Method and device for cell discrimination using artificial intelligence

Abstract

The present invention relates to a cell discrimination method and device using artificial intelligence, which can distinguish and determine the inherent characteristics of cells by learning the minute morphology of cells that changes in the early stages when various types of cells are cultured in various types of culture media and observing only cell images. The cell discrimination method using artificial intelligence according to the present invention includes an input step of inputting a cell image; and a discrimination step of discriminating which of various cell types, various culture conditions, and culture times the cell image corresponds to by using a discrimination model based on deep learning.

Description

[Technical field]

[0001] The present invention relates to a cell discrimination method and device using artificial intelligence, and more particularly to a cell analysis method and device using artificial intelligence that can learn the microscopic morphology of cells that changes in the early stages when various types of cells are cultured in various types of culture media, and distinguish and determine the unique characteristics of the cells by observing only the cell images. [Background technology]

[0002] Cell culture is an important technique in biological research, including molecular biology, and refers to culturing specific cells for the purpose of diagnosing or treating human diseases. Recently, there has been a demand for efficient mass cultivation of cells and tissues (collectively referred to as "cells") in the fields of pharmaceutical production, gene therapy, regenerative medicine, immunotherapy, and other fields. In particular, stem cells are the most actively researched subject in the field of bio-life, and since they differentiate into cells with specific functions depending on the environment and stimuli, it is essential to analyze and track the cell differentiation process during the cell culture process in order to find the mechanism of action and induction methods.

[0003] When conducting cell culture and differentiation experiments, researchers must confirm whether the cells have been cultured or differentiated with the original desired morphology and characteristics; however, it is nearly impossible to perfectly grasp the minute cellular changes with the naked eye, and there are limitations to judging whether the differentiation experiment was successful during the cell differentiation process.

[0004] Therefore, the inventors have made extensive efforts to determine the characteristics of cells using only cell images without additional equipment, and have confirmed that when analyzing minute cell changes using a Convolutional Neural Network (CNN) of artificial intelligence deep learning, the characteristics of cells can be analyzed with high accuracy, thereby completing the present invention.

[0005] The information provided in this Background section is intended solely to enhance understanding of the background of the invention and may not include information that constitutes prior art already known to those of ordinary skill in the art to which the invention pertains.

[0006] (Patent Document 1) KR10-2084683 B1 (Cell image analysis method and cell image processing device using artificial neural network) Summary of the Invention [Problem to be solved by the invention]

[0007] Therefore, the present invention has been devised to solve the above problems, and one object of the present invention is to provide a cell discrimination method and device using artificial intelligence deep learning, which can distinguish and determine cell characteristics for each time period during the process of inducing maintenance culture or differentiation of cells by applying various culture conditions to various cells including stem cells, obtaining cell images that change according to time period, and pre-learning and storing management using a convolutional neural network based on deep learning.

[0008] Other objects and advantages of the present invention will be described below and can be understood by the embodiments of the present invention. Also, the objects and advantages of the present invention can be realized by the means and combinations recited in the claims. [Means for solving the problem]

[0009] In order to achieve the above-mentioned objectives, a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention includes an input step of inputting a cell image; and a discrimination step of discriminating whether the cell image corresponds to any of various cell types, various culture conditions, and culture times using a discrimination model based on deep learning, wherein the discrimination step includes a step of extracting a first feature from the cell image; a step of extracting a second feature from the cell image; and a step of determining one or more of a cell type, culture conditions, and culture time for the cell image based on the extracted first and second features; and the discrimination model includes a first neural network for extracting the first feature from the cell image; a second neural network for extracting the second feature from the cell image; and a fully connected layer for determining one or more of a cell type, culture conditions, and culture time for the input cell image based on the extracted first and second features.

[0010] In a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, the cell images can be acquired by an imaging device at any one or more of the following time periods after cell culture: 1 hour to 1 hour 30 minutes, 3 hours to 3 hours 30 minutes, 6 hours to 6 hours 30 minutes, 12 hours to 12 hours 30 minutes, and 24 hours to 24 hours 30 minutes.

[0011] In a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, the first neural network may be implemented as a shallow-structure convolutional neural network formed of one convolutional layer and one pooling layer, and the second neural network may be implemented as a deep-structure convolutional neural network formed of four convolutional layers.

[0012] In a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, the various cell types include animal cells, including one or more of stem cell lines, human dermal fibroblast cell lines, epithelial cell lines, and immune cell lines, and human cells, and the various culture conditions may be different for each cell line.

[0013] In a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, the stem cell line includes one or more of mouse embryonic stem cells, mouse dedifferentiated stem cells, human embryonic stem cells, human dedifferentiated stem cells, human neural stem cells, human hair follicle stem cells, human mesenchymal stem cells, and human fibroblasts, the epithelial cell line includes human epidermal keratinocytes (HaCaT), the immune cell line includes T cells, and the human neural stem cells include human somatic cell-derived cell-transformed neural stem cells or human brain-derived neural stem cells.

