Optimizing image quality using statistical multivariate optimization methodology using desirability functions

In order to optimize image quality, Figures of Merit (FOM) have been developed, including Signal-to-Noise ratio (SNR), Contrast-to-Noise ratio (CNR), and CNR2-to-Dose ratio (CNR2/PED). Some FOMs are designed to describe the performance of system components: Detective Quantum Efficiency (DQE) and Noise Equivalent Quanta (NEQ) are examples. A single FOM has the downside that optimization is inherently driven by the design of the FOM and cannot be changed. In this paper, we propose using a multi-parametric methodology for optimizing multiple input factors and multiple response measurements. This methodology has been developed in the statistical community as an offshoot of MANOVA (Multivariate ANalysis Of VAriance) analysis. In this paper, we acquired 120 images with various techniques and measured four individual image quality metrics. We then developed multivariate prediction formula for each metric and determined the global optimum operating point, using desirability functions. We demonstrate the power of this methodology over single FOM metrics.