# US 12,217,488 B2 · Learning image generation device, learning image generation method, learning image generation program, learning method, learning device, and learning program > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,217,488 B2 - Document ID: US12217488B2 - Status: Granted - Kind: B2 - Application number: 17/562,988 - Filed: 2021-12-27 - Published: 2022-04-21 - Granted: 2025-02-04 - Earliest claimed: 2019-06-28 - Assignees: FUJIFILM Corporation - Applicants: FUJIFILM Corporation - Inventors: Sadato Akahori; Takuya Fuchigami - Web page: https://patentgrep.com/patent/US12217488B2 ## CPC - G06V 10/7747: {Organisation of the process, e.g. bagging or boosting} - G06T 7/0012: {Biomedical image inspection} - G06T 7/11: Region-based segmentation - G06V 10/7792: {the supervisor being an automated module, e.g. "intelligent oracle"} - G06T 2207/10081: Computed x-ray tomography [CT] - G06T 2207/10088: Magnetic resonance imaging [MRI] - G06T 2207/10104: Positron emission tomography [PET] - G06T 2207/20081: Training; Learning - G06T 2207/20084: Artificial neural networks [ANN] - G06T 2207/30016: Brain - G06V 2201/03: Recognition of patterns in medical or anatomical images - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL ## Abstract A learning image generation device including: a supervised data acquisition unit that acquires supervised data including a learning image and a correct learning image in which a correct region is defined in the learning image as a pair; and a variation learning image generation unit that generates a variation learning image in which a pixel value of a pixel belonging to the correct region is varied within a limitation range of an allowable pixel value of the pixel belonging to the correct region in the learning image. ## Claims ### Claim 1 A learning image generation device comprising: a memory; and a processor coupled to the memory, the processor configured to: acquire supervised data including a learning image and a correct region defined in the learning image as a pair; acquire a limitation range of allowable pixel values for a first designated pixel belonging to the correct region in the learning image; generate a plurality of variable pixel values for the first designated pixel belonging to the correct region in the learning image according to the limitation range; and generate a plurality of variation learning images, wherein a pixel value of a second designated pixel in each of the plurality of variation learning images is respectively a different pixel value among the plurality of variation pixel values, and wherein the first designated pixel in the learning image and the second designated pixel in each of the plurality of variation learning images have same pixel coordinates. ### Claim 2 (depends on 1) The learning image generation device according to claim 1, wherein the processor is configured to generate the plurality of variable pixel values in response to a normal distribution set for the correct region. ### Claim 3 A learning image generation method comprising: acquiring supervised data including a learning image and a correct region defined in the learning image as a pair; acquiring a limitation range of allowable pixel values for a first designated pixel belonging to the correct region in the learning image; generating a plurality of variable pixel values for the first designated pixel belonging to the correct region in the learning image according to the limitation range; and generating a plurality of variation learning images, wherein a pixel value of a second designated pixel in each of the plurality of variation learning images is respectively a different pixel value among the plurality of variation pixel values, and wherein the first designated pixel in the learning image and the second designated pixel in each of the plurality of variation learning images have same pixel coordinates. ### Claim 4 A non-transitory computer recording medium storing a learning image generation program causing a computer to: acquire supervised data including a learning image and a correct region defined in the learning image as a pair; acquire a limitation range of allowable pixel values for a first designated pixel belonging to the correct region in the learning image; generate a plurality of variable pixel values for the first designated pixel belonging to the correct region in the learning image according to the limitation range; and generate a variation learning images, wherein a pixel value of a second designated pixel in each of the plurality of variation learning images is respectively a different pixel value among the plurality of variation pixel values, and wherein the first designated pixel in the learning image and the second designated pixel in each of the plurality of variation learning images have same pixel coordinates. ### Claim 5 A learning method comprising: learning a model by using one or more first supervised data including a learning image and a correct region defined in the learning image as a pair, and one or more second supervised data including a plurality of variation learning images and a correct region defined in the learning image as a pair, wherein the plurality of variation learning images are generated by acquiring a limitation range of allowable pixel values for a first designated pixel belonging to the correct region in the learning image and generating a plurality of variable pixel values for the first designated pixel in the learning image according to the limitation range, wherein a value of a second designated pixel in each of the plurality of variation learning images is respectively a different pixel value among the plurality of variation pixel values, and wherein the first designated pixel in the learning image and the second designated pixel in each of the plurality of variation learning images have same pixel coordinates. ### Claim 6 (depends on 5) The learning method according to claim 5, wherein: the model is learned by using a plurality of the first