[Perceptual image dissimilarity--a novel metric for objective assessment of image quality in computed tomography with...

An iterative reconstruction (IR) technique in computed tomography (CT) is expected to play an important role in reducing the radiation dose while preserving both spatial resolution and contrast-to-noise ratio. However, images obtained by using the IR technique are known to have different visual appearances from those obtained by using the traditional filtered back-projection (FBP) reconstruction. This appearance is often figuratively described as "blocky," but it has not been objectively characterized further. In this paper, we propose a novel image quality metric, called "perceptual image dissimilarity" (PID), to characterize the visual dissimilarity between FBP and IR images. The PID was formulated as a grayscale transformation and subsequent structural similarity (SSIM)-based image quality measurement. The PID metric was validated using phantom images with three different modules. Sixty datasets, each consisting of an IR image and its corresponding noise-level-equivalent FBP image, were visually assigned "subjective dissimilarity scores" on a five level scale by six observers. The data sets were then quantitatively analyzed using both the PID and the traditional mean squared error (MSE) metrics. Our results show that the PID is highly consistent with the subjective dissimilarity score and thus delivers superior performance, whereas the MSE fails to quantify the observers' visual perception.