bpdl.metric_similarity module

Introducing some used similarity measures fro atlases and etc.

Copyright (C) 2015-2020 Jiri Borovec <jiri.borovec@fel.cvut.cz>

bpdl.metric_similarity.compare_atlas_adjusted_rand(a1, a2)[source]

using adjusted rand and transform original values from (-1, 1) to (0, 1) http://scikit-learn.org/stable/modules/generated/sklearn.metrics.adjusted_rand_score.html

Parameters

  • a1 – np.array<height, width>

  • a2 – np.array<height, width>

Return float

with 0 means no difference

>>> atlas1 = np.zeros((7, 15), dtype=int)
>>> atlas1[1:4, 5:10] = 1
>>> atlas1[5:7, 6:13] = 2
>>> atlas2 = np.zeros((7, 15), dtype=int)
>>> atlas2[2:5, 7:12] = 1
>>> atlas2[4:7, 7:14] = 2
>>> compare_atlas_adjusted_rand(atlas1, atlas1)
0.0
>>> compare_atlas_adjusted_rand(atlas1, atlas2) 
0.656...
bpdl.metric_similarity.compare_atlas_rnd_pairs(a1, a2, rand_seed=None)[source]

compare two atlases as taking random pixels pairs from both and evaluate that the are labeled equally of differently

Parameters

  • a1 – np.array<height, width>

  • a2 – np.array<height, width>

  • rand_seed – random initialization

Return float

with 0 means no difference

>>> atlas1 = np.zeros((7, 15), dtype=int)
>>> atlas1[1:4, 5:10] = 1
>>> atlas1[5:7, 6:13] = 2
>>> atlas2 = np.zeros((7, 15), dtype=int)
>>> atlas2[2:5, 7:12] = 1
>>> atlas2[4:7, 7:14] = 2
>>> compare_atlas_rnd_pairs(atlas1, atlas1)
0.0
>>> round(compare_atlas_rnd_pairs(atlas1, atlas2, rand_seed=0), 5)
0.37143
bpdl.metric_similarity.compare_matrices(m1, m2)[source]

sum all element differences and divide it by number of elements

Parameters

  • m1 – np.array<height, width>

  • m2 – np.array<height, width>

Return float

>>> seg1 = np.zeros((7, 15), dtype=int)
>>> seg1[1:4, 5:10] = 3
>>> seg1[5:7, 6:13] = 2
>>> seg2 = np.zeros((7, 15), dtype=int)
>>> seg2[2:5, 7:12] = 1
>>> seg2[4:7, 7:14] = 3
>>> compare_matrices(seg1, seg1)
0.0
>>> compare_matrices(seg1, seg2) 
0.819...
bpdl.metric_similarity.compare_weights(c1, c2)[source]
Parameters

  • c1 – np.array<height, width>

  • c2 – np.array<height, width>

Return float

>>> np.random.seed(0)
>>> compare_weights(np.random.randint(0, 2, (10, 5)),
...                 np.random.randint(0, 2, (10, 5)))
0.44
bpdl.metric_similarity.compute_classif_metrics(y_true, y_pred, metric_averages=('macro', 'weighted'))[source]

compute standard metrics for multi-class classification

Parameters
Return {str

float}:

>>> y_true = np.array([0] * 3 + [1] * 5)
>>> y_pred = np.array([0] * 5 + [1] * 3)
>>> dist_sm = compute_classif_metrics(y_true, y_pred)
>>> pair_sm = [(n, dist_sm[n]) for n in sorted(dist_sm.keys())]
>>> pair_sm 
[('ARS', 0.138...),
 ('accuracy', 0.75),
 ('confusion', [[3, 0], [2, 3]]),
 ('f1_macro', 0.800...), ('f1_weighted', 0.849...),
 ('precision_macro', 0.800...), ('precision_weighted', 0.75),
 ('recall_macro', 0.749...), ('recall_weighted', 0.749...),
 ('support_macro', None), ('support_weighted', None)]
>>> y_true = np.array([0] * 5 + [1] * 5 + [2] * 5)
>>> y_pred = np.array([0] * 5 + [1] * 3 + [2] * 7)
>>> dist_sm = compute_classif_metrics(y_true, y_pred)
>>> pair_sm = [(n, dist_sm[n]) for n in sorted(dist_sm.keys())]
>>> pair_sm 
[('ARS', 0.641...),
 ('accuracy', 0.866...),
 ('confusion', [[5, 0, 0], [0, 3, 2], [0, 0, 5]]),
 ('f1_macro', 0.904...), ('f1_weighted', 0.904...),
 ('precision_macro', 0.866...), ('precision_weighted', 0.866...),
 ('recall_macro', 0.861...), ('recall_weighted', 0.861...),
 ('support_macro', None), ('support_weighted', None)]
bpdl.metric_similarity.compute_labels_overlap_matrix(seg1, seg2)[source]

