| from matplotlib import pyplot as plt |
| import torchio as tio |
|
|
| from auto_detect_breast_mri.data.metadata import get_uka_metatensor |
| from auto_detect_breast_mri.data.transforms import RandomCropOrPad, ZNormalization, ImageToTensor |
| from auto_detect_breast_mri.data.breast_mri_dataset import BreastMRISubjects |
| from auto_detect_breast_mri.data.loaders import get_multiple_subjects_dataloader |
| from auto_detect_breast_mri.config import resolve_path |
|
|
| |
| path_base = resolve_path(None, "data_root", "root folder of the NIfTI data") |
| subset_path = resolve_path(None, "split_root", "folder holding the split files") |
| subset_file = "fold0/stratified_training_set-f0-0.csv" |
| feature_path = resolve_path(None, "metadata_file", "metadata export") |
| pre_image_shape = (32, 512, 512) |
| protocol = ['Sub_1'] |
| train_prop = 0.7 |
| train_fraction = 0.5 |
| batch_size = 4 |
| transform = tio.Compose([ |
| tio.RandomFlip((0,1,2), flip_probability=0.5), |
| RandomCropOrPad(pre_image_shape), |
| ZNormalization(per_channel=True, percentiles=(0.5, 99.5), masking_method=lambda x:x>0), |
| tio.RandomNoise(std=(0.25,0.5)), |
| ImageToTensor() |
| ]) |
|
|
| data_set = BreastMRISubjects(path_base, subset_path + subset_file, protocol=protocol, transform=transform) |
|
|
| feature_dataframe = get_uka_metatensor(0, feature_path) |
| train_loader, eval_loader, test_loader = get_multiple_subjects_dataloader(path_base, feature_dataframe, |
| pre_image_shape, transform, protocol, |
| subset_path, batch_size, stratified=True, |
| fraction=train_fraction, fold=0, subfold=0) |
|
|
| for i, batch in enumerate(train_loader): |
| image = batch['image']['data'] |
| label = batch['label'] |
| plt.imshow(image[0, 0, :, :, 14], cmap='gray', interpolation=None) |
| plt.show() |
| print(label[0]) |
| print("-------") |
|
|
