Method Article

Whole-brain Segmentation and Change-point Analysis of Anatomical Brain MRI—Application in Premanifest Huntington's Disease

DOI:

10.3791/57256

⸱

June 9th, 2018

 ,  ,  ,  ,  , 
1The Russell H. Morgan Department of Radiology and Radiological Science, Johns Hopkins University School of Medicine, 2Center for Imaging Science, Johns Hopkins University, 3Institute for Computational Medicine, Johns Hopkins University, 4Department of Applied Mathematics and Statistics, Johns Hopkins University, 5Division of Neurobiology, Departments of Psychiatry, Neurology, Neuroscience and Pharmacology, and Program in Cellular and Molecular Medicine, Johns Hopkins University School of Medicine, 6F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute, 7Department of Biomedical Engineering, Johns Hopkins University

In This Article

Summary

$$\rightleftharpoonup{xx}$$ $$\longleftharp{xx}$$, $$\longrightharp{xx}$$,

This paper describes a statistical model for volumetric MRI data analysis, which identifies the "change-point" when brain atrophy begins in premanifest Huntington's disease. Whole-brain mapping of the change-points is achieved based on brain volumes obtained using an atlas-based segmentation pipeline of T1-weighted images.

Abstract

$$\rightleftharpoonup{xx}$$ $$\longleftharp{xx}$$, $$\longrightharp{xx}$$,

Recent advances in MRI offer a variety of useful markers to identify neurodegenerative diseases. In Huntington's disease (HD), regional brain atrophy begins many years prior to the motor onset (during the "premanifest" period), but the spatiotemporal pattern of regional atrophy across the brain has not been fully characterized. Here we demonstrate an online cloud-computing platform, "MRICloud", which provides atlas-based whole-brain segmentation of T1-weighted images at multiple granularity levels, and thereby, enables us to access the regional features of brain anatomy. We then describe a regression model that detects statistically significant inflection points, at which regional brain atrophy starts to be noticeable, i.e. the "change-point", with respect to a disease progression index. We used the CAG-age product (CAP) score to index the disease progression in HD patients. Change-point analysis of the volumetric measurements from the segmentation pipeline, therefore, provides important information of the order and pattern of structural atrophy across the brain. The paper illustrates the use of these techniques on T1-weighted MRI data of premanifest HD subjects from a large multicenter PREDICT-HD study. This design potentially has wide applications in a range of neurodegenerative diseases to investigate the dynamic changes of brain anatomy.

Introduction

$$\rightleftharpoonup{xx}$$ $$\longleftharp{xx}$$, $$\longrightharp{xx}$$,

Magnetic resonance imaging (MRI) has substantially enhanced our ability to examine the brain anatomy and functions in neurodegenerative diseases1,2,3. T1-weighted structural MRI is one of most widely adopted imaging tools in routine clinical practice to assess the brain anatomy and related pathology. Quantitative analysis of the high-resolution T1-weighted images provides useful markers to measure anatomical changes during brain degeneration. In particular, segmentation based quantification approaches effectively reduces the image dimensionality from voxel level (on the order of (106)) to anatomical structural level ((102)) for high-throughput neuroinformatics4,5. Automated brain segmentation can be achieved using atlas-based methods6,7,8,9 that map the pre-defined anatomical labels from an atlas onto the patient images. Among the atlas-based methods, multi-atlas algorithms10,11,12,13,14 have yielded superior segmentation accuracy and robustness. Our group has developed a fully automated T1 multi-atlas segmentation pipeline, with advanced diffeomorphic image registration algorithms15, multi-atlas fusion methods16,17, and rich multi-atlas libraries18. The pipeline has been distributed on a cloud-computing platform, MRICloud19, since 2015, and it has been used to study neurodegenerative diseases, such as Alzheimer's disease (AD)20,21, Primary Progressive Aphasia22, and Huntington's Disease23.

