library("rplatypus")
cases <- list.dirs("dicom_export", recursive = FALSE)
report <- series_report(cases)
report[!report$ok, c("path", "problem")] # the ones to deal with first
table(report$sorted_by)Check folders of DICOM slices before training on them
Description
One row per directory, saying whether it is a usable series and what is wrong with it if it is not. Problems are reported as a column rather than raised, because a hundred cases out of an archive will contain a few with a missing slice, two series in one folder, or duplicated files - and stopping at the first turns an afternoon’s work into a week’s.
Usage
series_report(paths)
Arguments
paths
|
Directories, each holding the slices of one series. |
Details
Reads no pixels. Every check runs on the DICOM headers, so this is usable on a hundred gigabytes of data and is the first thing worth running on a new dataset.
Value
A data frame with one row per directory: path, ok, slices, sorted_by, spacing_1/2/3, series_uid and problem.
What it catches
A missing slice, which does not make a volume shorter - it puts everything past the gap in the wrong place, and no metric would show it. Two series in one folder, which is the normal state of an archive export: a scout, a reconstruction and a phase sitting together, which stacked would interleave two anatomies. Duplicated slices, from a copy or from several echoes mixed together. Slices of different size, angle or spacing.
It also reports how the slices were ordered. “position” means the geometry decided, which is what you want. “instance_number” means the files carry no positions and the order came from a number that is only required to be unique - worth knowing before trusting the result.
See Also
volume_info() for NIfTI files, split_dataset() for dividing the cases up.