US 12,169,926 B2
System and method for disease diagnosis using neural network
Joon Young Cho, Seoul (KR); and Sun Woo Kim, Seongnam-si (KR)
Assigned to DEEP BIO INC., Seoul (KR)
Appl. No. 17/266,090
Filed by DEEP BIO INC., Seoul (KR)
PCT Filed Aug. 7, 2019, PCT No. PCT/KR2019/009844
§ 371(c)(1), (2) Date Feb. 4, 2021,
PCT Pub. No. WO2020/032559, PCT Pub. Date Feb. 13, 2020.
Claims priority of application No. 10-2018-0092011 (KR), filed on Aug. 7, 2018.
Prior Publication US 2021/0304405 A1, Sep. 30, 2021
Int. Cl. G06T 7/00 (2017.01); A61B 5/00 (2006.01); G06N 3/045 (2023.01); G16H 30/40 (2018.01); G16H 50/20 (2018.01); G16H 50/30 (2018.01)
CPC G06T 7/0012 (2013.01) [A61B 5/7264 (2013.01); G06N 3/045 (2023.01); G16H 30/40 (2018.01); G16H 50/20 (2018.01); G16H 50/30 (2018.01)] 17 Claims
OG exemplary drawing
 
1. A disease diagnosis system implemented in a system including a processor and a storage device for storing a neural network, and using a biometric image obtained from a biological sample provided on a slide, and the neural network, the diagnosis system comprising:
the neural network configured to generate a patch-level diagnosis result indicating whether a disease exists in each of predetermined patches generated by dividing the biometric image obtained from the slide in a predetermined size, the biometric image being divided and processed by the neural network: a) receiving the biometric image that includes three channels of RGB color data values for each of a plurality of pixels, and b) receiving, on a fourth channel, an input value for each pixel included in the plurality of patches, wherein the fourth channel comprises a gray data value of each pixel in the plurality of patches;
a heatmap generation module configured to generate a patch-level heatmap image corresponding to the slide on the basis of the patch-level diagnosis result of each of the plurality of patches included in the slide;
a tissue mask generation module configured to generate a tissue mask image corresponding to the biometric image obtained from the slide on the basis of a hue-saturation-value (HSV) model corresponding to the biometric image obtained from the slide; and a visualization module configured to generate a disease diagnosis visualization image corresponding to the biometric image obtained from the slide on the basis of the patch-level heatmap image and the tissue mask image,
wherein the neural network comprises:
a micro neural network using training data for a specific tile; and
a macro neural network using training data for a micro tile including the specific tile and at least one adjacent tile of the specific tile.