Robust Self-Similarity Descriptor for Multimodal Image Registration
Thuvanan Borvornvitchotikarn, Werasak Kurutach · 2018
Similarity measures of medical image registration (MIR) are challenged by the rotational deformations. They can affect the accuracy of MIR and also the effectiveness of medical diagnosis. In this work, we enhance the local self-similarity measure of MIR in terms of robustness against rotation variants by utilizing the index of neighbors at the different rotated locations. We have evaluated our proposed method on 3D MRI brain modalities (T1, T2, and PD modes). The experimental results have shown that our approach can reduce the error rate of rigid image registration with rotational transformations (on the average of 0.72 ± 0.09 mm.). Moreover, the proposed method achieves the best alignment accuracy compared to state-of-the-art methods.