US 12,106,510 B2
Method and system for detecting landmarks in medical images
Parmeet Bhatia, Paoli, PA (US); Yimo Guo, Chester Springs, PA (US); Gerardo Hermosillo Valadez, West Chester, PA (US); Zhigang Peng, Ambler, PA (US); and Yu Zhao, Athens, GA (US)
Assigned to SIEMENS HEALTHINEERS AG, Forchheim (DE)
Filed by Siemens Healthcare GmbH, Erlangen (DE)
Filed on Mar. 3, 2021, as Appl. No. 17/190,674.
Claims priority of application No. 20162636 (EP), filed on Mar. 12, 2020.
Prior Publication US 2021/0287363 A1, Sep. 16, 2021
Int. Cl. G06T 7/00 (2017.01); A61B 90/00 (2016.01); G06T 3/60 (2006.01); G06T 7/70 (2017.01); G06T 7/73 (2017.01); G06V 10/22 (2022.01); G06V 10/40 (2022.01); G06V 10/82 (2022.01); G06V 30/24 (2022.01)
CPC G06T 7/73 (2017.01) [A61B 90/36 (2016.02); G06T 3/60 (2013.01); G06T 7/0012 (2013.01); G06T 7/70 (2017.01); G06V 10/22 (2022.01); G06V 10/40 (2022.01); G06V 10/82 (2022.01); G06V 30/2504 (2022.01); A61B 2090/363 (2016.02); G06T 2207/30004 (2013.01)] 21 Claims
OG exemplary drawing
 
1. A computer-implemented method for detecting one or more anatomic landmarks in medical image data, comprising:
receiving medical image data depicting a body part of a patient;
determining a first set of anatomic landmarks from a first representation of the medical image data at a first resolution by applying a first trained function to the first representation of the medical image data; and
determining a second set of anatomic landmarks from a second representation of the medical image data at a second resolution by applying a second trained function to the second representation of the medical image data, the second trained function using the first set of anatomic landmarks, the second resolution being higher than the first resolution, and the second trained function being different than the first trained function, wherein
the first trained function is configured to define one or more first subspaces in the first representation and specify sequences of actions based on a learned policy to reposition the one or more first sub-spaces in the first representation to parse the first representation to determine the first set of anatomic landmarks in one or more iterations of repositioning the one or more first sub-spaces,
the second trained function is configured to define one or more second sub-spaces in the second representation based on landmark locations of the first set of anatomic landmarks and specify sequences of actions based on a learned policy to reposition the one or more second sub-spaces in the second representation to parse the second representation to determine the second set of anatomic landmarks in one or more iterations of repositioning the one or more second sub-spaces,
an action of the sequences of actions based on the learned policy of the first trained function includes changing a position of the one or more first subspaces by at least one pixel,
an action of the sequences of actions based on the learned policy of the second trained function includes changing a position of the one or more second sub-spaces by at least one pixel,
the first resolution is selected from a set of one or more resolutions by a third trained function based on an intrinsic resolution of the medical image data,
the first resolution is lower than the intrinsic resolution of the medical image data, and
the first trained function corresponds to the first resolution and is selected from a set of trained functions corresponding to each resolution of the set of resolutions.