On October 31st, UVA Medical Imaging Research hosted Subok Park, PhD, from the Food and Drug Administration (FDA), who presented a lecture on tasked-based optimization of medical imaging systems for signal detection with application to digital breast tomosynthesis.
Task-Based Statistical Approach
A task-based statistical approach has been increasingly used to optimize imaging systems for clinically relevant tasks including signal detection. One of the many advantages of this approach is that rigorous and comprehensive in-silico imaging trials can be conducted to optimize the system before patients are exposed to unnecessary radiation. The essential components of this approach are:
- Clinically relevant task of interest
- A model observer to perform the task, which incorporates system characteristics and data variability
- Appropriate figure of merit to accurately measure observer performance
- An ensemble of models to represent the realism of patient anatomy and abnormalities.
It is essential to characterize and model each of the components as accurately as possible and incorporate variabilities found in each stage of the imaging chain such that system optimization is performed rigorously for relevant tasks found in clinical settings. Digital Breast Tomosynthesis (DBT), also called 3D mammography, has become available in the US market only a few years ago. DBT is a limited-angle tomography which inevitably has many different optimal settings.
The task-based approach is well suited for judiciously optimizing trade-offs between device performance and patient safety/comfort for such a system. In this presentation, both the general task based approach and its application to the optimization and assessment of 3D mammography systems were discussed at length.
Comments