US 12,471,851 B2
Method and system for providing cancer diagnosis information using artificial intelligence-based liquid biopsy of exosome
Yeon Ho Choi, Seoul (KR); Hyun Ku Shin, Seoul (KR); Jae Na Park, Seongnam-si (KR); and Soon Woo Hong, Seoul (KR)
Assigned to Exopert Corporation, Seoul (KR)
Appl. No. 17/311,718
Filed by Exopert Corporation, Seoul (KR)
PCT Filed Jan. 13, 2020, PCT No. PCT/KR2020/000568
§ 371(c)(1), (2) Date Jun. 8, 2021,
PCT Pub. No. WO2020/180003, PCT Pub. Date Sep. 10, 2020.
Claims priority of application No. 10-2019-0024870 (KR), filed on Mar. 4, 2019.
Prior Publication US 2022/0022816 A1, Jan. 27, 2022
Int. Cl. G01N 33/48 (2006.01); A61B 5/00 (2006.01); A61B 5/1455 (2006.01); G01N 21/65 (2006.01); G06N 3/08 (2023.01); G16H 50/20 (2018.01)
CPC A61B 5/7264 (2013.01) [A61B 5/0059 (2013.01); A61B 5/1455 (2013.01); G01N 21/658 (2013.01); G06N 3/08 (2013.01); G16H 50/20 (2018.01)] 6 Claims
 
1. A process-implemented method for providing cancer diagnosis information using artificial intelligence-based liquid biopsy of exosome, the method comprising:
measuring, by a Surface Enhanced Raman Spectroscopy (SERS) measurement device, a cultured cell exosome SERS signal from cultured cancer cells and cultured normal cells;
training, by a processor, a deep learning model using the cultured cell exosome SERS signal, the deep learning model being configured to distinguish between cancer cell exosome signals and normal cell exosome signals, wherein the deep learning model comprises five convolution layers, including:
a first convolutional layer configured to receive input data;
a second to fifth convolutional layer, each comprising a plurality of sub-convolutional layers, wherein skip connections are inserted between every two of the sub-convolutional layers; and
a fully connected layer connected sequentially after the fifth convolutional layer;
measuring, by the SERS measurement device, a blood exosome SERS signal from a subject's blood sample;
analyzing, by the processor, the blood exosome SERS signal using the deep learning model trained with the cultured cell exosome SERS signal to generate feature data, wherein the analyzing comprises processing the blood exosome SERS signal through the five convolutional layers and a first fully connected layer; and
analyzing, by the processor, a similarity between blood exosome feature data and cultured cell exosome feature data analyzed using the deep learning model to generate cancer diagnosis information,
wherein the similarity analysis comprises calculating a Mahalanobis distance, where Dnormal is a Mahalanobis distance from the blood exosome data to normal cell exosome data, and Dcancer is a Mahalanobis distance from the blood exosome data to cancer cell exosome data.