SPARSITY-DRIVEN IDEAL OBSERVERS FOR GUIDING IMAGING HARDWARE OPTIMIZATION

点击翻译按钮获取中文摘要

 DESCRIPTION (provided by applicant): The broad objective of this R21 application is to develop and investigate a novel method for optimizing hardware of modern computed imaging systems with respect to signal detection tasks. Specifically, we will establish an efficient method for computing a sparsity-driven Bayesian ideal observer (IO) test statistic that exploits object information that is relevant to modern sparse reconstruction methods. Significance: Modern reconstruction methods, referred to as sparse reconstruction methods, exploit the fact that objects of interest can often be described by sparse representations and have proven to be highly effective at reconstructing images from under-sampled measurement data. Conventional wisdom dictates that imaging hardware should be optimized by use of an IO that exploits full statistical knowledge of the class of objects to- be-imaged, without consideration of the reconstruction method to-be-employed. However, accurate and tractable models of the complete object statistics are often difficult to determine in practice. Moreover, in computed imaging approaches that employ compressive sensing concepts, imaging hardware and image reconstruction are innately coupled technologies. Accordingly, we propose to investigate a practical approach in which the hardware is optimized by use of the same low-level statistical information about the object that enables sparse reconstruction. This will facilitate reductions in data-acquisition times and/or radiation doses for a wide range of modern medical imaging systems. Challenges: There remain several impediments to computing IO performance to guide hardware optimization in practice. Perhaps most fundamental is the need to know the full probability density function of the object. Unrealistic assumptions, such as Gaussian-distributed object backgrounds, are generally required for analytical computation of IO performance. Markov chain Monte Carlo (MCMC) techniques are available for computing IO performance, but are not routinely employed due to their extreme computational burdens. Solutions: We will formulate sparsity-driven IOs (SD-IOs) to guide hardware optimization that assume knowledge of low-level statistical properties of the object that are related to sparsity. The SD-IO will explit the same statistical information regarding the object that is utilized by highly effective sparse image reconstruction methods. To efficiently compute SD-IO performance, we will estimate the posterior distribution by use of computational tools developed recently for variational Bayesian inference with sparse linear models. Subsequently, the SD-IO test statistic will be computed semi-analytically. Aims: The specific aims of this project are as follows. Aim 1: To develop and validate a method for computing SD-IO signal detection performance Aim 2: To investigate the use of the SD-IO for guiding optimization of data-acquisition parameters