Quality Assessment in Magnetic Resonance Images

Abstract

Assessing quality of medical images is critical because the subsequent course of actions depend on it. Extensive use of clinical magnetic resonance (MR) imaging warrants a study in image indices used for MR images. The quality of MR images assumes particular significance in the determination of their reliability for diagnostics, response to therapies, synchronization across different imaging cycles, optimization of interventional imaging, and image restoration. In this paper, we review various techniques developed for the assessment of MR image quality. The reported quality indices can be broadly classified as subjective/objective, automatic/semi-automatic, region-of-interest/non-region-of-interest-based, full-reference/no-reference and HVS incorporated/non-HVS incorporated. The trade-of across the various indices lies in the computational complexity, assumptions, repeatability, and resemblance to human perception. Because images are eventually viewed by the human eye, it is found that it is important to incorporate aspects of human visual response, sensitivity, and characteristics in computing quality indices. Additionally, no-reference metrics are the most relevant due to the lack of availability of a golden standard against which images could be compared. Techniques that are objective and automatic are preferred for their repeatability and to eliminate avoidable errors due to factors like stress, which arise in human intervention.