Abstract
This Report proposes a framework, based on statistical decision theory, within which imaging system performance may be measured, optimized and compared. Imaging system assessment depends on the task for which the system is intended and can, therefore, be cast in terms of task performance, which can be measured in two stages.
First, task performance can be measured in terms of how well an ideal Bayesian decision maker would perform the task using only the acquired data, i.e., before it is presented as an image to a human observer. This can be done for certain well-defined imaging tasks, and leads to a description of the quality of the acquired data in terms of an ideal observer signal-to-noise ratio. The second stage involves measurement of the performance of the task by observers using display data. In this case, the assessment is made in terms of a receiver operating characteristic (ROC) curve and the quality of the displayed data can be specified in terms of a signal-to-noise ratio. The two stages of performance assessment are complementary and their related roles are explored in the Report. The analysis of the acquired data has the advantage of allowing one to investigate the effect on performance of altering various parameters of the imaging system. In certain circumstances, this approach may permit calculation of the best achievable signal-to-noise ratio, but, at present, it dose not necessarily predict the behavior of the human observer. This approach also requires laboratory measurements of system parameters such as modulation transfer function and noise power or Wiener spectrum and these might require special resources. The ROC curve approach provides a thorough assessment of the quality of the displayed data that takes the observer’s behavior into account, but it may be demanding of time and resources. Thus, it is proposed that the analysis based on the acquired data may provide not only an assessment of the potential performance of the imaging system, but also guidance as to those imaging conditions under which it would be most profitable to perform the ROC analysis. Finally, the proposed framework set out in the report links the purely objective measures of device performance to the subjective assessment of image quality and, further, offers the potential for moving to higher levels of efficacy analysis involving cost-benefits analysis of clinical data.