Once the high-resolution images are segmented into brain structures, regional features, such as volumes, can be used to establish mathematical models to characterize the neuroanatomical changes. A change-point analysis method was recently established by our group to analyze the temporal order, in which statistically significant brain morphometric changes occur, based on longitudinal and/or cross-sectional MRI data. This statistical model was first developed to quantify shape-based diffeomorphometry over age in AD patients21,24; and it was later adapted to investigate brain structural changes in Huntington's disease (HD), as well as to describe brain developmental changes in neonatal brains25. In HD patients, the change-point was defined in with respect to the CAG-age product (CAP) score, as an indicator of the extent of exposure to the CAG expansion in HTT 26. It is well-known that striatal atrophy is one of the earliest markers in HD, followed by the globus pallidus27. Yet, the changes in striatum in relation to other gray and white matter structures across the brain remains unclear. Such relation is crucial for us to understand the disease progression. Change-point analysis of volumetric changes in all brain structures will likely provide systematic information of brain atrophy in premanifest phase of HD.

Here we demonstrate the procedures to perform whole-brain segmentation using MRICloud (www.mricloud.org), and steps to perform change-point analysis of volumetric data in premanifest HD subjects. The MRI data were collected from a large population multicenter PREDICT-HD study28,29 with approximately 400 controls and premanifest HD subjects. The combination of atlas-based segmentation and change-point analysis brings unique information about the spatiotemporal order of the brain structural changes and the disease progression pattern across the brain. The techniques are potentially applicable to a range of neurodegenerative diseases with various biomarkers to map the brain degeneration.

Access restricted. Please log in or start a trial to view this content.

