An Information-Theoretic Spatial Resolution Criterion for Qualitative Images

Innovation in image reconstruction techniques is measured by improvements in reliable image quality criteria, such as contrast and resolution. Traditional image quality criteria are designed for linear techniques with quantitative values. However, many newer ultrasound algorithms are nonlinear and produce images with nontraditional statistics and distributions; their underlying information content is considered more important than the raw image values themselves. Here, we demonstrate that such qualitative images are fundamentally incompatible with ultrasound spatial resolution metrics designed for quantitative images, such as the width and separability of point targets and the speckle autocorrelation length. We further propose a new spatial resolution metric based on the self-mutual information of translating image patches, called the autoinformation length. The autoinformation length captures both linear and nonlinear dependencies between image patches and is invariant under dynamic range transformations, making it well-suited for evaluating nonlinear imaging methods.