Convert NII.GZ to TGA

Technical note

How NII.GZ and TGA differ in practice

NII.GZ stores a volume rather than a picture: a three-dimensional grid of voxels, each with a physical size in millimetres, plus an orientation matrix that ties the grid to the patient's anatomy. TGA holds a flat image, so the conversion cuts the volume into slices along the axis most commonly used for viewing and writes each slice as its own numbered file. A 1 mm structural brain scan has somewhere around 180 slices, so a single NII.GZ comes back as a folder of about that many images, in order from one side of the scan to the other. The standard NII.GZ header has no fields for a patient's name, birth date or accession number, only an 80-character free-text description, which is one reason research pipelines convert DICOM to NIfTI before data is shared. That does not make every output anonymous. A head scan contains the face, and a 3D rendering of the volume can reconstruct it well enough to recognise someone, which is why public datasets are usually defaced first. Slices that pass through the front of the head carry the same risk.