Protocol

$$\rightleftharpoonup{xx}$$ $$\longleftharp{xx}$$, $$\longrightharp{xx}$$,

1. Atlas-based Whole Brain Segmentation

  1. Data preparation
    1. Convert three-dimensional (3D) T1-weighted images, typically acquired with MPRAGE (magnetization-prepared rapid gradient-echo) sequence, from vendor-specific DICOM (Digital Imaging and Communication) format to Analyzed format. Note that the cloud computation requires users' data to be transferred to remote clusters. According to the Health Insurance Portability and Accountability Act (HIPPA), remove the patients' personal identification information from the image files.
      NOTE: MRICloud offers a DICOM-to-Analyze converter (https://braingps.mricloud.org/t1prep) for the file format conversion as well as the de-identification of personal health information.
      1. Double-click to open Dcm2Analyze.exe. A popup window will open (Figure 1A).
      2. Specify the input DICOM data directory path as input and Analyzed image path and file name as output.
      3. Click "Go" to complete the conversion.
        NOTE: The conversion needs to be performed for each individual patient, and it would be beneficial to output all converted images in one study folder for batch processing (1.2.2.1).
  2. Multi-atlas based T1 image segmentation using MRICloud19.
    1. Log-in "Brain GPS" (register for first-time users) from https://mricloud.org. Select "Segmentation" tool from the main menu (Figure 1B). There are two application programming interface (API) options under "Segmentation": "T1-MultiAtlas" for single T1 image segmentation, and "T1-MultiAtlas Batch" for batch processing. The change-point analysis requires large population data, and thereby, batch processing is often the choice.
    2. Submit jobs on "T1-MultiAtlas Batch" API.
      1. Compress multiple Analyzed image files into a zip file. Click "+.zip" in Figure 1B to upload the zip file.
        NOTE: The current cloud portal limits the number of images to 30 in each zip folder. Large datasets can be separated into multiple zip folders to be processed. Special request can be made to accommodate more images per zip file. In the future, we may potentially increase the limited number or even remove the limit when more remote computational resources become available.
      2. Fill in the required fields in Figure 1B.
        1. Processing server: Choose "Computational Anatomy Science Gateway".
        2. Slice Type: Choose from "Sagittal", "Axial", or "Sagittal converted to Axial".
          NOTE: "Sagittal" refers to images acquired with sagittal orientation with in-plane view in Anterior-Posterior and Head-Foot directions. "Axial" refers to images acquired axial orientation with in-plane view in Left-Right and Anterior-Posterior directions. "Sagittal converted to Axial" refers to images that were originally acquired in axial orientation but later converted to sagittal images (those are typically "Axial" images with long neck area).
        3. Multi-atlas library: Choose the atlas library with closest age range to the user data to optimize the segmentation accuracy. We provide prebuilt multi-atlas datasets18 with different age ranges, e.g., "Pediatric 4-8yr", "Pediatric 8-12yr", "Adult 22-50", and "Adult 50-90", as well as different atlas versions. Information about atlas versions can be found in https://braingps.mricloud.org/atlasrepo.
    3. Check the job status through "My job status" (Figure 1C). Once the jobs are finished, a "Download Results" button will appear that allows users to download the segmentation results as a .zip file.
    4. Visualize the results. For single subject, the volumes obtained from the segmentation result can be visualized online (not feasible for batch processing results).
      1. Click "View result" button in Figure 1C. The webpage will turn to the visualization interface (Figure 1D). The axial, sagittal, and coronal views of the segmentation map are overlapped on the T1-weighted anatomical image. 3D rendering of the segmented brain structures are shown in the upper left window. Color of the overlaying segmentation map indicates the z-score of the structural volumes.
      2. Adjust the visualization options, including overlay ON/OFF, opacity of the overlay, zoom in and out, and slice positions from the upper right panel in Figure 1D.
      3. Group finest segmentation parcels into different granularity. In our atlases, we defined five levels of granularity with two types ontology relations18,30. A tree-view of the hierarchical multi-level anatomical definitions is shown in the lower left panel (Figure 1D). Click the level numbers to expand the structures at the corresponding ontology level. The segmentation maps will simultaneously switch to the corresponding ontology level.
        NOTE: Bi-monthly MRICloud workshops are help at Johns Hopkins University School of Medicine to offer hands-on tutorials of the online operations described above. Workshop Information can be found from https://braingps.mricloud.org/workshops.
    5. Download the segmentation results for subsequent analysis. Unzip the results to a user study directory, e.g., put together the segmentation results from all the subjects in one study directory, each in their individual subject folders.
      NOTE: The results include
    • example.img: original T1 image in axial orientation.
    • example_MNI.img: T1 image in MNI space following the Talairach coordinates.
    • example_7Labels.img and example_7Labels_MNI.img: Coarse segmentation with 7 parcels (gray matter, white matter, cerebrospinal fluid, lateral ventricle, skull, and background) in native and MNI space, respectively.
    • example_283Labels_M2.img and example_283Labels_M2_MNI.img: fine segmentation with 283 parcels (atlas version V9B) in native and MNI space, respectively. The exact number of labels depends on the atlas version.
    • example_corrected_MNI_stats.txt and example_MNI_stats.txt: statistics of volumes of each brain parcels in native and MNI space.
    • multilevel_lookup_table.txt: Multilevel ontology definition of the brain parcels.
  3. Perform batch processing to obtain brain volumes in a population. Use an in-house Matlab (www.mathworks.com) batch processing script to extract brain volumes from individual result folders and combine the volumetric data of all subjects at all granularity levels to a spreadsheet. A graphic user interface (GUI) is used to specify the input and output.
    1. Open Matlab.exe, run Main.fig, and a GUI will pop out (Figure 2A).
    2. In the "T1 Volume extraction from MRICloud" panel (upper panel in Figure 2A), specify the inputs, including the study directory where the downloaded the segmentation results are saved (see 1.2.5); and the multi-level lookup table file path and file name.
    3. Specify the output spreadsheet file path and file name where the volume data will be write to.
    4. Click "Extract volume" button to run the analysis. Results can be checked in the user-defined spreadsheet.
      NOTE: In addition, an R package is developed to process the MRICloud outputs and perform further statistical analysis, provided by Dr. Brian Caffo31. The package can be downloaded from https://github.com/bcaffo/MRIcloudT1volumetrics.

