Fig. 6:

Comparison of different metrics for registration of T2 and GradMag images. As seen on the left, the substantially different tissue and edge contrast between fixed and moving images make a difficult problem for registration. Our deep metric was derived from data that is only approximately registered by following a 3 step IMR. Plotting of each metric as a function of translation in x, y, z directions (t_x, t_y, t_z) is depicted. Based on these plots, mutual-information-based metrics have a noisier response function, compared to our derived deep metric, which is smoother. The capture range of our deep metric is also comparable to mutual-information-based metrics.