Input Data

Data Format

Brainchop is designed with the current version to support T1-weighted MRI volume segmentation. The input must be a T1 brain volume in the Nifti format.

Preprocessing Data

T1 image needs to be in volumetric shape 256x256x256, scaled and resampled to 1x1x1 mm voxels as a preprocessing step for proper results. This preprocessing can be made automatically in Brainchop tool using mri_convert.js, or can be made by using FreeSurfer's CLI command:

                        mri_convert original_T1.nii  new_T1.nii.gz  -c

Also conform, a python version of mri_convert, can be used for the T1 image correction from the terminal such that:

                         python conform.py  -i <input> -o <output>

After selecting the T1 weighted image for 3D segmentation, Brainchop applies different preprocessing steps underneath, such as:

• Converting data to JavaScript readable format.
• Converting input data to tensors to deploy tfjs framework for the 3D inference.
• Convert tensors to batches for Grey Matter/White Matter inference.
• Apply image normalization to enhance model accuracy.