2. Change-point Analysis of Volumetric Data

NOTE: Please refer the theory and mathematical details of the change-point model to our previous publications21,24,23. Briefly, for HD data, a regression model is established to define a critical change-point (in terms of CAP score), starting from which, the linear regression between CAP and volume becomes statistical significant, while correcting for the effects of patient's age, gender, and intracranial volume. In-house Matlab scripts, along with a GUI (Figure 2A, lower panel), was developed to compute the change-points for individual structures and perform statistical analysis. The scripts are available to users upon request.

  1. Calculate change-points for individual brain structures.
    1. In the "Change-point analysis" panel in Figure 2A, specify the file path and name of the multilevel volume spreadsheet, which is generated according to 1.3.
    2. Specify the file path and name of the output text file, which the change-point results will be written to.
    3. Choose the level of granularity and type of ontology definition in the drop-down box (Level 1 - 5), at which the change-point analysis will be performed.
    4. Click "Calculate change-point" button to perform the change-point analysis, as described in reference23, and the resultant change-points will be saved in the output text file.
  2. Statistical evaluations of the change-points.
    1. In the Matlab GUI (Figure 2A, lower panel), specify parameters for statistical tests, including the number of permutation (default 10,000), number of bootstrap (default 10,000), and p-value threshold (after the FDR correction, default 0.05).
    2. Click "Statistical test" button to run the tests. After this step, the p-value (by permutation test), before and after false-discovery rate (FDR) correction, as well as the standard deviation and the 95% confidence interval (by bootstrap) of the change-points will be written to the output text file as extra columns. See details of statistical test procedures in23,24.
  3. Generate change-point maps (optional). In the Matlab GUI (Figure 2A, lower panel), click "Map change-point" button to generate change-point maps. Statistically significant change-point values will be mapped onto the MNI-space anatomical image to visualize the spatial pattern. This mapping can be done at different granularity levels, depending on the anatomical level specified in 2.1. The change-point maps can be overlaid on T1-weighted images using MRIcro (Figure 2B) (http://www.cabi.gatech.edu/mricro/mricro/).

Access restricted. Please log in or start a trial to view this content.

Results

$$\rightleftharpoonup{xx}$$ $$\longleftharp{xx}$$, $$\longrightharp{xx}$$,

Using the procedures described in 1.1-1.3, whole brain segmentation maps can be obtained from MRICloud. In the current version of atlas (V9B), 283 parcels are segmented at the finest granularity (level 5), which can be grouped to different levels of granularity, e.g., from hemisphere to lobules and parcels, according to specific ontology definitions. Figure 3 shows two types of multi-level segmentations at five levels, in axial and coronal views. For...

Access restricted. Please log in or start a trial to view this content.

Discussion

$$\rightleftharpoonup{xx}$$ $$\longleftharp{xx}$$, $$\longrightharp{xx}$$,

As demonstrated in this paper, whole-brain segmentation of brain MRI can be conveniently achieved using our online platform MRICloud. T1-weighted MRI based volumetric marker has shown to be robust and sensitive to a range of neurodegenerative diseases1,2,3. The volumetric measures are used for various downstream analysis, such as mathematical modeling, and feature-selection and classification analysis to assist clinical diagnosi...

Access restricted. Please log in or start a trial to view this content.

Disclosures

$$\rightleftharpoonup{xx}$$ $$\longleftharp{xx}$$, $$\longrightharp{xx}$$,

The authors have nothing to disclose.

Acknowledgements

$$\rightleftharpoonup{xx}$$ $$\longleftharp{xx}$$, $$\longrightharp{xx}$$,

We thank the PREDICT-HD investigators, particularly, Dr. Hans Johnson and Dr. Jane S. Pauslen from University of Iowa, for their generosity in sharing the MRI data and constructive discussion on the data analysis and results.

