Eddy-current-induced distortion correction using maximum reconciled mutual information in diffusion MR imaging.

Area of Science:

• Medical Imaging
• Neuroimaging
• Diffusion Tensor Imaging

Background:

• Diffusion-weighted (DW) images in diffusion tensor imaging (DTI) are susceptible to distortions from head movements and eddy currents.
• Accurate registration of DW images is crucial for white matter structure tracking and tractography.
• Traditional registration methods using Shannon entropy-based measures like MMI/NMI can prematurely converge due to ignoring local image complexity.

Purpose of the Study:

• To address limitations of traditional DTI registration methods.
• To introduce a novel similarity measure, Maximum Reconciled Mutual Information (MRMI), for improved DW image registration.
• To enhance the accuracy and robustness of tractography and anatomical boundary alignment in DTI.

Main Methods:

• Developed a registration algorithm framework utilizing MRMI as the similarity measure.
• MRMI combines both global and local image information to overcome limitations of traditional entropy-based measures.
• Evaluated registration performance by comparing alignment quality and quantitative metrics against traditional methods.

Main Results:

• The MRMI method demonstrated superior alignment of DW images with b0 images compared to traditional methods.
• Fitted fractional anisotropy (FA) maps generated using MRMI showed closer approximation to true brain boundaries.
• Quantitative analysis confirmed a significant advantage of MRMI over other methods in terms of normalized mutual information (NMI) between b0 and aligned DW images.

Conclusions:

• MRMI-based registration achieves high spatial matching between b0 and DW images, enhancing robustness and accuracy.
• The proposed method offers a significant improvement over traditional diffusion image registration techniques.
• MRMI shows promise as a valuable addition to existing DTI software and multimodal medical image registration tools.