Store a nuclei and a membrane channel in the tfrecord for every example and use it as additional input during training.
Relevant background
Mike said for him it would be easier to tell if a cell is positive/negative for a given marker if he sees the nuclei and cell membrane that are visible in other channels. Thus we want to include two additional maps:
nuclei: which contains the mean over selected nuclei marker channels
membrane: which contains the mean over selected membrane marker channels
Also we'll include one additional feature called tissue_type.
Design overview
SegmentationTFRecords will get two additional inputs nuclei_channels and membrane_channels that expects a list of marker names. The class also gets a new class function make_composite(marker_list) that loads the images specified in the lists above and returns the mean over the channels. Then these two new channels will be stored as two new features nuclei_img and membrane_img in each example in the tfrecord. In class ModelBuilder we'll change prep_batches to accept a list of features. On first glance, augmentations should work with this additional channels without any changes.
Code mockup
TBD
Required inputs
nuclei_channels (list): contains the channel names that we turn into the nuclei channel
membrane_channels (list): containts the channel names that we turn into the membrane channel
Output files
.tfrecord file that contains two additional features nuclei_img and membrane_img.
Timeline
Give a rough estimate for how long you think the project will take. In general, it's better to be too conservative rather than too optimistic.
[ ] A couple days
[x] A week
[ ] Multiple weeks. For large projects, make sure to agree on a plan that isn't just a single monster PR at the end.
Estimated date when a fully implemented version will be ready for review:
Estimated date when the finalized project will be merged in:
Instructions
Store a nuclei and a membrane channel in the tfrecord for every example and use it as additional input during training.
Relevant background
Mike said for him it would be easier to tell if a cell is positive/negative for a given marker if he sees the nuclei and cell membrane that are visible in other channels. Thus we want to include two additional maps:
Also we'll include one additional feature called
tissue_type
.Design overview
SegmentationTFRecords
will get two additional inputsnuclei_channels
andmembrane_channels
that expects a list of marker names. The class also gets a new class functionmake_composite(marker_list)
that loads the images specified in the lists above and returns the mean over the channels. Then these two new channels will be stored as two new featuresnuclei_img
andmembrane_img
in each example in the tfrecord. In classModelBuilder
we'll changeprep_batches
to accept a list of features. On first glance, augmentations should work with this additional channels without any changes.Code mockup
TBD
Required inputs
nuclei_channels (list): contains the channel names that we turn into the nuclei channel membrane_channels (list): containts the channel names that we turn into the membrane channel
Output files
.tfrecord file that contains two additional features
nuclei_img
andmembrane_img
.Timeline Give a rough estimate for how long you think the project will take. In general, it's better to be too conservative rather than too optimistic.
Estimated date when a fully implemented version will be ready for review:
Estimated date when the finalized project will be merged in: