FastSurferCNN.utils.metrics¶

class FastSurferCNN.utils.metrics.DiceScore(num_classes, class_ids=None, device=None, one_hot=False, output_transform=<function DiceScore.<lambda>>)[source]¶

Accumulating the component of the dice coefficient i.e. the union and intersection.

Parameters:

opcallable(): A callable to update the accumulator. Method’s signature is (accumulator, output). For example, to compute arithmetic mean value, op = lambda a, x: a + x.
output_transformcallable(), optional: A callable that is used to transform the Engine’s process_function’s output into the form expected by the metric. This can be useful if, for example, you have a multi-output model and you want to compute the metric with respect to one of the outputs.
devicestr or torch.device, optional: Device specification in case of distributed computation usage. In most cases, it can be defined as “cuda:local_rank” or “cuda” if already set torch.cuda.set_device(local_rank). By default, if a distributed process group is initialized and available, the device is set to cuda.

Methods

`compute`([per_class, class_idxs])	Compute the Dice score.
`reset`()	Reset the state of the object.
`update`(output)	Update the internal state based on the output.

compute(per_class=False, class_idxs=None)[source]¶: Compute the Dice score.

reset()[source]¶: Reset the state of the object.

update(output)[source]¶

Update the internal state based on the output.

Parameters:

outputtuple of torch.Tensor: Tuple of predictions and labels.

FastSurferCNN.utils.metrics.dice_score(pred, gt, validate=True)[source]¶

Calculates the Dice Dissimilarity between pred and gt (best 0).

Parameters:

prednp.ndarray: Predicted image.
gtnp.ndarray: Ground truth image.
validatebool: If True, use the scipy implementation of the Dice Similarity. If False, use the numpy implementation.

Returns:

float: Dice Similarity between pred and gt.

FastSurferCNN.utils.metrics.hd(result, reference, voxelspacing=None, connectivity=1)[source]¶

Computes the (symmetric) Hausdorff Distance (HD) between the binary objects in two images. It is defined as the maximum surface distance between the objects.

Parameters:

resultarray_like: Input data containing objects. Can be any type but will be converted into binary: background where 0, object everywhere else.
referencearray_like: Input data containing objects. Can be any type but will be converted into binary: background where 0, object everywhere else.
voxelspacingfloat or sequence of floats, optional: The voxelspacing in a distance unit i.e. spacing of elements along each dimension. If a sequence, must be of length equal to the input rank; if a single number, this is used for all axes. If not specified, a grid spacing of unity is implied.
connectivityint: The neighbourhood/connectivity considered when determining the surface of the binary objects. This value is passed to scipy.ndimage.morphology.generate_binary_structure and should usually be \(> 1\). Note that the connectivity influences the result in the case of the Hausdorff distance.

Returns:

hdfloat: The symmetric Hausdorff Distance between the object(s) in result and the object(s) in reference. The distance unit is the same as for the spacing of elements along each dimension, which is usually given in mm.
hd50float: The 50th percentile of the Hausdorff Distance.
hd95float: The 95th percentile of the Hausdorff Distance.