Performance evaluation of image registration

From the medical point of view, the only meaningful measure of system performance is correct diagnosis. In ventilation and perfusion (V/Q) lung scan analysis, the goal is to identify the degree of match or mismatch, which is dependent on V/Q ratios. High registration performance is a critical factor for computation of the ratios. Thus, the method used to evaluate system performance is important. Here, the authors present an exclusive-or (XOR) function to evaluate registration error improvement and investigate several commonly used evaluation methods to compare registration results. Three measurements are compared with respect to accuracy, reliability, and computational cost. Both direct and indirect methods are studied to assess registration performance. The direct method requires direct user interaction such as manual identification of corresponding points or expert judgment. The indirect method is an automated approach with a minimum amount of user interaction. Examples include the correlation coefficient function and signal-to-noise ratio (SNR) approach