US 12,105,174 B2
Technique for determining a cardiac metric from CMR images
Puneet Sharma, Princeton Junction, NJ (US); and Lucian Mihai Itu, Brasov (RO)
Assigned to Siemens Healthineers AG, Forchheim (DE)
Filed by SIEMENS HEALTHINEERS AG, Forchheim (DE)
Filed on Aug. 27, 2021, as Appl. No. 17/446,223.
Claims priority of application No. 20465558 (EP), filed on Sep. 17, 2020.
Prior Publication US 2022/0082647 A1, Mar. 17, 2022
Int. Cl. G01R 33/563 (2006.01); G01R 33/567 (2006.01); G06N 3/08 (2023.01)
CPC G01R 33/56366 (2013.01) [G01R 33/567 (2013.01); G06N 3/08 (2013.01)] 24 Claims
OG exemplary drawing
 
1. A neural network system for determining at least two cardiac metrics from cardiac magnetic resonance (CMR) images, the neural network system comprising:
an input layer configured to receive at least one CMR image representative of a rest perfusion state and at least one CMR image representative of a stress perfusion state; and
an output layer configured to output at least two different cardiac metrics based on the at least one CMR image representative of the rest perfusion state and the at least one CMR image representative of the stress perfusion state,
wherein the neural network system with interconnections between the input layer and the output layer is trained by multi-task learning using a plurality of datasets, each of the datasets comprising an instance of the at least one CMR image representative of the rest perfusion state and the at least one CMR image representative of the stress perfusion state for the input layer and the at least two different cardiac metrics for the output layer,
wherein the neural network system comprises at least two sub-networks corresponding to the at least two different cardiac metrics, and
wherein the interconnections comprise cross-connections between the at least two sub-networks at the input layer and/or at least one hidden layer between the input layer and the output layer.