Skip to content

Latest commit

 

History

History
68 lines (51 loc) · 4.33 KB

wsi_registration.md

File metadata and controls

68 lines (51 loc) · 4.33 KB

WSI registration (Alignment)

WSI registration is necessary when we do pairwise WSIs analysis. Tissue correspondence can be established by aligning two WSIs. According to how we get the WSI pairs, two scenarios can be summarized:

  1. scan the slide and then treat the slide with other chemicals and scan it again
  2. scan a serials of slides on which are the continuous cuts of tissue.

In current version, we just consider the first scenario. This is a rigid image registration problem, just shifting, even without rotation. Because the scanners are designed to maximizing stability of slide loading and mounting, and only x and y directions have movements (z direction for focusing). Piratically the rotation angle is small enough to be ignored.

To standardize the offset saving and loading, we introduced a csv file to maintain the WSI pairs and their offsets.

We provide two ways to get the shifting offset:

1. Automatic registration

Here is the an example of how to automatically align two WSIs and save/load the offset to/from the csv file. It's a easy-to-use version of our previous work.

  • Note that rotation is not taken into account in current version of this automatic registration.
  • In most WSI scanner, the way of slides mounting determined that there should be minimum rotation, practically almost zero.
  • So if the slide is scanned and re-scanned, the obtained two WSIs should be no rotation.
  • But if the WSIs are not obtained in scan and re-scan way, the rotation shouldn't be ignored, and you can not use this automatic registration.

You need to define a tissue detector to identify the foreground of a WSI, from which image patches will be extracted, and shifting offset will be calculated based on these patches. Additionally, you may also need a WSI_CaseManager to help you to find the WSI counterpart, which maintains the correspondence in a MS Excel file.

from wsitools.file_management.wsi_case_manager import WSI_CaseManager   # import dependent packages
from wsitools.file_management.offset_csv_manager import OffsetCSVManager
from wsitools.tissue_detection.tissue_detector import TissueDetector
from wsitools.wsi_registration.auto_wsi_matcher import MatcherParameters, WSI_Matcher
fixed_wsi = "/projects/MELF/7bb50b5d9dcf4e53ad311d66136ae00f.tiff"
#float_wsi = "/projects/MELF-Clean/8a26a55a78b947059da4e8c36709a828.tiff"
float_wsi_root_dir = "/projects/MELF-Clean/"

case_mn = WSI_CaseManager()
float_wsi = case_mn.get_counterpart_fn(fixed_wsi, float_wsi_root_dir)
_, fixed_wsi_uuid, _ = case_mn.get_wsi_fn_info(fixed_wsi)
_, float_wsi_uuid, _ = case_mn.get_wsi_fn_info(float_wsi)

offset_csv_fn = "../file_management/example/registration_offsets.csv"
offset_csv_mn = OffsetCSVManager(offset_csv_fn)
offset_tmp, state_indicator = offset_csv_mn.lookup_table(fixed_wsi_uuid, float_wsi_uuid)
if state_indicator == 0 or staticmethod == 1:     # Auto registration does not exist
    tissue_detector = TissueDetector("GNB", threshold=0.5)
    matcher_parameters = MatcherParameters()
    matcher = WSI_Matcher(tissue_detector, matcher_parameters)
    offset = matcher.match(fixed_wsi, float_wsi)
    offset_csv_mn.update_auto_registration(fixed_wsi_uuid, float_wsi_uuid, offset)
    print("Automatic registration does not exist in file: %s." % offset_csv_fn)
    print("Add automatic registration result: (%.2f %.2f) to this file" % (offset[0], offset[1]))
else:
    print("Automatic registration is already in the csv file: %s" % offset_csv_fn)
    print("Quarried registration result: (%.2f %.2f)" % (offset_tmp[0], offset_tmp[1]))  # could be ground truth

if state_indicator == 0 or state_indicator == 2:
    print("Ground truth does not exist in file: %s. Need to update ground truth" % offset_csv_fn)
    print("Get ground truth from 'Annotation' and call OffsetCSVManager.update_ground_truth(...)")
else:
    print("Looking up ground truth offset from %s" % offset_csv_fn)
    print("Found ground truth offset: (%.2f %.2f)" % (offset_tmp[0], offset_tmp[1]))

2. QuPath annotation

We provide this Annotate with QuPath and Export Annotations document shows how to annotate with QuPath, export the annotation and calculate offset with python.