Comparison of image quality measures in cerebral magnetic resonance angiography

Magnetic resonance angiography (MRA) is a less expensive and less invasive alternative to conventional x-ray angiography in the imaging of the cerebral vasculature. However, x-ray angiography (XRA) provides better detail of small cerebral vessels than that currently obtainable with MRA. At the University of Utah a large research effort is directed towards improving MRA's depiction of the small cerebral vasculature. An important element in this research is a measure of image quality. This dissertation addressed the development of image quality measures for cerebral MRA. In this dissertation image quality was defined in terms of observer ability to detect segments of cerebral arteries. This detection was quantified using a two alternative forced choice experiment. The experiment used x-ray angiography images as a truth standard for defining the size, location, and orientation of the cerebral vasculature. It examined whether the two alternative forced choice results could be predicted by intrinsic image measures (i.e., signal-difference-to-noise ratio, contrast, and signal-difference-to-background-variation ratio) or by a ranking experiment where an observer qualitatively ranked the vasculature depicted in regions-of-interest within the images. These evaluation measures were applied to four imaging effects: (1) decreasing the receiver bandwidth for MR acquisition, (2) using magnetization transfer to reduce background tissue signal, (3) using zero-filled interpolation to improve display of the vasculature information, and (4) using phase constrained reconstruction techniques to reconstruct the MR angiography data. The intrinsic image measures and ranking measures did not strongly predict which effects significantly improved image quality defined by the two alternative forced choice. Only in the analysis of the magnetization transfer and zero-filled interpolation effects were all measures in agreement. When examined locally at the regions-of-interest, all the measures were found to be weakly predictive of each other, suggesting that all measures used common information and that each test was also sensitive to different additional information. The two alternative forced choice and ranking results were in closest agreement with contrast measurements, suggesting that these data were not primarily noise limited. The two alternative forced choice and ranking results were both sensitive to misregistration between the x-ray angiography images and the magnetic resonance angiography images and to the presence of additional vessels within the regions-of-interest. To increase confidence in the two alternative forced choice measure misregistration and signal ambiguity problems need to be corrected.