US 11,948,297 B1
Racially unbiased deep learning-based mammogram analyzer
Timothy Cogan, McKinney, TX (US); Richard Stubblefield, Diablo, CA (US); and Lakshman Tamil, Plano, TX (US)
Assigned to MedCognetics, Inc., Plano, TX (US)
Filed by MedCognetics, Inc., Plano, TX (US)
Filed on Jul. 15, 2021, as Appl. No. 17/305,864.
Claims priority of provisional application 63/052,411, filed on Jul. 15, 2020.
Int. Cl. G06T 7/00 (2017.01); A61B 6/50 (2024.01); G06N 3/04 (2023.01); G06N 3/08 (2023.01); G16H 30/20 (2018.01)
CPC G06T 7/0012 (2013.01) [A61B 6/502 (2013.01); G06N 3/04 (2013.01); G06N 3/08 (2013.01); G16H 30/20 (2018.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30068 (2013.01)] 36 Claims
OG exemplary drawing
 
1. A system comprising:
a convolutional neural network comprising a plurality of layers arranged in a hierarchy, wherein an input of the convolutional neural network is at a bottom of the hierarchy of layers and an output is at a top of the hierarchy of layers;
a first plurality of layers of the convolutional neural network which is trained by federated learning using a first set of training data comprising first breast images comprising multiple ethnicities;
a second layer, trained by active learning using a second set of training data comprising second breast images from only a different single ethnicity, a first ethnicity; and
a third layer, trained by active learning using a third set of training data comprising third images from only a single ethnicity, a second ethnicity that is different from the first ethnicity,
wherein the second and third layers are at the same level in the hierarchy of the convolutional neural network, and the second and third layers are above the first layers of the convolutional neural network,
the convolutional neural network comprising the first layers, second layer, and third layer comprises a machine learning model,
a breast image for analysis is input to the input of the convolutional neural network,
the output of the convolutional neural network provides a diagnosis probability of breast cancer,
a fourth layer and fifth layer are within the first layers of the convolutional neural network, and the fifth layer is above the fourth layer,
the breast image for analysis is input to the fourth layer, the fourth layer extracts first different features in the breast image, the fifth layer stores first weightings for the first different features in the breast image extracted by the fourth layer,
the first weightings for the first different features are input to the second and third layers,
given the first weightings from the fifth layer, the second layer outputs second weightings for second different features, and
given the first weightings from the fifth layer, the third layer outputs third weightings for third different features.