Lesion Segmentation Training

If the Brain Atrophy tool does not perform well in segmenting FLAIR hyperintense white-matter lesions, perhaps because your images have unusual contrast, you can train the segmentation using your own images. Training is in three steps:

1. Prepare the images. You will need a T₁-weighted and a FLAIR image for each training subject. You should:
   • Place each subject's T₁-weighted and a FLAIR image in its own folder.
   • Rename the T₁-weighted and a FLAIR image to have some consistency across all subjects. For example, you could name them "t1.nii" and "flair.nii".
   In the Brain Atrophy tool, for each subject, setup as for cross-sectional atrophy with WM lesions, but select . Set a Basename to use for the registered prior image, for example "Training". Now press the button, which will perform all image registrations needed.

2. When the stage 1 is complete, you need to outline the hyperintense lesions on the registered FLAIR image. This image will have the same name as the input FLAIR image, but with the prefix "r" to indicate that it has been registered to the T₁-weighted image.

   Load the registered FLAIR image into Jim's main display, and launch the ROI Toolkit. Use Contoured, Irregular ROIs to outline the WM hyperintensities on every slice. Then save the ROIs to file, using the default file name (i.e., if your registered FLAIR image is called "rflair.nii", then the ROI file name must be "rflair.roi").

   Repeat for all subjects in your training set. The number of subjects needed for training will vary according to their lesion loads, but we suggest that at least 20 subjects will be needed, and possibly more.

3. When you have a complete set of subjects, run the training. The tool for training is only available as the stand-alone tool RegionTrain.

   

   In this tool set:
   • Name tags for the T₁-weighted and FLAIR images. The image names for every subject must contain these tags. For example, you could have a series of patients, each in their own separate folder, with name "Pat1t1.nii" & "Pat1flair.nii", "Pat2t1.nii" & "Pat2flair.nii" etc.
   • The image Basename. This must be the same Basename that you used in step 1 (e.g., "Training").
   • Select all the folders that contain the training subjects' images and ROI file. There will be one folder for each training subject. You can select a single "top-level" folder if all the different subjects are within this folder.
   • Set the name of the file to which you want to save the region statistics. This should have a file extension ".ser".
   Then click the button. Training will start, processing each training subject in turn. When, complete, the region statistics file will be created, and can be used as input to the atrophy assessment with WM lesions, either cross-sectional or longitudinal.