Accepts an 3D numpy array and shows median slices in all three planes :param img_numpy:

:param ls: list of modality paths, where the last path is the segmentation map :param dataset_name: which dataset is used :param mode: t

Converts NxDxHxW label image to NxCxDxHxW, where each label gets converted to its corresponding one-hot vector :param input: 4D input image (

Checks a path if exist in the other list

:param iter: iteration or partial epoch :param epoch: epoch of training :param loss: any loss numpy :param mode: tra

Flattens a given tensor such that the channel axis is first. The shapes are transformed as follows: (N, C, D, H, W) -> (C, N * D * H * W)

:param img_numpy: 3D medical image modality :param labels: 3D medical image labels :param alpha: scaling factor of gaussian filter :p

Returns total 3d input volumes (t1 and t2 or more) and segmentation maps 3d total vol shape : torch.Size([1, 144, 192, 256])

Returns a random rotated array in the same shape :param img_numpy: 3D numpy array :param min_angle: in degrees :param max_angle: in d

:param img_numpy: :param min_percentage: :param max_percentage: :return: zoom in/out aigmented img

:param iter: iteration or partial epoch :param loss: any loss torch.tensor.item() :param channel_score: per channel score or

Comparative 2d vizualization of median slices: axial, saggital and transpose. Save to png file and to tensorboard :param predictions:

Random center crop near the label area

Full training logic

Cut off outliers that are related to detected black in the image (the air area)

:param iter: iteration or partial epoch :param epoch: epoch of training :param loss: any loss numpy :param mode: tra

Accepts an image tensor and normalizes it :param normalization: choices = "max", "mean" , type=str

Resize the 3d numpy array to the dim size :param out_dim is the new 3d tuple

setup GPU device if available, move model into configured device

Training logic for an epoch :param epoch: Current epoch number

Computes DiceCoefficient as defined in https://arxiv.org/abs/1606.04797 given a multi channel input and target. Assumes the input is a norma

Converts label image to CxHxW, where each label gets converted to its corresponding one-hot vector :param target is of shape

Dynamically computes padding to keep input conv size equal to the output for stride = 1 :return: