostic skills across case and lesion type. We will accomplish these goals through three specific aims: 1. For multiple scanner models and protocols, demonstrate the success of our protocol optimization engine. 2. For abdominal CT, determine case, lesion, and reader predictors of radiologist diagnostic performance. 3. Develop adaptive learning and assessment techniques to address case and reader variability. The proposed work is significant because it will use objective and quantitative metrics, as well as leading- edge education and adaptive learning technology, to improve diagnostic performance and consistency in low- dose CT imaging. This work is innovative because, for the first time, the case, lesion and reader features leading to decreased diagnostic performance will be characterized and then mitigated with state-of-the-art adaptive assessment and training techniques. The results of this work will allow any imaging facility to optimize their dose levels without compromising the lifesaving diagnostic information obtained from CT.