US 12,243,287 B2
Method for configuring an image evaluation device and also image evaluation method and image evaluation device
Markus Michael Geipel, Munich (DE); Florian Büttner, Munich (DE); Christoph Tietz, Ottobrunn (DE); Gaby Marquardt, Hausen (DE); and Daniela Seidel, Baiersdorf (DE)
Assigned to Siemens Healthcare Diagnostics Inc., Tarrytown, NY (US)
Appl. No. 17/280,863
Filed by Siemens Healthcare Diagnostics Inc., Tarrytown, NY (US)
PCT Filed Sep. 16, 2019, PCT No. PCT/IB2019/057772
§ 371(c)(1), (2) Date Mar. 26, 2021,
PCT Pub. No. WO2020/065436, PCT Pub. Date Apr. 2, 2020.
Claims priority of application No. 18197650 (EP), filed on Sep. 28, 2018.
Prior Publication US 2022/0012531 A1, Jan. 13, 2022
Int. Cl. G06F 18/214 (2023.01); G06F 18/22 (2023.01); G06F 18/2415 (2023.01); G06N 3/045 (2023.01); G06V 10/764 (2022.01); G06V 10/82 (2022.01); G06V 20/69 (2022.01)
CPC G06V 10/764 (2022.01) [G06F 18/214 (2023.01); G06F 18/22 (2023.01); G06F 18/2415 (2023.01); G06N 3/045 (2023.01); G06V 10/82 (2022.01); G06V 20/698 (2022.01); G06V 2201/03 (2022.01)] 12 Claims
OG exemplary drawing
 
1. A method for configuring an image evaluation device for determining an object type and an object sub-type of an imaged object, comprising:
inputting a multiplicity of training images respectively assigned to an object type and an object sub-type into a first neural network module for recognizing image features,
inputting training output data sets of the first neural network module into a second neural network module for recognizing object types based on recognized image features, wherein:
the first and second neural network modules are jointly trained such that training output data sets of the second neural network module at least approximately reproduce the object types assigned to the training images, and
for a respective object type:
inputting training images assigned to the respective object type into a trained first neural network module, the trained first neural network module only receiving training images for a respective object type,
generating a training output data set of the trained first neural network module for a respective training image of the respective object type,
assigning the object sub-type of the respective training image to the training output data set of the trained first neural network module, and
inputting the training output data set of the trained first neural network module into a dedicated sub-type recognition module for only the respective object type for recognizing object sub-types of the respective object type on the basis of image features recognized by the trained first neural network module.