This work is supported by NIH grants R21 NS098018, P50 NS16375, NS40068, R01 NS086888, R01 NS084957, P41 EB015909, P41 EB015909, R01 EB000975, R01 EB008171, and U01 NS082085.

Access restricted. Please log in or start a trial to view this content.

Materials

List of materials used in this article
NameCompanyCatalog NumberComments
MATLABMathworksN/AVersion 2015b and above
Dell WorkstationDellDell Precision T5500 (Intel Xeon CPU)

References

$$\rightleftharpoonup{xx}$$ $$\longleftharp{xx}$$, $$\longrightharp{xx}$$,
  1. Paulsen, J. S., et al. Clinical and Biomarker Changes in Premanifest Huntington Disease Show Trial Feasibility: A Decade of the PREDICT-HD Study. Front Aging Neurosci. 6, 78(2014).
  2. Jack, C. R. Jr, et al. Hypothetical model of dynamic biomarkers of the Alzheimer's pathological cascade. Lancet Neurol. 9 (1), 119-128 (2010).
  3. Laakso, M. P., et al. Hippocampal volumes in Alzheimer's disease, Parkinson's disease with and without dementia, and in vascular dementia: An MRI study. Neurology. 46 (3), 678-681 (1996).
  4. Miller, M. I., Faria, A. V., Oishi, K., Mori, S. High-throughput neuro-imaging informatics. Front Neuroinform. 7, 31(2013).
  5. Mori, S., Oishi, K., Faria, A. V., Miller, M. I. Atlas-based neuroinformatics via MRI: harnessing information from past clinical cases and quantitative image analysis for patient care. Annu Rev Biomed Eng. 15, 71-92 (2013).
  6. Klein, A., Mensh, B., Ghosh, S., Tourville, J., Hirsch, J. Mindboggle: automated brain labeling with multiple atlases. BMC Med Imaging. 5, 7(2005).
  7. Heckemann, R. A., Hajnal, J. V., Aljabar, P., Rueckert, D., Hammers, A. Automatic anatomical brain MRI segmentation combining label propagation and decision fusion. Neuroimage. 33 (1), 115-126 (2006).
  8. Artaechevarria, X., Munoz-Barrutia, A., Ortiz-de-Solorzano, C. Combination strategies in multi-atlas image segmentation: application to brain MR data. IEEE Trans Med Imaging. 28 (8), 1266-1277 (2009).
  9. Lotjonen, J. M. P., et al. Fast and robust multi-atlas segmentation of brain magnetic resonance images. Neuroimage. 49 (3), 2352-2365 (2010).
  10. Rohlfing, T., Brandt, R., Menzel, R., Maurer, C. R. Evaluation of atlas selection strategies for atlas-based image segmentation with application to confocol microscopy images of bee brains. Neuroimage. 21 (4), 1428-1442 (2004).
  11. Fischl, B., et al. Whole brain segmentation: Automated labeling of neuroanatomical structures in the human brain. Neuron. 33 (3), 341-355 (2002).
  12. Collins, D. L., Holmes, C. J., Peters, T. M., Evans, A. C. Automatic 3-D model-based neuroanatomical segmentation. Human Brain Mapping. 3 (3), 190-208 (1995).
  13. Dawant, B. M., et al. Automatic 3-D segmentation of internal structures of the head in MR images using a combination of similarity and free-form transformations: Part I, methodology and validation on normal subjects. Ieee Transactions on Medical Imaging. 18 (10), 909-916 (1999).
  14. Wu, G., et al. A generative probability model of joint label fusion for multi-atlas based brain segmentation. Med Image Anal. 18 (6), 881-890 (2014).
  15. Miller, M. I., Trouve, A., Younes, Y. Diffeomorphometry and geodesic positioning systems for human anatomy. Technology. 2, (2013).
  16. Tang, X., et al. Bayesian Parameter Estimation and Segmentation in the Multi-Atlas Random Orbit Model. PLoS One. 8 (6), e65591(2013).
