Multi-Parametric MR Image Registration in Glioma Brain Tumors Using Multi-Similarity (RC and NMI) Measures Based on Wavelet Transform

The objective of this study is to align multi-parametric MR images of brain tumors using wavelet transformation and multi-similarity (RC and NMI) measures.
In this work, we implemented a multi-level non-rigid registration technique with multi-similarity measures for registration of perfusion- and diffusion–derived (rCBV and ADC) maps to morphological FLAIR images. To evaluate the performance of our proposed algorithm, we used synthetic data to test the robustness of the method to noise and intensity inhomogeneity. Finally, the algorithm was applied to multiparametric (FLAIR/rCBV-/ADC-maps) of 10 patients with glial tumors.
Evaluation of the proposed method on synthetic and real data revealed that this approach has a large capture range and is more robust against noise and intensity inhomogeneity without increasing the load and complexity of registration algorithm. The results for synthetic data contaminated with noise and intensity inhomogeneity based on Hausdorff Distance (HD), Root Mean Square Error (RMSE) and Baddeley's delta image metric (Δ) improved by 8%, 8% and 21% respectively. For real data, the overall performances based on RMSE and HD metrics were 28% and 10% for ADC to FLAIR registration, and 40% and 14% for rCBV map to FLAIR registration.
In this work, through the proposed multi-similarity measure combined with each other in different wavelet decomposition levels, the capture range of multiparametric image registration algorithm and robustness against noise and intensity inhomogeneity artifacts could be improved.