[0014] In the cell discrimination method using artificial intelligence deep learning according to an embodiment of the present invention, the mouse embryonic stem cells are cultured under at least one of a culture condition including a leukaemia inhibitory factor (LIF) medium, a culture condition including an insulin-transferrin-selenium supplement (ITS) medium, and a culture condition in which the LIF medium is removed. The culture conditions for the mouse dedifferentiated stem cells include at least one of a culture condition including PD0325901, SB431542, thiazovivin, ascorbic acid, and the like. the culture conditions include at least one of the following: culture conditions including PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; culture conditions in which PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media are removed; and culture conditions including ITS media; the human embryonic stem cells or the human dedifferentiated stem cells are cultured under at least one of the following: culture conditions including PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; culture conditions in which PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media are removed; and culture conditions including ITS media; and the human somatic cell-derived cell transformed neural stem cells are cultured under at least one of the following: DMEM / F12, N2, B27, bFGF, EGF, thiazovivin, valproic acid The culture conditions include one or more of the following: a culture condition including DMEM / F12, N2, B27, bFGF, and EGF; and a culture condition including DMEM / F12 and ITS media. The human brain-derived neural stem cells are cultured in a basal medium, an induced neural stem cell growth supplement, or a 5-AZA (azacitidine) medium.The culture conditions include at least one of the following: a culture condition including a DMEM medium, 10% FBS, Pen / Strep, L-glutamine, and streptomycin, and a culture condition including a DMEM medium and an ITS media; the human hair follicle stem cells include at least one of the following: a culture condition including a DMEM medium, 10% FBS, Pen / Strep, L-glutamine, and streptomycin, and a culture condition including a DMEM medium and an ITS media; the human mesenchymal stem cells include at least one of the following: a culture condition including a DMEM medium, 10% FBS, NEAA, and Pen / Strep, and a culture condition including a DMEM medium and an ITS media; the human fibroblasts include at least one of the following: a DMEM medium, 10% FBS, Pen / Strep, and NEAA; the HaCaT cells include at least one of the following: a DMEM medium, 10% FBS, Pen / Strep, L-glutamine, and streptomycin; the T cells include at least one of the following: RPMI Culture conditions can include 1640 media with Pen / Strep, β-mercaptoethanol, and L-glutamine.

[0015] In a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, the discrimination model includes a dataset for image learning, and the dataset may include a training image set of 1,000, 1,500, and 2,000, a validation image set of 800, and a test image set of 100.

[0016] In a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, the discrimination model may adopt a training image set of 2,000.

[0017] According to one embodiment of the present invention, a cell discrimination device using artificial intelligence deep learning includes an input unit to which a cell image is input; a discrimination unit to discriminate whether the cell image corresponds to any of various cell types, various culture conditions, and culture times using a discrimination model based on deep learning; and an output unit to provide a discrimination result of the discrimination unit to a user terminal, wherein the discrimination unit operates through a step of extracting a first feature from the cell image; a step of extracting a second feature from the cell image; and a step of determining one or more of a cell type, culture conditions, and culture time for the cell image based on the extracted first and second features, and the discrimination model includes a first neural network for extracting the cell region from the cell image; a second neural network for extracting the cell membrane region from the cell image; and a fully connected layer for determining one or more of a cell type, culture conditions, and culture time for the input cell image based on the extracted first and second features. Effect of the Invention

[0018] As described above, the cell discrimination method and device using artificial intelligence according to the present invention applies various culture conditions to various cells including stem cells to obtain cell images that change according to time periods, and performs pre-learning and storage management using a convolutional neural network based on deep learning, thereby achieving the effect of distinguishing and determining cell characteristics according to each time period during the process of maintaining and culturing cells or inducing their differentiation.

[0019] Specifically, the following effects can be achieved.

[0020] First, since the system captures images of cells that change during culture under given culture conditions for each type of cell at different times, this system will be highly useful in automatically grasping the characteristics of cells and monitoring whether the cells are being cultured to the desired shape and characteristics in future unmanned automated cell culture.

[0021] Secondly, when researchers in each laboratory conduct experiments using various cells and various culture conditions, CNN technology can be used to detect minute cellular changes, providing consistent and uniform cell culture, enabling accurate cell culture and differentiation experiments using existing established cell culture conditions. This will create a new bio market for cell culture through the construction of unmanned automated cell culture systems in the future.

[0022] Third, the Resnet50 algorithm is utilized for deep learning-based model training, which can reduce model training time and improve accuracy. [Brief description of the drawings]

