A Survey Over Image Quality Analysis Techniques for Brain MR Images

Abstract

AIM: We carry out a study on image quality methods, organize them in a logical approach, provide the mathematical framework, and finally discuss their performance. Objective image quality measures are frequently used in image processing. This review is not dedicated to subjective tests as they are very difficult and time consuming processes. MATERIALS AND METHODS: The quality studies are performed during the pre-processing step through the assessment of the de-noising efficacy, during the processing step as segmentation operation and as methods that evaluated its performance or in pattern-recognition. Extensive studies have taken hybrid metrics into account such as structural similarity index (SSIM), mean SSIM, feature similarity, the quality index based on local variance, and objective metrics such as signal-to-noise ratio, peak signal-to-noise ratio, mean square error, mean absolute error, contrast to noise ratio, root mean square error, Bhattacharyya coefficient or mutual information. RESULTS: These methods are compared in the context of brain MR images based on the reported performances. The most frequently used objective metric in the evaluation of the quality of processed MR image is SNR because it is slightly tissue-depended. From the hybrid metrics, the most used is SSIM. CONCLUSION: This paper summarized the objective and hybrids metrics that are Human Vision System - based characteristics. Also, it discusses on the notion of image quality assessment. The problems faced by various metrics are highlighted and the advantage of utilized a certain metric or a tandem of certain metrics are emphasized.