I am working with the latest MITK 2018-04 release on ubuntu 16.04.
My main focus is on correcting fully automatically created segmentations, rather than creating segmentations from scratch.
Here are some thoughts that work accelerate my workflow when correcting fully automatically created segmentatios maps. (order with respect to priority)
- when one decides to save an image/ a segmentation (in my project, where each patient has is own folder) it would easy my life if the folder suggested from MITK would be the one where I loaded the images from. Then I would not have to find the right patient folder whenever I save a file
- the new feature "moving through time with arrow keys" is extremely useful. Therefore it would be desirable to be able to use it without having to click into the display window each time any action outside the display window is performed
- Region growing: tool works only if the label selected is not already present in the current slice. If I want to apply this tool I first have to erase the current segmentation and then create a new one. This is closely related to the wish of having a "region shrinking" tool
- shortcut to hide segmentation maps (all at once instead of just one by clicking). Even if opacity is set to <100% it is sometimes necessary to see exactly what is "behind" the segmentation map.
- Label selection: I would be helpful to be able to change the active label without having to "leave" the image. Comparable with the short cuts for the tool selection. Thinkable: press "l" and click segmentations label of choice -> this label will be active
- Delete all labels present in image instead of deleting all of them separately.
- Delete the same label for all slides sharing either the same time stamp or z-coordinate. Idea: a wrong label might be present in all or at least many time steps for the same z-coordinate
- be able to copy segmentation maps from one slice to another. Rational behind this: segmentation map for a structure of interest might be missing not only in the current but also in the neighboring (spacial and temporal) slices. The segmentation map is then likely to be similar in these slices. Being able to copy with lead to more consistent segmentation maps.
- recognize images (nifti in my case. Historically grown as far as i know. I talked to stefan and the advertised nrrd as Ralf did before...) that contains a segmentation and convert it automatically or allow for user defined settings with one click (e.g. 1. "reinit", 2. convert to segmentation, 3. set opacity of segmentation to 30%)
- point above could be integrated a "segmentation correction tool" besides "segmentation" and "multilabel segmentation"
- change opacity: by right clicking on an object in the data manager the opacity can be changed. Imagine the opacity is 100%. If you decided to go for opacity 20% you might want to click at one fifth of the opacity slider to reduce the opacity with one click only. What actually happens is that the opacity will be successively reduce from100% -> 90% -> 80%...
- A segmentation can be transformed into a polygon. This is handy to get an better impression of the object (organ, nodule, ...) in 3d. For multilabel segmentations it would be handy to have a polygon which keeps a color per label (instead of on color for all label, as it is at the moment)
- in case you want to focus more on hearts, it would be handy to see not only one study (short axis) but also be able to see a second study (long axis). That would especially become useful if one could show where the current slice in the short axis view would be in the long axis view (with or without taking misalignment due to breathing into account)
For the future
- having some of the best segmenters of the world in our department one could consider intelligent correction systems. Idea: where a fully automatic approach fails without user interaction, a semi-automatic approach should be able to tackle this problem after getting help form the rater