US 12,488,603 B2
Systems and methods for automatically identifying features of a cytology specimen
Hamid Reza Tizhoosh, Waterloo (CA); Seyed Rohollah Moosavitayebi, Oshawa (CA); and Clinton James Vaughan Campbell, Oakville (CA)
Assigned to HAMID REZA TIZHOOSH, Rochester, MN (US); and CLINTON JAMES VAUGHAN CAMPBELL, Oakville (CA)
Filed by Hamid Reza Tizhoosh, Rochester, MN (US); and Clinton James Vaughan Campbell, Oakville (CA)
Filed on Apr. 14, 2022, as Appl. No. 17/720,678.
Claims priority of provisional application 63/175,819, filed on Apr. 16, 2021.
Prior Publication US 2022/0335736 A1, Oct. 20, 2022
Int. Cl. G06K 9/00 (2022.01); G06T 7/00 (2017.01); G06V 10/25 (2022.01); G06V 10/74 (2022.01); G06V 10/82 (2022.01); G06V 20/69 (2022.01)
CPC G06V 20/69 (2022.01) [G06T 7/0012 (2013.01); G06V 10/25 (2022.01); G06V 10/761 (2022.01); G06V 10/82 (2022.01); G06T 2207/20084 (2013.01); G06T 2207/30024 (2013.01)] 22 Claims
OG exemplary drawing
 
1. A method of improving a feature identification process for a cytology specimen, the method comprising:
dividing a digital whole slide image of the cytology specimen into a plurality of image portions;
evaluating each image portion of the plurality of image portions with a feature detection neural network trained to identify one or more relevant image portions from the plurality of image portions that contain a region of interest for the cytology specimen, whereby reducing a number of image portions from being further processed when absent the region of interest for the cytology specimen;
automatically selecting a base set of relevant image portions from the one or more relevant image portions, the base set of relevant image portions containing an initial set of relevant image portions for identifying features for the cytology specimen;
evaluating the base set of relevant image portions with a feature identification neural network trained to generate a base cell data comprising a predicted feature type for each feature identified within that image portion and an accuracy likelihood associated with the predicted feature type;
evaluating one or more relevant image portions outside the base set of relevant image portions with the feature identification neural network to generate a supplemental cell data comprising the predicted feature type for each feature within the one or more relevant image portions and the associated accuracy likelihood for the predicted feature type;
determining whether the supplemental cell data and the base cell data satisfy a similarity threshold indicative that the base cell data is sufficiently representative of features present in the cytology specimen;
in response to determining the similarity threshold is not satisfied:
updating the base set to include the supplemental cell data, and generating the updated base cell data with the updated base set with the feature identification neural network; and
continuing to apply the feature identification neural network to another one or more relevant image portions to generate the supplemental cell data, and determining whether the supplemental cell data and the updated base cell data satisfy the similarity threshold until the similarity threshold is satisfied; and
otherwise, providing the feature types portion identified for the cytology specimen.