Using human and model performance to compare MRI reconstructions

M. Dylan Tisdall

doi:10.1109/TMI.2006.881374

Using human and model performance to compare MRI reconstructions

2006-11-01

23

PRE

AI

DOI：10.1109/TMI.2006.881374

Original

Original request for help

Save

Abstract

En 中文

Magnetic resonance imaging (MRI) reconstruction techniques are often validated with signal-to-noise ratio (SNR), contrast-to-noise ratio, and mean-to-standard-deviation ratio measured on example images. We present human and model observers as a novel approach to evaluating reconstructions for low-SNR magnetic resonance (MR) images. We measured human and channelized Hotelling observers in a two-alternative forced-choice signal-known-exactly detection task on synthetic MR images. We compared three reconstructions: magnitude, wavelet-based denoising, and phase-corrected real. Human observers performed approximately equally using all three reconstructions. The model observer showed very close agreement with the humans over the range of images. These results contradict previous predictions in the literature based on SNR. Thus, we propose that human observer studies are important for validating MRI reconstructions. The model's performance indicates that it may provide an alternative to human studies.

Keywords:

denoising

magnetic resonance imaging (MRI)

observers

signal detection

signal processing

Using human and model performance to compare MRI reconstructions

Abstract

AI Summary

Journal

Organization