  17. Wang, H., et al. Multi-Atlas Segmentation with Joint Label Fusion. IEEE Trans Pattern Anal Mach Intell. 35 (3), 611-623 (2013).
  18. Wu, D., et al. Resource atlases for multi-atlas brain segmentations with multiple ontology levels based on T1-weighted MRI. Neuroimage. 125, 120-130 (2015).
  19. Mori, S., et al. MRICloud: Delivering High-Throughput MRI Neuroinformatics as Cloud-Based Software as a Service. Computing in Science & Engineering. 18 (5), 21-35 (2016).
  20. Wu, D., Ceritoglu, C., Miller, M. I., Mori, S. Direct estimation of patient attributes from anatomical MRI based on multi-atlas voting. Neuroimage-Clinical. 12, 570-581 (2016).
  21. Miller, M. I., et al. Network Neurodegeneration in Alzheimer's Disease via MRI Based Shape Diffeomorphometry and High-Field Atlasing. Front Bioeng Biotechnol. 3, 54(2015).
  22. Faria, A. V., et al. Content-based image retrieval for brain MRI: an image-searching engine and population-based analysis to utilize past clinical data for future diagnosis. Neuroimage Clin. 7, 367-376 (2015).
  23. Wu, D., et al. Mapping the order and pattern of brain structural MRI changes using change-point analysis in premanifest Huntington's disease. Hum Brain Mapp. , (2017).
  24. Younes, L., Albert, M., Miller, M. I., Team, B. R. Inferring changepoint times of medial temporal lobe morphometric change in preclinical Alzheimer's disease. Neuroimage Clin. 5, 178-187 (2014).
  25. Wu, D., et al. Mapping the critical gestational age at birth that alters brain development in preterm-born infants using multi-modal MRI. NeuroImage. 149, 33-43 (2017).
  26. Zhang, Y., et al. Indexing disease progression at study entry with individuals at-risk for Huntington disease. Am J Med Genet B Neuropsychiatr Genet. 156b (7), 751-763 (2011).
  27. Vonsattel, J. P., et al. Neuropathological classification of Huntington's disease. J Neuropathol Exp Neurol. 44 (6), 559-577 (1985).
  28. Paulsen, J. S., et al. Detection of Huntington's disease decades before diagnosis: the Predict-HD study. J Neurol Neurosurg Psychiatry. 79 (8), 874-880 (2008).
  29. Paulsen, J. S., et al. Prediction of manifest Huntington's disease with clinical and imaging measures: a prospective observational study. Lancet Neurol. 13 (12), 1193-1201 (2014).
  30. Djamanakova, A., et al. Tools for multiple granularity analysis of brain MRI data for individualized image analysis. Neuroimage. 101, 168-176 (2014).
  31. Caffo, B. A package for T1 volumetric analysis of MRIcloud output. R package version 0.0.1. , Available from: https://github.com/bcaffo/MRIcloudT1volumetrics (2017).
  32. Aljabar, P., Heckemann, R. A., Hammers, A., Hajnal, J. V., Rueckert, D. Multi-atlas based segmentation of brain images: atlas selection and its effect on accuracy. Neuroimage. 46 (3), 726-738 (2009).
  33. Huntington Study Group. Unified Huntington's Disease Rating Scale: reliability and consistency. Mov Disord. 11 (2), 136-142 (1996).
  34. Liang, Z., et al. Evaluation of Cross-Protocol Stability of a Fully Automated Brain Multi-Atlas Parcellation Tool. PLoS One. 10 (7), e0133533(2015).

Access restricted. Please log in or start a trial to view this content.

Reprints and Permissions

Request permission to reuse the text or figures of this JoVE article

Request Permission

Tags

T1 weighted MRIMRICloud PlatformAtlas based SegmentationVolumetric AnalysisCAP ScoreMultiAtlas BatchMATLAB GUI

Related Articles