TIRL: Automated Non-Linear Registration of Stand-Alone Histological Sections to Whole-Brain MRI

Presented During:


Poster No:

1270 

Submission Type:

Abstract Submission 

Authors:

Istvan Huszar1,2, Menuka Pallebage-Gamarallage2, Sean Foxley3, Benjamin Tendler1,2, Anna Leonte4, Marlies Hiemstra5, Jeroen Mollink5,1,2, Adele Smart2, Sarah Bangerter-Christensen6, Hannah Brooks2, Martin Turner2, Olaf Ansorge2, Karla Miller1,2, Mark Jenkinson1,2

Institutions:

1FMRIB, Wellcome Centre for Integrative Neuroimaging, University of Oxford, Oxford, United Kingdom, 2Nuffield Department of Clinical Neurosciences, University of Oxford, Oxford, United Kingdom, 3Department of Radiology, University of Chicago, Chicago, IL, 4University Medical Center Groningen, University of Groningen, Groningen, the Netherlands, 5Department of Anatomy, Donders Institute for Brain, Cognition and Behaviour, Radboud UMC, Nijmegen, the Netherlands, 6Brigham Young University, Provo, UT, United States

First Author:

Istvan Huszar, MD  
FMRIB, Wellcome Centre for Integrative Neuroimaging, University of Oxford|Nuffield Department of Clinical Neurosciences, University of Oxford
Oxford, United Kingdom|Oxford, United Kingdom

Co-Author(s):

Menuka Pallebage-Gamarallage, PhD  
Nuffield Department of Clinical Neurosciences, University of Oxford
Oxford, United Kingdom
Sean Foxley, PhD  
Department of Radiology, University of Chicago
Chicago, IL
Benjamin Tendler, PhD  
FMRIB, Wellcome Centre for Integrative Neuroimaging, University of Oxford|Nuffield Department of Clinical Neurosciences, University of Oxford
Oxford, United Kingdom|Oxford, United Kingdom
Anna Leonte, MSc  
University Medical Center Groningen, University of Groningen
Groningen, the Netherlands
Marlies Hiemstra, MSc  
Department of Anatomy, Donders Institute for Brain, Cognition and Behaviour, Radboud UMC
Nijmegen, the Netherlands
Jeroen Mollink, PhD  
Department of Anatomy, Donders Institute for Brain, Cognition and Behaviour, Radboud UMC|FMRIB, Wellcome Centre for Integrative Neuroimaging, University of Oxford|Nuffield Department of Clinical Neurosciences, University of Oxford
Nijmegen, the Netherlands|Oxford, United Kingdom|Oxford, United Kingdom
Adele Smart  
Nuffield Department of Clinical Neurosciences, University of Oxford
Oxford, United Kingdom
Sarah Bangerter-Christensen  
Brigham Young University
Provo, UT, United States
Hannah Brooks  
Nuffield Department of Clinical Neurosciences, University of Oxford
Oxford, United Kingdom
Martin Turner, MA MBBS PhD FRCP  
Nuffield Department of Clinical Neurosciences, University of Oxford
Oxford, United Kingdom
Olaf Ansorge, MD, PhD  
Nuffield Department of Clinical Neurosciences, University of Oxford
Oxford, United Kingdom
Karla Miller, PhD  
FMRIB, Wellcome Centre for Integrative Neuroimaging, University of Oxford|Nuffield Department of Clinical Neurosciences, University of Oxford
Oxford, United Kingdom|Oxford, United Kingdom
Mark Jenkinson, BSc BE DPhil  
FMRIB, Wellcome Centre for Integrative Neuroimaging, University of Oxford|Nuffield Department of Clinical Neurosciences, University of Oxford
Oxford, United Kingdom|Oxford, United Kingdom

Introduction:

Advanced MRI methods are sensitive to tissue properties at much finer scales than the resolution of a clinical MRI scan. Consequently, it is of great interest what the MRI signal can reveal about the healthy tissue microstructure, and how it is affected by disease. Conversely, most existing biophysical models concern healthy conditions, and radiological findings in patients are rarely followed up by post-mortem histological validation, leaving the radiological and histopathological understanding of human neurodegeneration detached. Bridging the gap requires precise alignment between MRI and histology that so far has been predominantly addressed for serial sections. However, the costs of this technique are prohibitively high to take disease heterogeneity into account by studying a multitude of brains. Here we report a customisable image registration platform to automate the alignment of conventional small-slide histological sections to whole-brain post-mortem MRI as an alternative.

