[Paper Review] MRI-PET Registration with Automated Algorithm in Pre-clinical Studies

Nathanael L. Baisa, Stéphanie Bricq+1 more|arXiv (Cornell University)|May 19, 2017

Medical Imaging Techniques and Applications9 references3 citations

TL;DR

This paper presents a fully automated MRI-PET registration method for pre-clinical small animal studies using PCA-initialized global affine registration followed by local B-spline transformation, significantly improving accuracy over non-PCA-initialized methods. The approach achieves sub-millimeter target registration error (TRE) on brain and kidney datasets, outperforming manual initialization in both speed and precision.

ABSTRACT

Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) automatic 3-D registration is implemented and validated for small animal image volumes so that the high-resolution anatomical MRI information can be fused with the low spatial resolution of functional PET information for the localization of lesion that is currently in high demand in the study of tumor of cancer (oncology) and its corresponding preparation of pharmaceutical drugs. Though many registration algorithms are developed and applied on human brain volumes, these methods may not be as efficient on small animal datasets due to lack of intensity information and often the high anisotropy in voxel dimensions. Therefore, a fully automatic registration algorithm which can register not only assumably rigid small animal volumes such as brain but also deformable organs such as kidney, cardiac and chest is developed using a combination of global affine and local B-spline transformation models in which mutual information is used as a similarity criterion. The global affine registration uses a multi-resolution pyramid on image volumes of 3 levels whereas in local B-spline registration, a multi-resolution scheme is applied on the B-spline grid of 2 levels on the finest resolution of the image volumes in which only the transform itself is affected rather than the image volumes. Since mutual information lacks sufficient spatial information, PCA is used to inject it by estimating initial translation and rotation parameters. It is computationally efficient since it is implemented using C++ and ITK library, and is qualitatively and quantitatively shown that this PCA-initialized global registration followed by local registration is in close agreement with expert manual registration and outperforms the one without PCA initialization tested on small animal brain and kidney.

Motivation & Objective

• Address the challenge of accurate multi-modal image registration in small animal pre-clinical studies where standard human-based algorithms fail due to high anisotropy and low intensity contrast.
• Develop a robust, fully automatic registration pipeline that handles both rigid (e.g., brain) and deformable (e.g., kidney, heart) organs without pre-processing steps.
• Improve registration accuracy and convergence speed by integrating PCA to estimate initial translation and rotation parameters for mutual information-based optimization.
• Validate the method on diverse small animal datasets, including brain and kidney, to demonstrate generalizability across anatomical structures.
• Provide a computationally efficient solution using C++ and ITK for use in high-throughput oncology and drug development studies.

Proposed method

• Apply a multi-resolution pyramid with three levels to the image volumes for global affine registration to enhance convergence and reduce noise sensitivity.
• Use mutual information as the similarity metric between MRI and PET images, with Parzen window density estimation for joint probability computation.
• Introduce PCA on 3D spatial positions selected via the EM algorithm to estimate initial transformation parameters (translation and rotation), improving registration initialization.
• Implement a two-level multi-resolution B-spline grid for local non-rigid registration, refining the global alignment with spatially adaptive deformation.
• Apply the registration pipeline in sequence: PCA-initialized global affine registration followed by local B-spline refinement, both using the ITK library.
• Optimize the registration using gradient-free algorithms with mutual information as the cost function, minimizing divergence between modalities.

Experimental results

Research questions

• RQ1 Can a PCA-initialized global affine registration improve convergence and accuracy in MRI-PET registration of small animal datasets compared to non-initialized methods?
• RQ2 How effective is the combination of global affine and local B-spline transformations in registering both rigid and deformable small animal organs such as brain and kidney?
• RQ3 To what extent does the proposed method reduce target registration error (TRE) in multi-modal pre-clinical imaging without requiring pre-segmentation or masking?
• RQ4 How does the method perform on highly anisotropic MRI volumes and noisy PET images common in small animal studies?
• RQ5 Can the algorithm achieve reliable registration across diverse anatomical structures (e.g., brain, kidney, chest) without modality-specific tuning?

Key findings

• The PCA-initialized global affine registration reduced mean target registration error (TRE) to 0.56 mm (0.38 mm in brain) and 0.61 mm (0.47 mm in kidney), significantly outperforming the non-PCA-initialized method (1.07 mm and 1.23 mm, respectively).
• The addition of local B-spline registration further improved alignment, reducing TRE by up to 15% in the brain and 25% in the kidney compared to global-only registration.
• The method achieved faster convergence, requiring fewer iterations (42 vs. 46) and less time (36 sec vs. 40 sec) in brain registration when using PCA initialization.
• In monomodal MRI-MRI registration (brain and kidney), the PCA-initialized method achieved sub-millimeter TRE (0.07 mm in brain, 0.15 mm in kidney), demonstrating high precision in same-modality alignment.
• The mutual information (MI) quality metric improved with PCA initialization (e.g., -0.408 in brain vs. -0.402 without), indicating better similarity between registered images.
• The algorithm successfully registered highly anisotropic datasets (e.g., 0.156×0.156×1.8 mm³ in brain MRI) and noisy PET scans (0.387×0.387×0.775 mm³), showing robustness to image quality variations.