Demo: 2D image RI2FL

2D RI2FL demo

from bioMONAI.data import *
from bioMONAI.transforms import *
from bioMONAI.core import *
from bioMONAI.core import Path
from bioMONAI.data import get_image_files, get_target, RandomSplitter
from bioMONAI.nets import BasicUNet, DynUNet
from bioMONAI.losses import *
from bioMONAI.metrics import *

from monai.utils import set_determinism

set_determinism(0)

device = get_device()
print(device)

cuda

Create Dataloader

bs, size = 16, 128

path = Path('../_data/Babesia/')
path_x = path/'RI'
path_y = path/'TRITC'


data_ops = {
    'blocks':       (BioImageBlock(cls=BioImageProject), BioImageBlock(cls=BioImage)),
    'get_items':    get_image_files,
    'get_y':        get_target(path_y, same_filename=False, signal_file_prefix='RI', target_file_prefix='TRITC'),
    'splitter':     RandomSplitter(valid_pct=0.2),
    'item_tfms':    [ScaleIntensity(minv=0.0, maxv=1.0),RandCrop2D(size), RandRot90(prob=0.5), RandFlip(prob=0.75)],
    'bs': bs,
}

data = get_dataloader(
    path_x, 
    show_summary=True,
    **data_ops,
    )

# print length of training and validation datasets
print('train images:', len(data.train_ds.items), '\nvalidation images:', len(data.valid_ds.items))

Setting-up type transforms pipelines
Collecting items from ../_data/Babesia/RI
Found 275 items
2 datasets of sizes 220,55
Setting up Pipeline: BioImageProject.create -> Tensor2BioImage -- {}
Setting up Pipeline: get_target.<locals>.generate_target_path -> BioImage.create -> Tensor2BioImage -- {}

Building one sample
  Pipeline: BioImageProject.create -> Tensor2BioImage -- {}
    starting from
      ../_data/Babesia/RI/O11_RI_frame20.tiff
    applying BioImageProject.create gives
      BioImageProject of size 1x512x512
    applying Tensor2BioImage -- {} gives
      BioImageProject of size 1x512x512
  Pipeline: get_target.<locals>.generate_target_path -> BioImage.create -> Tensor2BioImage -- {}
    starting from
      ../_data/Babesia/RI/O11_RI_frame20.tiff
    applying get_target.<locals>.generate_target_path gives
      ../_data/Babesia/TRITC/O11_TRITC_frame20.tiff
    applying BioImage.create gives
      BioImage of size 1x512x512
    applying Tensor2BioImage -- {} gives
      BioImage of size 1x512x512

Final sample: (BioImage([[[134., 134., 134.,  ..., 134., 134., 134.],
         [134., 134., 134.,  ..., 133., 134., 134.],
         [134., 134., 134.,  ..., 133., 133., 134.],
         ...,
         [133., 133., 134.,  ..., 133., 133., 133.],
         [134., 133., 134.,  ..., 133., 133., 133.],
         [133., 134., 134.,  ..., 133., 133., 134.]]]), BioImage([[[19., 19., 20.,  ..., 15., 16., 15.],
         [19., 20., 19.,  ..., 14., 16., 14.],
         [19., 18., 20.,  ..., 14., 14., 17.],
         ...,
         [19., 21., 20.,  ..., 17., 18., 16.],
         [19., 19., 20.,  ..., 17., 16., 17.],
         [20., 19., 19.,  ..., 16., 17., 17.]]]))


Collecting items from ../_data/Babesia/RI
Found 275 items
2 datasets of sizes 220,55
Setting up Pipeline: BioImageProject.create -> Tensor2BioImage -- {}
Setting up Pipeline: get_target.<locals>.generate_target_path -> BioImage.create -> Tensor2BioImage -- {}
Setting up after_item: Pipeline: ScaleIntensity -> RandCrop2D -- {'size': (128, 128), 'lazy': False, 'p': 1.0} -> RandRot90 -- {'prob': 0.5, 'max_k': 3, 'spatial_axes': (0, 1), 'ndim': 2, 'lazy': False, 'p': 1.0} -> RandFlip -- {'prob': 0.75, 'spatial_axis': None, 'ndim': 2, 'lazy': False, 'p': 1.0} -> ToTensor
Setting up before_batch: Pipeline: 
Setting up after_batch: Pipeline: 

Building one batch
Applying item_tfms to the first sample:
  Pipeline: ScaleIntensity -> RandCrop2D -- {'size': (128, 128), 'lazy': False, 'p': 1.0} -> RandRot90 -- {'prob': 0.5, 'max_k': 3, 'spatial_axes': (0, 1), 'ndim': 2, 'lazy': False, 'p': 1.0} -> RandFlip -- {'prob': 0.75, 'spatial_axis': None, 'ndim': 2, 'lazy': False, 'p': 1.0} -> ToTensor
    starting from
      (BioImageProject of size 1x512x512, BioImage of size 1x512x512)
    applying ScaleIntensity gives
      (MetaTensor of size 1x512x512, MetaTensor of size 1x512x512)
    applying RandCrop2D -- {'size': (128, 128), 'lazy': False, 'p': 1.0} gives
      (MetaTensor of size 1x128x128, MetaTensor of size 1x128x128)
    applying RandRot90 -- {'prob': 0.5, 'max_k': 3, 'spatial_axes': (0, 1), 'ndim': 2, 'lazy': False, 'p': 1.0} gives
      (MetaTensor of size 1x128x128, MetaTensor of size 1x128x128)
    applying RandFlip -- {'prob': 0.75, 'spatial_axis': None, 'ndim': 2, 'lazy': False, 'p': 1.0} gives
      (MetaTensor of size 1x128x128, MetaTensor of size 1x128x128)
    applying ToTensor gives
      (MetaTensor of size 1x128x128, MetaTensor of size 1x128x128)

Adding the next 3 samples

No before_batch transform to apply

Collating items in a batch

No batch_tfms to apply
None
train images: 220 
validation images: 55

data.show_batch(max_n=2, cmap='hot')

Setting affine, but the applied meta contains an affine. This will be overwritten.

Load and train a 2D model

from bioMONAI.nets import Deeplab, DeeplabConfig

config_2d = DeeplabConfig(
    dimensions=2,
    in_channels=1,  
    out_channels=1,
    backbone="resnet10",  
    aspp_dilations=[1, 6, 12, 18]
)
model = Deeplab(config_2d)
 
loss = MSSSIML1Loss(2, levels=2) #CombinedLoss(alpha=0, beta=0.5)
metrics = [SSIMMetric, MSELoss]

trainer = fastTrainer(data, model, loss_fn=loss, metrics=metrics, show_summary=False)

trainer.fit_flat_cos(500)

trainer.show_results(cmap='gray')

# trainer.save('tmp-model')

Test data

Evaluate the performance of the selected model on unseen data. It’s important to not touch this data until you have fine tuned your model to get an unbiased evaluation!