Design and evaluation of a breast specific collimator using response surface methodology and Monte Carlo simulations

Design optimization of collimators for various cameras and scanning procedures has been a challenge for many years. The experimental optimization studies require high cost scanning procedures or time consuming complex Monte Carlo simulations. This fact creates a need for efficient design of experiments. In this specific example, we utilized the response surface methodology to maximize the lesion detectability with respect to collimator parameters, namely the hole length, hole size and septal thickness for the simulated breast specific scintigraphy collimator to yield the highest contrast-to-noise ratio (CNR) for a subcentimetric lesion. CNR was chosen as the metric since it takes into consideration both the resolution and the sensitivity, the well-known trade-off in nuclear medicine imaging. Moreover, we performed a sensitivity analysis to external factors such as lesion depth and studied the working principles of the optimizer. As a result, we can state that the optimization works as a matched filter to match the lesion size. The matched filter is known as the optimal linear filter for maximizing the CNR when additive stochastic noise is present. For both lesion depths, a significant increase in CNR was observed with the optimized collimator parameters compared to commercial general purpose and high resolution collimators. As a conclusion, we could develop a better understanding of the collimator optimization process by exploring the solution space systematically by using a sequential search scheme to generate the response model fit.