[0023] [Figure 1] 1 is a flowchart showing a cell discrimination method using artificial intelligence (deep learning) according to one embodiment of the present invention. [Diagram 2] FIG. 2 is an illustrative diagram showing cell images acquired by time period in a cell discrimination method using artificial intelligence (deep learning) according to an embodiment of the present invention. [Diagram 3] FIG. 2 is an illustrative diagram showing the structure of a discrimination model in a cell discrimination method using artificial intelligence (deep learning) according to one embodiment of the present invention. [Figure 4] 1 is a flowchart showing a process of determining one or more of a cell type, culture conditions, and culture time by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention. [Diagram 5] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,000. [Figure 6]This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,000. [Figure 7] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,000. [Figure 8] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,000. [Figure 9] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,000. [Figure 10] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 11] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 12] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 13] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 14] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 15] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 16] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 17] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 18] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 19] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 20] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 21]This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 22] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 23] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 24] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Diagram 25] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 26] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 27] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 28] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Figure 29] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 1,500. [Diagram 30] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 31] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 32] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 33] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 34] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 35] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 36]This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 37] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 38] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 39] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 40] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 41] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 42] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 43] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 44] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 45] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Diagram 46] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 47] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 48] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 49] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 50] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 51]This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 52] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 53] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 54] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 55] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 56] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 57] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 58] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 59] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 60] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 61] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 62] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 63] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 64] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 65] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 66]This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 67] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 68] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 69] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 70] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 71] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 72] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 73] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 74] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 75] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 76] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 77] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 78] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 79] This is a graph showing the training results obtained by applying a discrimination model in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, which is the result of training using a training image set of 2,000. [Figure 80] 1 is a graph showing the results of comparing the training accuracy of a training set in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention. [Figure 81] 1 is a graph showing precision and recall of cell morphology learning when various cells are cultured under each media condition according to the present invention. [Figure 82]1 is a graph showing precision and recall of cell morphology learning when various cells are cultured under each media condition according to the present invention. [Figure 83] 1 is a graph showing precision and recall of cell morphology learning when various cells are cultured under each media condition according to the present invention. [Figure 84] 1 is a graph showing precision and recall of cell morphology learning when various cells are cultured under each media condition according to the present invention. [Figure 85] 1 is a graph showing precision and recall of cell morphology learning when various cells are cultured under each media condition according to the present invention. [Figure 86] 1 is a graph showing precision and recall of cell morphology learning when various cells are cultured under each media condition according to the present invention. [Figure 87] FIG. 1 is a block diagram showing a cell discrimination device using artificial intelligence (deep learning) according to one embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0024] The terms used in this specification will be briefly explained and the present invention will be specifically described.

[0025] The terms used in the present invention are selected from currently widely used general terms as much as possible in consideration of the functions in the present invention, but they may differ depending on the intentions or precedents of the engineers in the field, the emergence of new technologies, etc. In addition, in certain cases, there are terms arbitrarily selected by the applicant, and in this case, the meanings are described in detail in the corresponding description of the invention. Therefore, the terms used in the present invention should be defined based on the meanings that the terms have and the overall content of the present invention, rather than simply the names of the terms.

[0026] Throughout the specification, when a part "includes" a certain component, this does not mean excluding other components, but may further include other components, unless otherwise specified to the contrary. Furthermore, the terms "module" and "unit" described in the specification mean a unit that processes at least one function or operation, which may be embodied in hardware or software, or a combination of hardware and software. Furthermore, throughout the specification, when a part is said to be "connected" to another part, this includes not only "directly connected" but also "connected with another component between them."

[0027] Hereinafter, the present invention will be described in detail with reference to the accompanying drawings so that a person skilled in the art to which the present invention pertains can easily carry out the present invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. In order to clearly explain the present invention in the drawings, parts that are not related to the description are omitted, and similar parts are designated by similar reference numerals throughout the specification.

[0028] The present invention will now be described in detail with reference to the accompanying drawings.

[0029] As one embodiment of the present invention, a cell discrimination method using artificial intelligence (deep learning) may be provided.

[0030] FIG. 1 is a flowchart showing a cell discrimination method using artificial intelligence (AI) deep learning according to one embodiment of the present invention, FIG. 2 is an illustrative diagram showing cell images obtained by time period in a cell discrimination method using artificial intelligence (AI) deep learning according to one embodiment of the present invention, FIG. 3 is an illustrative diagram showing the structure of a discrimination model in a cell discrimination method using artificial intelligence (AI) deep learning according to one embodiment of the present invention, and FIG. 4 is a flowchart showing a process of applying a discrimination model to determine one or more of cell type, culture conditions, and culture time in a cell discrimination method using artificial intelligence (AI) deep learning according to one embodiment of the present invention.

[0031] First, referring to FIG. 1, a cell discrimination method using artificial intelligence (deep learning) according to one embodiment of the present invention may include an input step (S100) in which a cell image 10 is input; and a discrimination step (S200) in which the cell image 10 corresponds to one of various cell types, various culture conditions, and culture times using a discrimination model 100 based on deep learning.

[0032] In this specification, the cell image 10 refers to an image obtained by using an imaging device such as an optical microscope, etc. There is no limitation on the method of capturing the cell image using the imaging device.

[0033] The cell images 10 can be acquired at predetermined time intervals during cell culture.

[0034] For example, the cell image 10 can be acquired by an imaging device at any one or more of the following time periods after cell culture: 1 hour to 1 hour 30 minutes, 3 hours to 3 hours 30 minutes, 6 hours to 6 hours 30 minutes, 12 hours to 12 hours 30 minutes, and 24 hours to 24 hours 30 minutes.

[0035] In cell culture, although this differs for each cell line, the change in cell morphology during proliferation is greatest initially, and therefore it is best to obtain cell images within 24 to 25 hours, preferably within 24 hours.

