Difference entropy as a similarity metric for medical images nonrigid registration

In the present work a similarity metric based on difference image entropy has been investigated for medical images nonrigid registration using block-matching algorithm. The used image histogram size has shown to affect registration accuracy, allowing registration improvements for difference entropy metric, which has shown best results when compared to others two well known metrics in literature, namely, normalized cross correlation and mutual information. The displacement field was smoothed with a Gaussian convolution to constrain discontinuities, allowing parametric control of the smoothness in nonrigid transformation. Therefore the registration parameters can be adjusted to allow either local or global image transformation. The proposed method was evaluated for echocardiographic and breast magnetic resonance images presenting better accuracy. Keywords-nonrigid registration; similarity metric; ultrasound image;