# US 12,112,558 B2 · Learning model generation method, identification method, learning model generation system, identification system, and non-transitory computer-readable storage medium > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,112,558 B2 - Document ID: US12112558B2 - Status: Granted - Kind: B2 - Application number: 17/445,773 - Filed: 2021-08-24 - Published: 2022-03-03 - Granted: 2024-10-08 - Earliest claimed: 2020-08-28 - Assignees: Hamamatsu Photonics K.K. - Applicants: Hamamatsu Photonics K.K. - Inventors: Kibo Ote; Fumio Hashimoto; Osamu Yasuhiko; Hidenao Yamada - Web page: https://patentgrep.com/patent/US12112558B2 ## CPC - G06V 20/695: {Preprocessing, e.g. image segmentation} - G06V 20/698: {Matching; Classification} - G06N 20/00: Machine learning - G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract The method includes an acquisition step for model generation of acquiring a hologram for model generation in which an image of a cell is captured and which is used to generate a learning model, a generation step for model generation of generating a new hologram for model generation having a different focal position from the acquired hologram for model generation and increasing the number of holograms for model generation, and a learning model generation step of generating a learning model by performing machine learning using the increased holograms for model generation. ## Claims ### Claim 1 A learning model generation method of generating a learning model for identifying a cell on the basis of a captured image of the cell, the method comprising: acquiring a hologram for model generation in which an image of the cell is captured, the hologram for model generation being used to generate the learning model; generating a new hologram for model generation having a different focal position from the acquired hologram for model generation and increasing a number of holograms for model generation; and generating the learning model by performing machine learning using at least the acquired hologram and the new hologram as inputs for model generation, wherein the learning model is operative to output a type of cell or cell classification. ### Claim 2 (depends on 1) The learning model generation method according to claim 1, wherein the method includes increasing the number of holograms for model generation by adding or removing an aberration other than a change in a focal position to or from the hologram for model generation. ### Claim 3 (depends on 1) The learning model generation method according to claim 1, wherein the method includes generating the learning model by performing machine learning using a plurality of increased holograms for model generation collectively as one input. ### Claim 4 (depends on 1) The learning model generation method according to claim 1, wherein the type of cell or cell classification includes whether the cell is a cancer cell or a non-cancer cell. ### Claim 5 (depends on 4) The learning model generation method according to claim 4, wherein the type of cell or cell classification further includes a cancer type corresponding to the cancer cell. ### Claim 6 An identification method of identifying a cell that is a target for identification on the basis of a captured image of the cell, the method comprising: acquiring a hologram for identification in which an image of the cell that is a target for identification is captured; generating a new hologram for identification having a different focal position from the acquired hologram for identification and increasing a number of holograms for identification; and identifying a type of cell or cell classification for the cell using a learning model on the basis of at least the acquired hologram and the new hologram. ### Claim 7 (depends on 6) The identification method according to claim 6, wherein the method includes increasing the number of holograms for identification by adding or removing an aberration other than a change in a focal position to or from the hologram for identification. ### Claim 8 (depends on 6) The identification method according to claim 6, wherein the method includes identifying a type of cell or cell classification for the cell using the learning model generated by performing machine learning using at least the acquired hologram and the new hologram as inputs for model generation, wherein the learning model is operative to output the type of cell or cell classification. ### Claim 9 (depends on 6) The identification method according to claim 6, wherein the method includes identifying a type of cell or cell classification for the cell for each increased hologram that is used as an input for identification, and finally identifying the type of cell or cell classification for the cell on the basis of an identification result for each hologram for identification. ### Claim 10 (depends on 6) The identification method according to claim 6, wherein the method includes identifying a type of cell or cell classification for the cell using a plurality of increased holograms for identification collectively as one input. ### Claim 11 (depends on 6) The identification method according to claim 6, wherein the type of cell or cell classification includes whether the cell is a cancer cell or a non-cancer cell. ### Claim 12 (depends on 11) The identification method according to claim 11, wherein the type of cell or cell classification further includes a cancer type corresponding to the cancer cell. ### Claim 13 A learning model generation system configured to generate a learning model for identifying a cell on the basis of a captured image of the cell, the system comprising circuitry configured to: acquire a hologram for model generation in which an image of the cell is captured, the hologram for model generation being used to generate the learning model; generate a new hologram for model generation having a different focal position from the acquired hologram for model generation and increase a number of holograms for model generation; and generate the learning model by performing machine learning using at least the acquired hologram and the new hologram as inputs for model generation, wherein the learning model is operative to output a type of cell or cell classification. ### Claim 14 (depends on 13) The learning model generation system according to claim 13, wherein the type of cell or cell classification includes whether the cell is a cancer cell or a non-cancer cell. ### Claim 15 (depends on 14) The learning model generation system according to claim 14, wherein the type of cell or cell classification further includes a cancer type corresponding to the cancer cell. ### Claim 16 An identification system configured to identify a cell that is a target for identification on the basis of a captured image of the cell, the system comprising circuitry configured to: acquire a hologram for identification in which an image of the cell that is a target for identification is captured; generate a new hologram for identification having a different focal position from the acquired hologram for identification and increase a number of holograms for identification; and identify a type of cell or cell classification for the cell using a learning model on the basis of at least the acquired hologram and the new hologram. ### Claim 17 (depends on 16) The identification system according to claim 16, wherein the type of cell or cell classification includes whether the cell is a cancer cell or a non-cancer cell. ### Claim 18 (depends on 17) The identification system according to claim 17, wherein the type of cell or cell classification further includes a cancer type corresponding to the cancer cell. ### Claim 19 A non-transitory computer-readable storage medium storing a learning model generation program causing a computer to operate as a learning model generation system configured to generate a learning model for identifying a cell on the basis of a captured image of the cell, the program causing the computer to: acquire a hologram for model generation in which an image of the cell is captured, the hologram for model generation being used to generate the learning model; generate a new hologram for model generation having a different focal position from the acquired hologram for model generation and increase a number of holograms for model generation; and generate the learning model by performing machine learning using at least the acquired hologram and the new hologram as inputs for model generation, wherein the learning model is operative to output a type of cell or cell classification. ### Claim 20 A non-transitory computer-readable storage medium storing an identification program causing a computer to operate as identification system configured to identify a cell that is a target for identification on the basis of a captured image of the cell, the program causing the computer to: acquire a hologram for identification in which an image of the cell that is a target for identification is captured; generate a new hologram for identification having a different focal position from the acquired hologram for identification and increase a number of holograms for identification; and identify a type of cell or cell classification for the cell using a learning model on the basis of at least the acquired hologram and the new hologram.