[0036] Referring to FIG. 2, the cell images 10 are acquired at minimum time intervals, that is, 1 hour, 3 hours, 6 hours, 12 hours, and 24 hours, after cell culture.

[0037] For reference, Figure 2 shows cell images taken at 1 hour, 3 hours, 6 hours, 12 hours, and 24 hours after culturing mouse embryonic stem cell (mES) B6 cells in LIF (leukaemia inhibitory factor) media for maintenance and ITS (insulin-transferrin-selenium supplement) media for differentiation.

[0038] In the present invention, various culture conditions are applied to various cells including stem cells to obtain cell images that change over time periods, and deep learning technology is applied to the obtained cell images to analyze minute cellular changes, thereby making it possible to distinguish and determine cell characteristics over each time period during the process of maintaining and culturing cells or inducing differentiation.

[0039] Various deep learning technologies may be applied to the cell discrimination method using artificial intelligence deep learning according to an embodiment of the present invention. In other words, by learning a cell image based on the deep learning technology, it is possible to discriminate which of various cell types, various culture conditions, and culture times the cell image 10 corresponds to by using a discrimination model 100 generated.

[0040] Preferably, in the present invention, the discrimination model 100 can be generated based on a convolutional neural network (CNN) among various deep learning techniques.

[0041] More preferably, in the present invention, the discriminant model 100 can be generated based on the Resnet50 algorithm among various convolutional neural networks (CNNs).

[0042] 3, the discrimination model 100 may include a first neural network 110 for extracting a first feature from the cell image 10, a second neural network 120 for extracting a second feature from the cell image 10, and a fully connected layer 130 for determining one or more of a cell type, culture conditions, and culture time for the input cell image 10 based on the extracted first and second features. Here, the fully connected layer may correspond to a classifier that determines one or more of a cell type, culture conditions, and culture time for the cell image 10 based on the extracted feature information.

[0043] Here, the first feature of the cell image 10 may be a large feature, for example a cell morphology feature, and the second feature of the cell image 10 may be a small feature, for example a cell edge feature.

[0044] Also, the first neural network 110 may be implemented as a shallow-structured convolutional neural network formed of one first convolutional layer 112 (Conv1) and one pooling layer 113, and the second neural network 120 may be implemented as a deep-structured convolutional neural network formed of four second to fifth convolutional layers 121, 122, 123, and 124 (Conv2, Conv3, Conv4, and Conv5). Here, the pooling layer 113 may perform a pooling operation to reduce the size of output data from the convolutional layer 112 or to emphasize specific data. The pooling layer 113 may include a max pooling layer and an average pooling layer.

[0045] Specifically, in the discrimination model 100, the cell image 10 used as an input may be, for example, 240×320 pixels in size.

[0046] When the cell image 10 is input to the first neural network 110, image features are extracted through the convolution layer 112 and the pooling layer 113. At this time, since the size of the image is reduced after processing, a zero padding 111 process is performed to maintain the size of the image, and the image size is temporarily enlarged before subsequent processing begins. For example, the size of the cell image 10 is enlarged from 240×320 to 246×236 through the zero padding 111 process. As a result, an image of 120×160×64 pixels is input to the convolution layer 112, and then an image of 60×80×64 pixels is input to the second neural network 120 through a pooling operation in the pooling layer 113.

[0047] The second to fifth convolutional layers 121, 122, 123, and 124 of the second neural network 120 extract image features using a filter of size 1×1 or 3×3.

[0048] For example, in the second convolutional layer 121 (Conv2), an image pattern is analyzed through (1×1, 64), (3×3, 64), and (1×1, 256) filters, and this analysis is repeated three times. That is, the second convolutional layer 121 (Conv2) comprises nine layers.

[0049] In the third convolutional layer 122 (Conv3), the image pattern is analyzed through (1×1, 128), (3×3, 128), and (1×1, 512) filters, and this analysis is repeated four times. That is, the third convolutional layer 122 (Conv3) consists of 12 layers.

[0050] In the fourth convolutional layer 123 (Conv4), the image pattern is analyzed through (1×1, 256), (3×3, 256), and (1×1, 1024) filters, and this analysis is repeated six times. That is, the fourth convolutional layer 123 (Conv4) comprises 18 layers.

[0051] In the fifth convolutional layer 124 (Conv5), the image pattern is analyzed through (1×1, 512), (3×3, 512), and (1×1, 2048) filters, and this analysis is repeated three times. That is, the fifth convolutional layer 124 (Conv5) consists of nine layers.

[0052] Therefore, the second neural network 120 is composed of a total of 48 layers.

[0053] Similarly, the first neural network 110 is composed of two layers: a first convolutional layer 112 (Conv1) and a pooling layer 113.

[0054] Figure 4 is a flowchart showing a process of applying a discrimination model to determine one or more of a cell type, culture conditions, and culture time in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention.

