def label_unions_and_duplicates(rois, overlap_threshold):
'''Detect unions and duplicates and label ROIs.'''
masks = create_roi_mask_array(rois)
valid_masks = np.ones(masks.shape[0]).astype(bool)
ms = mask_set.MaskSet(masks=masks)
# detect and label duplicates
duplicates = ms.detect_duplicates(overlap_threshold)
for duplicate in duplicates:
index = duplicate[0]
if "duplicate" not in rois[index].labels:
rois[index].labels.append("duplicate")
valid_masks[index] = False
# detect and label unions only for remaining valid masks
valid_idxs = np.where(valid_masks)
ms = mask_set.MaskSet(masks=masks[valid_idxs].astype(bool))
unions = ms.detect_unions()
if unions:
union_idxs = list(unions.keys())
idxs = valid_idxs[0][union_idxs]
for idx in idxs:
if "union" not in rois[idx].labels:
rois[idx].labels.append("union")
return rois
def identify_valid_masks(mask_array):
ms = mask_set.MaskSet(masks=mask_array.astype(bool))
valid_masks = np.ones(mask_array.shape[0]).astype(bool)
# detect duplicates
duplicates = ms.detect_duplicates(overlap_threshold=0.9)
if len(duplicates) > 0:
valid_masks[duplicates.keys()] = False
# detect unions, only for remaining valid masks
valid_idxs = np.where(valid_masks)
ms = mask_set.MaskSet(masks=mask_array[valid_idxs].astype(bool))
unions = ms.detect_unions()
if len(unions) > 0:
un_idxs = unions.keys()
valid_masks[valid_idxs[0][un_idxs]] = False
return valid_masks
def label_unions_and_duplicates(roi_objs, masks=None, duplicate_threshold=0.9,
union_threshold=0.7, max_dist=10, set_size=2):
"""
Modified from allensdk.internal.brain_observatory.roi_filter.py
Returns ROI objects with unions and duplicates labelled.
Required args:
- roi_objs (ROI objects): ROI objects
Optional args:
- masks (3D array) : ROI mask arrays. If None provided, they
are recreated from the ROI objects
default: None
- duplicate_threshold (float): threshold for identifying ROI duplicated
(only the first of each set is labelled
a duplicate)
default: 0.9
- union_threshold (float) : threshold for identifying ROIs that are
unions of several ROIs
default: 0.7
- set_size (int) : number of ROIs forming sets to be checked
for possibly being unions
default: 2
- max_dist (num) : max distance between ROIs to be checked
for possibly being unions
default: 10
Returns:
- roi_objs (ROI objects): ROI objects labelled for union, duplicate,
empty and border overlapping mask conditions
"""
roi_objs = copy.deepcopy(roi_objs)
if masks is None:
masks = roi_masks.create_roi_mask_array(roi_objs)
# get indices for non empty ROIs
non_empty_mask = np.asarray([
roi_obj.mask is not None for roi_obj in roi_objs]).astype(bool)
non_empty_idx = np.where(non_empty_mask)[0]
# label empty ROIs
for idx in np.where(~non_empty_mask)[0]:
roi_objs[idx].labels.append("empty")
ms = mask_set.MaskSet(masks=masks[non_empty_idx])
# detect and label duplicates
duplicates = ms.detect_duplicates(duplicate_threshold)
for duplicate in duplicates:
orig_idx = non_empty_idx[duplicate[0]]
if "duplicate" not in roi_objs[orig_idx].labels:
roi_objs[orig_idx].labels.append("duplicate")
# detect and label unions
unions = ms.detect_unions(set_size, max_dist, union_threshold)
if unions:
union_idxs = list(unions.keys())
for idx in union_idxs:
orig_idx = non_empty_idx[idx]
if "union" not in roi_objs[orig_idx].labels:
roi_objs[orig_idx].labels.append("union")
return roi_objs