Metrics

Functions:

dice(im1, im2[, empty_score])

Computes the dice coefficient/F1 score, a measure of average similarity.

h_index(list_distributions)

Computes the h-index given a mixture of ditributions.

h_index_single_weighted(list_distributions)

Computes the single weighted h-index given a mixture of ditributions.

hausdorff_distance(mask_pred, mask_gt)

Calculates the Hausdorff distance for a given image, provided a ground truth.

jaccard(mask_gt, mask_pred)

Calculates the IoU/jaccard index for a pair of masks.

percent_content_captured(mask_a, mask_b)

Computes the content of mask_a captured in mask_b and returns it as a percent.

total_error(mask_gt, mask_pred[, weight_fn, ...])

Computes the percent of misclassified pixels on an image.
cell_analysis_tools.metrics.dice(im1, im2, empty_score=1.0)[source]

Computes the dice coefficient/F1 score, a measure of average similarity.

Parameters:

  • im1 (array-like, bool) – Any array of arbitrary size. If not boolean, will be converted.

  • im2 (array-like, bool) – Any other array of identical size. If not boolean, will be converted.

  • empty-score (int) – Any other array of identical size. If not boolean, will be converted.

Returns:

Dice coefficient as a float on range [0,1].

Maximum similarity = 1

No similarity = 0

Both are empty (sum eq to zero) = empty_score

Return type:

float

Note

The order of inputs for dice is irrelevant. The result will be identical if im1 and im2 are switched.

cell_analysis_tools.metrics.h_index(list_distributions) → float[source]

Computes the h-index given a mixture of ditributions.

Parameters:

list_distributions (list) – List of np.arrays containing estimated subdistributions

Returns:

calculated H-index

Return type:

float

Notes

When comparing H-index between datasets be sure to set the n_components paramter of the GaussianMixtureModel to the same value.

References

http://www.microscopist.co.uk/wp-content/uploads/2021/06/FLIM-review.pdf

cell_analysis_tools.metrics.h_index_single_weighted(list_distributions) → float[source]

Computes the single weighted h-index given a mixture of ditributions.

Parameters:

list_distributions (list) – List of np.arrays containing estimated subdistributions

Returns:

calculated H-index

Return type:

float

Notes

When comparing H-index between datasets be sure to set the n_components paramter of the GaussianMixtureModel to the same value.

cell_analysis_tools.metrics.hausdorff_distance(mask_pred, mask_gt)[source]

Calculates the Hausdorff distance for a given image, provided a ground truth. Both arrays must have the same number of columns. see https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance.directed_hausdorff.html :param mask_pred: predicted segmentation mask :type mask_pred: np.ndarray :param mask_gt: ground truth mask :type mask_gt: np.ndarray

Returns:

  • ddouble – The directed Hausdorff distance between arrays u and v,

  • index_1int – index of point contributing to Hausdorff pair in u

  • index_2int – index of point contributing to Hausdorff pair in v

cell_analysis_tools.metrics.jaccard(mask_gt, mask_pred)[source]

Calculates the IoU/jaccard index for a pair of masks.

Logic is implemented to require images of the same size.

Parameters:

  • mask_pred (np.ndarray) – Ground truth numpy ndarray image.

  • mask_gt (np.ndarray) – Current image numpy ndarray.

Returns:

jaccard_index – Calculated Jaccard index/distance.

Return type:

float

cell_analysis_tools.metrics.percent_content_captured(mask_a, mask_b)[source]

Computes the content of mask_a captured in mask_b and returns it as a percent. (A intersect B)/A # From the paper Mitohacker <https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7642274/>

Parameters:

  • mask_a (ndarray) – boolean mask.

  • mask_b (ndarray) – boolean mask.

Returns:

percent – percent of mask_a content captured by mask_b.

Return type:

float

cell_analysis_tools.metrics.total_error(mask_gt, mask_pred, weight_fn=1, weight_fp=1)[source]

Computes the percent of misclassified pixels on an image. In the case where one is more important than the other, a weighted average may be used: c0FP + c1FN

‘https://stats.stackexchange.com/questions/273537/f1-dice-score-vs-iou <https://stats.stackexchange.com/questions/273537/f1-dice-score-vs-iou>’_

Parameters:

  • mask_gt (bool ndarray) – groudn truth mask.

  • mask_predicted (bool ndarray) – DESCRIPTION.

  • weight_fn (int, optional) – scalar weight applied to false negatives. The default is 1.

  • weight_fp (int, optional) – scalar weight applied to false positives. The default is 1.

Returns:

percent – total error of the predicted mask given a ground truth mask.

Return type:

float