[0055] 4, the discrimination step (S200) may include the steps of: extracting a first feature in the cell image 10 (S210); extracting a second feature in the cell image 10 (S220); and determining one or more of a cell type, a culture condition, and a culture time by applying the discrimination model 100 based on the first feature and the second feature extracted from the cell image 10. Here, the first feature of the cell image 10 may be a large feature, for example, a cell morphology feature, and the second feature of the cell image 10 may be a small feature, for example, a cell edge feature.

[0056] In a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, the various cell types include animal cells and human cells, including one or more of stem cell lines, human dermal fibroblasts, epithelial cell lines, and immune cell lines, and the various culture conditions may be different for each cell line.

[0057] The stem cell line may include one or more of mouse embryonic stem cells (mES), mouse induced pluripotent stem cells (miPSCs), human embryonic stem cells, human dedifferentiated stem cells, human neural stem cells, human hair follicle stem cells, human mesenchymal stem cells, and human fibroblasts; the epithelial cell line may include human epidermal keratinocytes (HaCaT); the immune cell line may include T cells; and the human neural stem cells may include human somatic cell-derived cell-transformed neural stem cells or human brain-derived neural stem cells.

[0058] In the present invention, the culture conditions are briefly described as follows.

[0059] The mouse embryonic stem cells may be cultured under one or more of the following conditions: a culture condition containing a leukaemia inhibitory factor (LIF) medium, a culture condition containing an insulin-transferrin-selenium supplement (ITS) medium, and a culture condition in which the LIF medium is removed. Here, the LIF medium serves to maintain the characteristics of embryonic stem cells, and the ITS medium serves to induce differentiation.

[0060] The culture conditions for the mouse dedifferentiated stem cells may include at least one of the following: culture conditions including PD0325901 (MEK (mitogen-activated protein kinase) inhibitor), SB431542 (TGF-β (Transforming Growth Factor-β) inhibitor), thiazovivin, ascorbic acid (AA), and LIF media; culture conditions excluding PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; and culture conditions including ITS media. Here, the four small molecules PD0325901, SB431542, thiazovivin, and ascorbic acid function to maintain the characteristics and chromosome stability of mouse dedifferentiated stem cells.

[0061] The human embryonic stem cells or the human dedifferentiated stem cells may include at least one of the following culture conditions: culture conditions including PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; culture conditions excluding PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; and culture conditions including ITS media. Here, the four small molecule compounds PD0325901, SB431542, thiazovivin, and ascorbic acid function to maintain chromosome stability.

[0062] The human somatic cell-derived cell transformed neural stem cells may include any one or more of the following culture conditions: DMEM / F12 (Dulbecco's Modified Eagle's Medium), N2 (N2 supplement), B27 (serum supplement), bFGFF (basic fibroblast growth factor), EGF (epidermal growth factor), thiazovivin, valproic acid, purmorphamine, A8301, SB431542, CHIR99021, DZNep (Deazaneplanocin A), and 5-AZA (Azacitidine); culture conditions including DMEM / F12, N2, B27, bFGF, and EGF; and culture conditions including DMEM / F12 and ITS media. Here, the small molecule compounds thiazovivine, valproic acid, palmorfamine, A8301, SB431542, CHIR99021, DZNep, and 5-AZA function to maintain chromosomal stability.

[0063] The human brain-derived neural stem cells may be cultured in a culture condition comprising a basal medium, an induced neural stem cell growth supplement, and antibiotics, a culture condition comprising a basal medium and antibiotics, and a culture condition comprising a basal medium, antibiotics, and ITS media.

[0064] The human hair follicle stem cells may be cultured under one or more of the following conditions: DMEM media containing 10% FBS (fetal bovine serum), Pen / Strep (penicillin & streptomycin), L-glutamine, and streptomycin; and DMEM media containing ITS media.

[0065] The human mesenchymal stem cells may be cultured under one or more of the following culture conditions: DMEM media containing 10% FBS (Fetal bovine serum), NEAA (non-essential amino acids), and Pen / Strep; and DMEM media containing ITS media.

[0066] The human fibroblasts can be cultured in DMEM media containing 10% FBS, Pen / Strep, and NEAA.

[0067] The HaCaT cells can be cultured in DMEM media containing 10% FBS, Pen / Strep, L-glutamine, and streptomycin.

[0068] The T cells may be cultured in RPMI (Rosewell Park Memorial Institute, USA) 1640 media containing Pen / Strep, β-mercaptoethanol, and L-glutamine.

[0069] In a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention, the discrimination model 100 includes a dataset for image learning, and the dataset may include a training image set of 1,000, 1,500, and 2,000, a validation image set of 800, and a test image set of 100.

[0070] Preferably, as a result of training, the discriminant model 100 is able to adopt a training image set of 2,000 images.

[0071] [Training results] After cell culture under various cell types and various culture conditions, cell images were obtained at the shortest time intervals, i.e., 1 hour, 3 hours, 6 hours, 12 hours, and 24 hours, and the discrimination model (also called CNN model) according to the present invention was applied to compare the training accuracy.

[0072] As the data sets for the discriminant model, different training image sets of 1,000, 1,500, and 2,000 were used, and a common validation image set of 800 and a test image set of 100 were used.

