make super-resolution fly: quantitatively and at scale
-
task-specific super-resolution (SR)
- global tissue segmentation with priors
- local cortical segmentation
- hippocampus segmentation
- basal forebrain segmentation
- deep brain structure segmentation
- substantia nigra
- caudate, putamen, etc
-
general purpose methodology:
- simultaneous SR for intensity and segmentation probability pairs
- super-resolution multi-atlas segmentation (SRMAS)
- local joint label fusion for arbitrary segmentation libraries
- local joint label fusion for single templates with augmentation
tests provide a good example of use cases.
-
install:
python setup.py install
-
test:
python tests/test_segmentation.py
-
group 1: basal forebrain 75L, 76R
-
group2: hippocampus
- use deep hippocampus but 48L, 47R
-
group 3: cortex
- entorhinal: 117L, 116R
- parahippocampal: 171L 170R
- middle temporal gyrus: 155L 154R
- fusiform gyrus: 123L 122R
- total intracranial volume (ICV),
- caudate, => 37L, 36R
- putamen => 58L, 57R
- substantia nigra (SN) => is within 62L and 61R but should be dealt with separately
- globus pallidus => 56L, 55R
- corticospinal tract => TODO via registration from template
- supplementary motor area ( 183=left, 182=right - actually a subset of this but misses medial portion )
wlab = [36,37,57,58,61,62,55,56,183,182]
-
documentation
-
testing
- figure out how to distribute sr models
-
....