An Image Quality Assessment System for Evaluating MR Reconstruction Pipeline Using Single Image Acquisition Method

Abstract

Image quality assessment becomes critical for measuring the performance of Magnetic Resonance (MR) image reconstruction algorithms. The raw data acquired from MR scanner is processed through a series of components in the MR reconstruction pipeline to generate an image. The MR reconstruction pipeline includes the algorithms for reconstruction, quality enhancement, artifact reduction, and so on. The implementation of the pipeline requires evaluation of large number of filters and reconstruction algorithms, which demands a simple and effective mechanism to assess the quality of output at each stage. The main goal of the proposed work is to provide research groups and MR algorithm developers with protocols and tools necessary to conveniently assess the quality of MR images and to be able to compare the reconstruction methods among each other. This paper focuses on the implementation of an image quality assessment mechanism using the single image method by specifying Region of Interest (ROI) from object and background as suggested by National Electrical Manufacturers Association (NEMA). The single image method does not produce consistent results as the SNR values change drastically based on the background noise. It has been observed that, by careful selection of noise region, the drastic changes in values of SNR can be avoided with single acquisition techniques. This mechanism is supported by a visual aiding tool to carefully select signal and noise regions and avoid the errors introduced in blind selection of ROI. Using the single acquisition method and visual aiding tool, SNR comparison of images from two different reconstruction methods can be easily evaluated.