Evaluation of the limits of visual detection of image misregistration in a brain fluorine-18 fluorodeoxyglucose PET MRI study

Abstract

In routine clinical work, registration accuracy is assessed by visual inspection. However, the accuracy of visual assessment of registration has not been evaluated. This study establishes the limits of visual detection of misregistration in a registered brain fluorine-18 fluorodeoxyglucose positron emission tomography to magnetic resonance image volume. The “best” registered image volume was obtained by automatic registration using mutual information optimization. Translational movements by 1 mm, 2 mm, 3 mm and 4 mm, and rotational movements by 1°, 2°, 3° and 4° in the positive and negative directions in the x- (lateral), y- (anterior-posterior) and z- (axial) axes were introduced to this standard. These 48 images plus six “best” registered images were presented in random sequence to five observers for visual categorization of registration accuracy. No observer detected a definite misregistration in the “best” registered image. Evaluation for inter-observer variation using observer pairings showed a high percentage of agreement in assigned categories for both translational and rotational misregistrations. Assessment of the limits of detection of misregistration showed that a 2-mm translational misregistration was detectable by all observers in the x- and y-axes and 3-mm translational misregistration in the z-axis. With rotational misregistrations, rotation around the z-axis was detectable by all at 2° rotation whereas rotation around the y-axis was detected at 3–4°. Rotation around the x-axis was not symmetric with a positive rotation being identified at 2° whereas negative rotation was detected by all only at 4°. Therefore, visual analysis appears to be a sensitive and practical means to assess image misregistration accuracy. The awareness of the limits of visual detection of misregistration will lead to increase care when evaluating registration quality in both research and clinical settings.