When I was discussing the use
of JIST with Craigh, we came upon a new type of functionality, and
now Im struggling to find a way to implement it.
It would be great, maybe even essential, if a processing pipeline
not only returns the processed data, but also a set of quality
parameters. One might be interested in the mean error of some
fitting, or the registration distance needed to register two
volumes on each other. By eye-balling these quality parameters, one
would be able to point out clearly erroneous datasets.
Rather then to add a whole slew of new output parameters, it might
be easier (at least more backwards, compatible) to do this on some
type of logging level.
So, maybe introduce a specific logging level "quality assurance",
that is read out by a process and returned in a simple csv file or
such.
Another approach would be to add a single "Quality" parameter to
each module, which can contain a number of values. A single new
module could then collect all these quality parameters and return
them in a list view.
Again, the aim is to have a clear overview of the quality of all
crucial steps of the pipeline, where one would be able to spot
outlier datasets by their difference with respect to the output
scalars of regular datasets.
Let me know what you guys think, and what would be the best way to
proceed.
Daniel