[0073] [Training set of 1,000 images] B6 cells (referred to as B6 embryonic muscle cells) from mouse embryonic stem cells were cultured using LIF and ITS media, and the training results are summarized in Figures 5 to 9.

[0074] For reference, the graph on the left shows the accuracy and loss (or loss value) of training and validation. The accuracy and loss of training are denoted by train_acc and train_loss, respectively, and the accuracy and loss of validation are denoted by val_acc and val_loss, respectively.

[0075] Also, the confusion matrix on the right is a table showing the accuracy using 100 images from the test image set.

[0076] Referring to the left graphs of Figures 5 to 9, the training and validation accuracy was close to 1 in time slots 1, 3, and 6, but it can be seen that the training and validation accuracy gradually decreased in time slots 12 and 24. It can also be seen that the training and validation loss appeared unstable in all time slots, and increased to over 0.5 in particular in time slot 24.

[0077] Referring to the confusion matrix on the right side of Figures 5 to 9, most of the time periods except for the 24-hour period showed accuracy rates exceeding 80%, but this does not mean that the learning results of this model showed high accuracy.

[0078] As mentioned above, the training accuracy was not high when using a training set of 1,000 training images.

[0079] [Training set of 1,500 images] B6 cells (referred to as B6 embryonic muscle cells), which are mouse embryonic stem cells, were cultured using LIF and ITS media, and then cultured using media with LIF added and LIF removed (shown as LIF-), and the training results are summarized in Figures 10 to 14 and Figures 15 to 19.

[0080] In addition, mouse dedifferentiated stem cells (miPSCs-) were cultured using LIF and ITS media, and using media with LIF added and LIF removed (shown as LIF-), and the training results are summarized in Figures 20 to 24 and Figures 25 to 29. For reference, the "-" in "miPSCs-" means that the four chemicals were not added to the culture conditions.

[0081] As mentioned above, the training results using a set of 1,500 training images indicate that the accuracy of cell classification by the media was fairly high for B6 embryonic muscle cells, but it is not believed to be reliable enough to distinguish cells in actual cell cultures.

[0082] In addition, in the training of the dedifferentiated stem cells, the validation loss was found to increase due to overfitting, which can be seen as an indication that the training was not performed normally.

[0083] This was trained using a training set of 2,000 images.

[0084] [Training set of 2,000 images] B6 cells (also called B6 embryonic muscle cells), which are mouse embryonic stem cells, were cultured using LIF and ITS media, and then cultured using media with LIF added and LIF removed (shown as LIF-), and the training results are summarized in Figures 30 to 34 and Figures 35 to 39.

[0085] Looking at the graphs on the left in Figures 30 to 34, we can see that the accuracy of training and validation for all time periods is almost 1, and the loss has also decreased to almost 0. This means that model training and validation were performed normally, and the confusion matrix on the right also showed a higher accuracy result than the initial training result.

[0086] Referring to the graphs on the left in Figures 35 to 39, it can be seen that the accuracy of training and validation by time period is almost close to 1, and the loss also converges to 0. In other words, the training and validation processes are considered to have been performed normally, and the confusion matrix on the right shows an accuracy of over 95% up to 12 hours, and at 24 hours, the accuracy slightly decreased due to the size of the cells, but still showed an accuracy of over 90%.

[0087] J1 cells (also called J1 embryonic muscle cells) of mouse embryonic stem cells were cultured using LIF and ITS media, and also cultured using media with LIF added and LIF removed (shown as LIF-), and the training results are summarized in Figures 40 to 44 and Figures 45 to 49.

[0088] Referring to the graphs on the left in Figures 40 to 44, the training and validation accuracies were all close to 1. In the confusion matrix, the accuracy of distinguishing cells cultured in LIF media was 99% or 100%, and the accuracy of distinguishing cells cultured in ITS media was over 97%.

[0089] Referring to the left graphs in Figures 45 to 49, the accuracy of training and validation is close to 1, and the confusion matrix also shows an accuracy of over 99%.

[0090] Mouse dedifferentiated stem cells (miPSCs+Line1) were cultured using LIF and ITS media, and then cultured using media with LIF added (LIF) and media without LIF (LIF-), and the training results are summarized in Figures 50 to 54 and 55 to 59. For reference, the "+" in "miPSCs+" means that four types of chemicals were added to the culture conditions.

[0091] Referring to Figures 50 to 54, the accuracy of training and validation in the left graph is almost 1, but the validation loss is stable except for the 1st and 12th hour.

[0092] Referring to Figures 55 to 59, the accuracy of training and validation was generally good in the graphs on the left, but validation loss was unstable in time slots 1 and 24. This is most likely due to the generation of cell debris and the rapid proliferation rate of cells. The accuracy of error alignment was over 96% except for time slot 24, which appears to have been low due to limitations in distinguishing cell morphology due to the rapid proliferation rate of cells and debris as mentioned above.

[0093] Mouse dedifferentiated stem cells (miPSCs+Line2) were cultured using LIF and ITS media, and also using media with LIF added and LIF removed (shown as LIF-), and the training results are summarized in Figures 60 to 64 and Figures 65 to 69.

