Processing pipeline details

These pages describe what each part of the pipeline does, in what order, and which options change it. They stay at the level needed to choose options and read the reports and the methods boilerplate. The reasoning behind each correction, and the artifacts they address, will be covered in the forthcoming dMRI artifacts book.

The preprocessing stages. Anatomical processing runs alongside the DWI stages until coregistration; each stage has a page below.

QSIPrep builds one workflow per subject (or per session, with --subject-anatomical-reference sessionwise). It contains one anatomical workflow and one DWI workflow per output. The DWI workflow runs the stages below and, at the end, resamples the corrected data into ACPC space and writes the derivatives.

Stage	What happens	Page
Anatomical processing	Conform, brain extract, segment, AC-PC align and normalize the anatomical reference	Anatomical processing
Per-series preprocessing	Denoise and unring each series; concatenate; gradient nonlinearity field	Per-series preprocessing
Head motion and eddy currents	One of three backends	eddy (FSL), TORTOISE DIFFPREP, SHORELine
Susceptibility distortion	Estimated from a fieldmap, reverse phase encoding, or an anatomical reference; applied inside the head motion backend or after it	Susceptibility distortion correction
Coregistration and resampling	DWI reference, registration to the anatomical, one resampling into `ACPC` space, bias correction, merging	Coregistration and resampling

The DWI workflow for one output, with no fieldmap:

(Source code, png, svg, pdf)

The node names in these graphs are the ones that appear in the log (see Reading the log).