progress

Jan 31
Today I setup this github repo, worked through the simple ITK tutorials and went over the notes I had taken last year to try and remember what the scope of this project was. What I am going to work towards in the immediate future is getting a registraton workflow going on all of the patient data on a segment by segment basis, what I will need some guidance on is for the same spinal cord segment but but under different imaging sensitivities what how should I combine the registrations into a segmentation, is there a better segment to trust? or should I use more of a majority vote method. I also know I am going to get it wrong when it comes to mapping the mri images to the actual vertebrate they came from. Longer term I will need to infer what Sarah's code for extracting values from the segmented images should do and either tweak the code and my segmentations to work together or just write my own version and cut matlab completely out of the equation. I'm not going to try and fix sarah's registration code it is even more of black box now, I'm even hazier with matlab, and replacing it was part of the original scope of this project. There was certainly more within the scope of this project but I don't and probably didn't know what that was I sort of vaguely knew it had to do with finding medically relevant models to fit mri data to.

Feb 1
Played around with some of the segmentation methods in simpleITK, mostly the intensity threshold based ones all on some of the example data they provide. I think tomorrow I need to more quickly explore the rest of the segmentation toolbox and then try and see how I can get it to work on whichever sensitivity of the real data is the easiest.

Feb 2
Read the definitions for yesterday's segmentation algorithms, and tried a few more of the notebook examples. And think I've pretty much got the syntax figured out. Tomorrow I'll finish the edge detection exercise, look over the rest to see if there is something maybe relevant and then start with doing the white matter gray matter segmentation of an image from the real data set.

Feb 3
Only got through the last of the notebooks next day I'll start on the segmentation for the real data

Feb 6
Toyed with the real data and tried to see how similar images of the same spinal cord section were to each other

Feb 14
Loosely sorted out which images are less likely to be similar to each other and started working on seeing how well they register together to see how similar the images really are to each other.

Feb 15
Finished the translation registration of the images from the same spinal cord segment, tomorrow I'll try to get a primitive segmentation pipeline going

Feb 16
Made some seed regions and compared translation only registration to one with rotations

Feb 17
Start on segmentation of actual data

Feb 21
Got a workable automatic segmentation, there may be need to be manual removal of extra features or that may also go away with some slight parameter tweaking

Feb 22
Good usable segmentation script and started outputting segmentations based on the ActiveAx axon density images

Feb 23
Normalized image brightness to try and improve segmentation, also need to exclude the fissure from segmentations

Feb 24
Mostly read about others' approaches to gm segmentation particularly around the spinal cord grey matter segmentation challenge

Feb 26
Played with some of the registration tools to turn a one or two layer manual segmentation into an entire segmented volume

Feb 27
Re explored SCT as well as got my hands on DeepSeg, I am going to either use or implement something similar to https://doi.org/10.1016/j.neuroimage.2016.07.062 and if none of these are better than the seed growing approach I am going to bite the bullet and do it mostly manually

Mar 15
Used sct_deepseg with model seg_exvivo_gm-wm_t2 and looked into sct's templating pipeline

Mar 16
Breaking from segmentation and trying to register a hand drawn segmentation

# git issues to raise
- on 34 change slice_number=(0, image.GetSize()[1] - 1), to slice_number=(0, image.GetSize()[2] - 1), to match the previous cell
- results visualization example says Alaph blend rather than Alpha blend
- Using fiji the way I have been creates a memory leak where all old images remain in memory longer after they or fiji has been closed