[0094] Referring to Figures 60 to 64, the accuracy of training and validation in the graph on the left is close to 1, and except for one time period where the validation loss was a little uncertain, the loss was close to 0 in all time periods. The accuracy of the confusion matrix was over 98%, and the ITS media was over 93%, showing reliable results.

[0095] Referring to Figures 65 to 69, except for the fact that the validation loss was slightly unstable over the 24-hour period in the graph on the left, the training and validation accuracies were close to 1. The confusion matrix also showed that the accuracy of distinguishing between cells cultured in each medium was very high over all time periods.

[0096] Mouse dedifferentiated stem cells (miPSCs-) were cultured using LIF and ITS media, and also using media with LIF added and LIF removed (denoted as LIF-), and the training results are summarized in Figures 70 to 74 and Figures 75 to 79.

[0097] 70 to 74, in the graphs on the left, the validation loss rate was generally unstable from time 6 onwards, and in the final 24 hour period, even though the training accuracy was high, there was too much cell debris and the cells proliferated slowly, resulting in an excessively large validation loss rate. In the 1 hour period, the accuracy of cells cultured in LIF media was 79%, and in the 24 hour period, the accuracy of cells cultured in ITS was approximately 59%, a low accuracy. In the 24 hour period, there was too much cell debris to distinguish the cell morphology, and the cells proliferated quickly, resulting in low validation loss rate and test accuracy.

[0098] 75 to 79, the left graph shows that the training and validation accuracies are generally high, but the validation loss is unstable over the 24-hour period. The right confusion matrix also shows that the accuracy is not as high as the results for the remaining cells.

[0099] As shown above, the training using the 2,000 training image set did not suffer from the overfitting phenomenon seen in the training using the 1,500 training image set, and a significantly high accuracy was achieved even with mouse dedifferentiated stem cells (miPSCs), which are difficult to obtain high accuracy with.

[0100] These results provide the possibility of distinguishing cell morphological changes due to media type with high accuracy in future unmanned automated cell culture systems.

[0101] Figure 80 is a graph showing the results of comparing the training accuracy of a train set in a cell discrimination method using artificial intelligence deep learning according to one embodiment of the present invention.

[0102] Referring to FIG. 80, when comparing the training accuracy of the initial 1,000 and 2,000 train sets in terms of the accuracy of distinguishing between LIF (leukemia inhibitory factor) media for maintenance purposes and ITS (insulin-transferrin-selenite) media for differentiation purposes, it can be seen that the greater the number of trained data, i.e., the higher the accuracy is for the 2,000 train set.

[0103] 81 to 86 are graphs showing the precision and recall of cell morphology learning when various cells are cultured under each media condition according to the present invention. Here, precision is the ratio of actual true results among those classified as true by model learning, and recall is the ratio of actual true results predicted by the model.

[0104] Referring to Figure 81, in training using a set of 1,000 training images of cells cultured in LIF media (see the left graph in the top graph), the precision of cell morphology learning decreases over time. Also, the recall rate of cell morphology learning was highest in the three time periods, but the recall rate value was not high overall. Similarly, in ITS media for differentiation purposes, the precision and recall rates were highest in the three time periods (see the right graph in the top graph), but overall the values ​​were unstable and low.

[0105] However, when training was done using a set of 2,000 training images of cells cultured in LIF media and ITS media (see graph below), both the precision and recall of cell morphology learning were high.

[0106] Referring to Figure 82, in the training using a 2,000 training image set, high values ​​were observed for both precision and recall for cells cultured in LIF media and LIF-free media (LIF-). In addition, in the 24-hour period, precision was relatively low due to limitations in distinguishing cell morphology due to cell proliferation, but still showed high precision.

[0107] Referring to Figure 83, the precision and recall of distinguishing the morphology of cells [J1 mESCs (mouse embryonic stem cells)] (mouse embryonic stem cells J1) cultured under each media condition were high, close to 1, for all time periods in LIF media, ITS media, LIF media, and LIF(-) media. This means that the CNN model can also distinguish J1 mESCs cells (mouse embryonic stem cells J1), and therefore this algorithm can be applied to a variety of cells.

[0108] Referring to Figure 84, the precision and recall of cells (miPSCs (mouse induced pluripotent stem cells) Line 1) cultured under each media condition were lower in precision and recall at 12 hours in the LIF + 4 chemicals (e.g., containing PD0325901, SB431542, thiazovivin, ascorbic acid) media and ITS media for differentiation than at other times due to the generation of cell debris (see the top two graphs). Also, cells cultured in LIF + 4 chemicals media and media without LIF and 4 chemicals (LIF (-) + 4 chemicals (-)) had limitations in including only the cell morphology in the image at 24 hours due to cell proliferation and the generation of debris, resulting in lower precision and recall than at other times (see the bottom two graphs). However, high precision and recall were observed at other times.

[0109] Referring to Figure 85, the precision and recall rate of distinguishing cells (miPSCs (mouse induced pluripotent stem cells) Line 2) cultured under each media condition (same conditions as in Figure 84) based on their morphology are shown, and the precision and recall rates showed values ​​close to 1 throughout the entire time period.