Methods:

Tensor Image Registration Library (TIRL) was written in Python to provide an integrated registration workflow for various different kinds of images and 2D-to-3D problems. TIRL has a modular design (Fig. 1): every component of the workflow is an object that allows further customisation by subclassing. At the core of the library is the universal tensor image object (TImage), whose values are L-rank tensors defined on an N-dimensional domain. TImage can readily import data from most image formats (including MRI and microscopy) and also handles images that are larger than memory. TImage extends the concept of the sform matrix of the NIfTI format by defining image data on a voxel Domain, from which physical coordinates are calculated by a user-defined chain of linear and non-linear Transformation objects. These can be grouped arbitrarily for simultaneous optimisation by gradient-based or gradient-free algorithms in the SciPy and NLOpt libraries. Finally, TIRL allows mixing multiple cost functions and regularisation methods.
Supporting Image: fig1.png
 

Results:

To test TIRL, a 3-stage pipeline was built (Fig. 2A) to register 6 stand-alone histological sections (of 4 different stains) from a human brain to the corresponding post-mortem MR data [1, 2]. Dissection photographs were used as intermediates to automate initialisation steps. The first stage employed diffusion-regularised non-linear registration [3] driven by the MIND metric [4] and achieved 0.25-0.70 mm registration accuracy (stain-dependent) between histological sections and corresponding tissue block photographs. The second stage reinserted tissue blocks into coronal brain slice photographs by testing automatically detected insertion sites and performing MIND-driven linear and non-linear refinement steps with <0.2 mm final registration error. The third stage performed slice-to-volume registration of the coronal slice photographs to 3D MRI by automatic initialisation along the AP axis, then using the BOBYQA algorithm [5] to optimise 3D affine parameters based on the MIND metric. Finally, slice curvature was introduced by optimising a 3D deformation field defined by a set of radial basis functions on the slice surface. The results consistently outperformed manual alignment by corresponding anatomical landmarks. Based on simulated slices from the MRI, the accuracy of this stage was 0.04-0.89 mm in the range of expected initial deformations. Identical high-quality registrations were later obtained from a total of 143 slices in 15 brains. Final results are shown in Fig. 2.
Supporting Image: fig2_proportional.png
 

Conclusions:

A versatile image registration platform was developed and tested by successfully implementing an automated MRI-histology registration pipeline. On the one hand the pipeline provides grounds for large-scale validation of current imaging biomarkers. On the other hand, creating voxel-wise aligned MRI-histology datasets allows learning algorithms to derive quasi-histological information from future clinical MRI scans, which may increase the specificity of imaging biomarkers.

Disorders of the Nervous System:

Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s)

Modeling and Analysis Methods:

Image Registration and Computational Anatomy 1
Methods Development 2

Novel Imaging Acquisition Methods:

Multi-Modal Imaging

Keywords:

MRI
Other - histology; registration

1|2Indicates the priority used for review

My abstract is being submitted as a Software Demonstration.

Yes

Please indicate below if your study was a "resting state" or "task-activation” study.

Other

Healthy subjects only or patients (note that patient studies may also involve healthy subjects):

Patients

Was any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.

Yes

Was any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.

Not applicable

Please indicate which methods were used in your research:

Structural MRI
Optical Imaging
Postmortem anatomy

For human MRI, what field strength scanner do you use?

7T

Which processing packages did you use for your study?

FSL
Other, Please list  -   Python; TIRL

Provide references using author date format

1. Huszar, I.N., et al., Tensor Image Registration Library: Automated Non-Linear Registration of Sparsely Sampled Histological Specimens to Post-Mortem MRI of the Whole Human Brain. bioRxiv, 2019: p. 849570.
2. Pallebage-Gamarallage, M., et al., Dissecting the pathobiology of altered MRI signal in amyotrophic lateral sclerosis: A post mortem whole brain sampling strategy for the integration of ultra-high-field MRI and quantitative neuropathology. BMC Neurosci, 2018. 19(1): p. 11.
3. Modersitzki, J., Numerical Methods for Image Registration. 2003: Oxford University Press.
4. Heinrich, M.P., et al., MIND: modality independent neighbourhood descriptor for multi-modal deformable registration. Med Image Anal, 2012. 16(7): p. 1423-35.
5. Powell, M.J.D., The BOBYQA algorithm for bound constrained optimization without derivatives. 2009, Cambridge University.