Christian G. Graff1, Kyle J. Myers2
1Division of Imaging &
Applied Mathematics, U. S. Food & Drug Administration , Silver Spring,
MD, United States; 2Division of Imaging & Applied Mathematics,
U. S. Food & Drug Administration, Silver Spring, MD, United States
We develop a metric for image quality assessment based on task performance using the Ideal (Bayesian) Observer SNR. The Ideal Observer is an upper bound for human performance and measures the ability to perform a task (in this case lesion detection) using statistical decision theory. Ideal Observer performance can be assessed from raw data or image reconstructions to measure the effectiveness of both the data acquisition and image reconstruction. This has advantages over other metrics such as SNR and CNR which are only weakly correlated with a readers ability to perform clinical tasks.