Different analysis pipelines, different results... How to deal with analytical variability?
Abstract
Data processing is central to biomedical imaging and over the years the different tools and
algorithms available to study a dataset have multiplied. Recently, multiple studies in the literature have
demonstrated how the exact choice of analysis pipeline can have an impact on the conclusions of scientific
experiments. Those studies focused on varying different components of the analysis pipeline including
differences observed across scientific software, across preprocessing pipelines or even across different
operating systems. In addition, the development of many-analyst studies -- in which independent teams of
experts are tasked to answer the same research questions using the same input data -- have shed lights into
how much differences can be expected across pipelines selected by experts from the scientific community. In
this talk, we review different examples showcasing this analytical variability. We will reflect on the underlying reasons that can explain the existence of this variability.
Origin | Files produced by the author(s) |
---|---|
Licence |