Comparison of manual versus low-level segmentation methods used for the validation of NRR algorithms in subtraction CT angiography