compute overlap between tho segmentation atlasess) with same sizes

Parameters

  • seg1 – np.array<height, width>

  • seg2 – np.array<height, width>

Return ndarray

np.array<height, width>

>>> seg1 = np.zeros((7, 15), dtype=int)
>>> seg1[1:4, 5:10] = 3
>>> seg1[5:7, 6:13] = 2
>>> seg2 = np.zeros((7, 15), dtype=int)
>>> seg2[2:5, 7:12] = 1
>>> seg2[4:7, 7:14] = 3
>>> compute_labels_overlap_matrix(seg1, seg1)
array([[76,  0,  0,  0],
       [ 0,  0,  0,  0],
       [ 0,  0, 14,  0],
       [ 0,  0,  0, 15]])
>>> compute_labels_overlap_matrix(seg1, seg2)
array([[63,  4,  0,  9],
       [ 0,  0,  0,  0],
       [ 2,  0,  0, 12],
       [ 9,  6,  0,  0]])
bpdl.metric_similarity.relabel_max_overlap_merge(seg_ref, seg_relabel, keep_bg=True)[source]

relabel the second segmentation cu that maximise relative overlap for each pattern (object), if one pattern in reference atlas is likely composed from multiple patterns in relabel atlas, it merge them NOTE: it skips background class - 0

Parameters

  • seg_ref (ndarray) – segmentation

  • seg_relabel (ndarray) – segmentation

  • keep_bg (bool) –

Return ndarray

>>> atlas1 = np.zeros((7, 15), dtype=int)
>>> atlas1[1:4, 5:10] = 1
>>> atlas1[5:7, 3:13] = 2
>>> atlas2 = np.zeros((7, 15), dtype=int)
>>> atlas2[0:3, 7:12] = 1
>>> atlas2[3:7, 1:7] = 2
>>> atlas2[4:7, 7:14] = 3
>>> atlas2[:2, :3] = 5
>>> relabel_max_overlap_merge(atlas1, atlas2, keep_bg=True)
array([[1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0],
       [1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0],
       [0, 2, 2, 2, 2, 2, 2, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0],
       [0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0],
       [0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0]])
>>> relabel_max_overlap_merge(atlas2, atlas1, keep_bg=True)
array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0, 0],
       [0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0, 0]])
>>> relabel_max_overlap_merge(atlas1, atlas2, keep_bg=False)
array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 0],
       [0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 0],
       [0, 0, 0, 0, 0, 0, 0, 2, 2, 2, 2, 2, 2, 2, 0]])
bpdl.metric_similarity.relabel_max_overlap_unique(seg_ref, seg_relabel, keep_bg=True)[source]

relabel the second segmentation cu that maximise relative overlap for each pattern (object), the relation among patterns is 1-1 NOTE: it skips background class - 0

Parameters

  • seg_ref (ndarray) – segmentation

  • seg_relabel (ndarray) – segmentation

  • keep_bg (bool) –

Return ndarray

>>> atlas1 = np.zeros((7, 15), dtype=int)
>>> atlas1[1:4, 5:10] = 1
>>> atlas1[5:7, 3:13] = 2
>>> atlas2 = np.zeros((7, 15), dtype=int)
>>> atlas2[0:3, 7:12] = 1
>>> atlas2[3:7, 1:7] = 2
>>> atlas2[4:7, 7:14] = 3
>>> atlas2[:2, :3] = 5
>>> relabel_max_overlap_unique(atlas1, atlas2, keep_bg=True)
array([[5, 5, 5, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0],
       [5, 5, 5, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0],
       [0, 3, 3, 3, 3, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 0],
       [0, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 0],
       [0, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 0]])
>>> relabel_max_overlap_unique(atlas2, atlas1, keep_bg=True)
array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 0, 0],
       [0, 0, 0, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 0, 0]])
>>> relabel_max_overlap_unique(atlas1, atlas2, keep_bg=False)
array([[5, 5, 5, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0],
       [5, 5, 5, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0],
       [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0],
       [0, 3, 3, 3, 3, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0],
       [0, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 0],
       [0, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 0],
       [0, 3, 3, 3, 3, 3, 3, 2, 2, 2, 2, 2, 2, 2, 0]])