Open sriosq opened 1 week ago
We should do a couple of manual segs, and fine tune the existing model
Update:
There are some pockets of "lung"' out of place. Smoothing out the trachae segmentation now.
@evaalonsoortiz @mathieuboudreau This is the final smooth segmentation for db0_030.
Computing FM and assesing if this provides significant difference to the results.
Is the jaggedness at the edge of the lungs an image compression issue, or really what the segmentations are like now?
The lungs look like this:
I tried gaussian smoothign but it seems to not change. I don't want to over edit it I might change the segmentation drastically like before.
@mathieuboudreau Update:
Smoothing it outside segment and also the inside of the lungs, plus the skull and the trachea too!
Those look much better to me!
After a short meeting with Maxime B. we have seen that the trachea segmentation lacks the bifurcation which connects the trachea with lungs. This would be very challenging to segment with a trained model but with the anatomical T1w image it is possible to discern.
This may or may not be the reason why some subjects display fieldmap without the intensity peak around C7, but worth exploring the details.
A need for smoother lung segmentation as well as a revision of sharp discontinuities and wholes on the lung segmentation is also noted.