Starting with the input data
We check whether the material meets the requirements of the selected method. Format, resolution, orientation, imaging coverage and completeness all matter. Preparing an image for computation must take into account how the output will later be overlaid on the source material.
Running a specified version of the method
Each analysis should identify the algorithm version and parameter set. We are developing execution through controlled queues, with a record of each run and limits on the resources it uses.
We work with both machine learning models and numerical image analysis methods. The choice depends on the project's objective, the properties of the data and how the result will be assessed.
Returning to the image
We compare the output with the input data. We check its geometry and completeness, then prepare it for inspection and further work. Depending on the agreed scope, this may be a segmentation, a set of spatial maps or a scene for opening in 3D Slicer.
Assessing software execution and the value of a method separately
A successfully completed computation confirms that a particular process has run. Whether the output is useful for a given application requires a separate assessment. We therefore agree on test scope, comparison material and acceptance criteria at the start of a project.