Detection of degradation of magnetic resonance (MR) images: comparison of an automated MR image-quality analysis system with trained human observers.

RATIONALE AND
OBJECTIVES: The perceived need for magnetic resonance (MR) imaging quality control (QC) is occasionally minimized on the assumption that significant errors will be detected by the users. To evaluate the validity of this assumption, we compared the sensitivity of a test object and automated image analysis system for MR imaging QC with the sensitivity of trained human observers by evaluating images that were intentionally degraded.
METHODS: Parameters for imaging the test object and normal human volunteers were set to values that decreased the signal-to-noise ratio (SNR), caused distortion, and increased the slice thickness and separation.
RESULTS: The human observers were able to detect a 6-13% reduction in the SNR and distortions of more than 15% in human images. They were unable to identify 40% increases in the slice thickness. Automated analysis of test object images was able to detect all image degradations at the minimum levels applied.
CONCLUSION: The poor sensitivity of the human observers indicated that degradation, especially spatial measurements, could be significantly in error before being detected through visual analysis of clinical images. These errors would be detected by automated analysis of the test object used. Further investigation is needed to better define the accuracy with which quantitative image-quality analysis predicts the effects of degraded image quality on the ability of human observers to detect subtle abnormalities in clinical images.