supervised data in first learning, and the model is learned by replacing at least one first supervised data among the plurality of first supervised data with the second supervised data in second and subsequent learning. ### Claim 7 (depends on 5) The learning method according to claim 5, wherein the model is learned by using a plurality of the first supervised data in first learning, and the model is learned by adding at least one second supervised data in second and subsequent learning. ### Claim 8 (depends on 6) The learning method according to claim 6, wherein at least one of the second supervised data to be used or the number of the second supervised data is randomly set for each learning in the second and subsequent learning. ### Claim 9 (depends on 6) The learning method according to claim 6, wherein at least one of the second supervised data to be used or the number of the second supervised data is set in advance in the second and subsequent learning. ### Claim 10 (depends on 6) The learning method according to claim 6, wherein the model is learned by using only the plurality of first supervised data at least once in the second and subsequent learning. ### Claim 11 A learning device comprising: a memory; and a processor coupled to the memory, the processor configured to: acquire one or more first supervised data including a learning image and a correct region defined in the learning image as a pair; acquire one or more second supervised data including a plurality of variation learning images and a correct region defined in the learning image as a pair, wherein the plurality of variation learning images are generated by acquiring a limitation range of allowable pixel values for a first designated pixel belonging to the correct region in the learning image and generating a plurality of variable pixel values for the first designated pixel in the learning image according to the limitation range, wherein a value of a second designated pixel in each of the plurality of variation learning images is respectively a different pixel value among the plurality of variation pixel values, and wherein the first designated pixel in the learning image and the second designated pixel in each of the plurality of variation learning images have same pixel coordinates; and learn a model by using the acquired one or more first supervised data and the acquired one or more second supervised data. ### Claim 12 A learning device comprising: a memory; and a processor coupled to the memory, the processor configured to: acquire one or more first supervised data including a learning image and a correct region defined in the learning image as a pair; acquire one or more second supervised data including a plurality of variation learning images and a correct region defined in the learning image as a pair, wherein the plurality of variation learning images are generated by acquiring a limitation range of allowable pixel values for a first designated pixel belonging to the correct region in the learning image and generating a plurality of variable pixel values for the first designated pixel in the learning image according to the limitation range, wherein a value of a second designated pixel in each of the plurality of variation learning images is respectively a different pixel value among the plurality of variation pixel values, and wherein the first designated pixel in the learning image and the second designated pixel in each of the plurality of variation learning images have same pixel coordinates; and learn a model by using the acquired one or more first supervised data and the acquired one or more second supervised data, wherein the model is learned by using a plurality of the first supervised data in the first learning, and the model is learned by replacing at least one first supervised data among the plurality of first supervised data with the second supervised data in second and subsequent learning. ### Claim 13 A non-transitory computer recording medium storing a learning program causing a computer to: acquire one or more first supervised data including a learning image and a correct region defined in the learning image as a pair; acquire one or more second supervised data including a plurality of variation learning images and a correct region defined in the learning image as a pair, wherein the plurality of variation learning images are generated by acquiring a limitation range of allowable pixel values for a first designated pixel belonging to the correct region in the learning image and generating a plurality of variable pixel values for the first designated pixel in the learning image according to the limitation range, wherein a value of a second designated pixel in each of the plurality of variation learning images is respectively a different pixel value among the plurality of variation pixel values, and wherein the first designated pixel in the learning image and the second designated pixel in each of the plurality of variation learning images have same pixel coordinates; and learn a model by using the one or more first supervised data and the one or more second supervised data acquired by the supervised data acquisition unit. ### Claim 14 (depends on 1) The learning device according to claim 1, wherein an upper limit of the limitation range for generating the plurality of variable pixel values in the plurality of variation learning images is greater than an upper limit of a range of pixel values in the correct region in the learning image, and wherein a lower limit of the limitation range for generating the plurality of variable pixel values in the plurality of variation learning images is less than a lower limit of a range of pixel values in the correct region in the learning image. ### Claim 15 (depends on 1) The learning device according to claim 1, wherein the learning image is a medical image, and wherein an upper limit and a lower limit of the limitation range is set based on a type of the medical image and a type of lesion to be detected.