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

The overall goal of this image analysis pipeline is to perform quantitative volumetric analysis of T1 weighted MR images and to identify critical change-points when brain atrophy begins in neurodegenerative diseases. This method can help answer key questions in quantitative analysis of brain anatomical changes during the course of neurodegenerative disease. For example, the order and the pattern of brain atrophy in the Huntington’s disease.

The main advantage of this technique is that the combination of Atlas-based segmentation and change-point analysis bring unique spatial temporal information about brain degeneration from large population of studies. The volumetric analysis of brain MRI was achieved by MRICloud, which provides fully automated image segmentation and quantification through remote computation and web-based user interface. Change-point analysis of the brain MRI provides us a systematic view of the disease progression across the whole brain.

The applications of this technique extend toward a range of neural degenerative disease. To begin this procedure, convert three dimensional T1 weighted images from vendor-specific DICOM format to analyzed format. Double click to open Dcm2Analyze.exe.

A pop up window will open. Specify the input DICOM data directory path as input and analyzed image file path and name as output. Then, click go to complete the conversion.

For MultiAtlas-based T1 image segmentation using MRICloud, log in BrainGPS and select segmentation tool from the main menu. There are two options under segmentation, the T1-MultiAtlas for single T1 image segmentation, and T1-MultiAtlas Batch for batch processing. Submit jobs on T1-MultiAtlas Batch API by compressing multiple analyzed image files into a ZIP file, and then click zip to upload the ZIP file.

Now, fill in the required fields. For processing server, choose Computational Anatomy Science Gateway. For slice type, choose from Sagittal, Axial, or Sagittal converted to Axial.

For MultiAtlas library, choose the Atlas library with the closest age range to the user data to optimize the segmentation accuracy. Check the job status through my job status. Once the jobs are finished, a download results button will appear that allows users to download the segmentation results as a ZIP file.

Results will be saved in individual folders. Visualize the results by clicking the view results button. The web page will turn to the visualization interface.

The Axial, Sagittal, and Coronal views of the segmentation map are overlapped on the T1 weighted anatomical image. 3-D 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.

Adjust the visualization options, including overlay on/off, opacity of the overlay, zoom in and out, and slice positions from the upper right panel. To perform batch processing to obtain brain volumes in a population, use an in-house MATLAB program to extract individual brain volumes and combine them in a spreadsheet. In MATLAB, run main.

m, and a GUI will pop out. In the T1 volume extraction from MRICloud panel, specify the inputs, including the study directory where the downloaded segmentation results are saved, and the multi-level lookup table file path and file name. Specify the output directory of the spreadsheet file where the volume data will be written to.

Click the extract volume button to run the analysis. A message will show after successful completion. Results can be checked in the spreadsheet, where the rows contain individual subject data and the columns contain the structural volumes of these individuals defined at multiple anatomical granularities.

To calculate change-points for individual brain structures, use the same MATLAB GUI. In the change-point analysis panel, specify the spreadsheets that contain the multi-level volume data and the associated diagnostic information. Then, specify the directory of the output text file, which the change-point results will be written to.

Afterward, choose the level of granularity and type of ontology definition in the drop down box, at which the change-point analysis will be performed. Also, choose the main repressor to be used in the change-point analysis, for example, CAP score for Huntington’s disease. Click the calculate change-point button to perform the analysis and the resultant change-points will be saved in the user-defined output excel file.

To perform statistical evaluations of the change-points in the MATLAB GUI, specify the parameters for statistical tests, including the number of permutation and number of bootstrap. Click the statistical test button to run the tests. After this step, the p-values from the permutation test as well as the mean and standard deviation from the bootstrap operation will be written to the output text file as extra columns.

Optionally, to map the change-points into the brain and 3-D space, click map change-point. After this step, and change-point map will be generated and visualized by overlying on the T1 weighted image. Shown here are the representative results of the Atlas-based whole brain segmentation at multiple granularity levels.

Axial and Coronal views of multi-level segmentation maps are overlaid on T1 weighted anatomical images. The hierarchical anatomical relations between the hemisphere the cerebral nuclei, the basal ganglia of the cerebral nuclei, the striatum and globus pallidus of the basal ganglia, and putamen and caudate of the striatum are illustrated in 3-D. Here, the scatter plots demonstrate the change-point analysis of these structures, where the blue dots denote the z-scores of volumetric data after correcting for age, sex, and intercranial volumes.

The black curves are the fitted z-scores, with regression to the change-point dependent CAP scores, and the red lines indicate the positions of the change-points. Here are the whole-brain change-point maps at multiple granularity levels. The regions that show significant change-points are mapped onto a T1 weighed image, and the colors indicate the change-point values in unit of CAP score.

Once mastered, this procedure can be accomplished in several hours for the image segmentation part, depending on the availability of the cloud resource. And, another few hours for the change-point analysis, depending on the user-defined level of anatomical granularity. While attempting this procedure, it is important to always check the quality of the original image and accuracy of image segmentation before proceeding to the change-point analysis.

After its development, this technique paved the way for researchers in the field of medical image analysis to explore whole-brain anatomical change and its spatial temporal pattern in large MRI studies of neural degenerative diseases. After watching this video, you should have a good understanding of how to perform automated segmentation of T1 weighted images and how to use the MATLAB GUI to run change-point analysis of photometric data.