[0110] Therefore, even for the same cells, there may be differences in cell morphology depending on the cell passage or researcher, but the results of training with the CNN model are able to distinguish subtle differences in cell morphology and derive consistent results with high accuracy.

[0111] For reference, Line 1 and Line 2 were cultured under the same culture conditions but are different cell lines.

[0112] Referring to FIG. 86, the precision and reproducibility of distinguishing cells (miPSCs-) cultured under each media condition (same as the conditions in FIG. 82) based on their morphology are shown.

[0113] In media containing LIF, precision decreased over time, but recall increased (see the left graph of the top two graphs). Here, it appears that training accuracy decreased over time due to the generation of cell debris, but recall shows that the proportion of cells that the CNN model determined to be cultured in LIF media that actually matched increased, so training appears to have been performed normally to some extent.

[0114] Conversely, for cells cultured in ITS media, which induced differentiation, the precision increased over time, but the recall decreased (see the right graph of the top two graphs). Because the morphology of cells cultured in ITS media changed rapidly and a lot of cell debris was generated, the recall was slightly lower even though the training accuracy was high.

[0115] In LIF media and media without LIF (LIF(-)), the reproducibility of cells cultured in LIF media was generally high, except for one time period when it was low. This means that the CNN model was able to distinguish cells well under the two media conditions, even in cells where differences in cellular changes are difficult to distinguish by eye (see the two graphs at the bottom).

[0116] Meanwhile, Figure 87 is a block diagram showing a cell discrimination device using artificial intelligence deep learning according to one embodiment of the present invention.

[0117] Referring to Figure 87, a cell discrimination device using artificial intelligence deep learning according to one embodiment of the present invention may include an input unit 200 to which a cell image 10 is input, a discrimination unit 300 that uses a discrimination model 100 based on deep learning to discriminate which of various cell types, various culture conditions, and culture times the cell image 10 corresponds to, and an output unit 400 that provides the discrimination result of the discrimination unit 300 to a user terminal 20.

[0118] The cell image 10 input through the input unit 200 can be stored in a database 210 .

[0119] The user terminal 20 may refer to a device used by a user for cell discrimination. That is, the user terminal 20 may include any device that can provide the result of cell discrimination on the cell image 10 to the user through a display or a sound signal.

[0120] In addition, as an embodiment of the present invention, a computer readable recording medium having a program for implementing the above-mentioned method recorded thereon may be provided.

[0121] Meanwhile, the above-mentioned methods can be created as a computer executable program, and can be embodied in a general-purpose digital computer that runs the program using a computer readable medium. Also, the data structures used in the above-mentioned methods can be recorded in a computer readable medium through various means. A recording medium that records executable computer programs or codes for executing the various methods of the present invention should not be understood to include transitory objects such as carrier waves or signals. The computer readable medium can include recording media such as magnetic recording media (e.g., ROM, floppy disk, hard disk, etc.) and optically readable media (e.g., CD-ROM, DVD, etc.).

[0122] The above description of the present invention is for illustrative purposes only, and those skilled in the art will understand that the present invention can be easily modified into other specific forms without changing the technical concept or essential features of the present invention. Therefore, it should be understood that the above-described embodiments are illustrative and not limiting in all respects. For example, each component described as a single type may be implemented in a distributed manner, and similarly, each component described as a distributed type may be implemented in a combined form.

[0123] The scope of the present invention is indicated by the claims below rather than by the detailed description, and all modifications and variations derived from the meaning and scope of the claims and their equivalent concepts should be interpreted as being included in the scope of the present invention.

Claims

1. An input step in which a cell image is input; and using a deep learning based discrimination model to discriminate whether the cell images correspond to various cell types, various culture conditions, and culture times; The determining step includes: Extracting a first feature within the cell image; extracting a second feature within the cell image; and determining at least one of a cell type, a culture condition, and a culture time for the cell image based on the extracted first feature and the extracted second feature; The discrimination model is a first neural network for extracting the first features from the cell images; a second neural network for extracting the second features from the cell images; and a fully connected layer for determining one or more of a cell type, a culture condition, and a culture time for an input cell image based on the extracted first feature and the extracted second feature; The culture conditions include one or more of culture conditions including media for maintaining cell characteristics, culture conditions including media for inducing cell differentiation, and culture conditions that remove media for maintaining cell characteristics.

2. 2. The method of claim 1, wherein the cell images are obtained by an imaging device at one or more of the following time periods after cell culture: 1 hour to 1 hour 30 minutes, 3 hours to 3 hours 30 minutes, 6 hours to 6 hours 30 minutes, 12 hours to 12 hours 30 minutes, and 24 hours to 24 hours 30 minutes.

3. The first neural network is implemented as a shallow convolutional neural network formed of one convolutional layer and one pooling layer, 2. The method of claim 1, wherein the second neural network is implemented as a deep-structure convolutional neural network formed of four convolutional layers.

4. The various cell types include animal and human cells, including at least one of stem cell lines, human dermal fibroblasts, epithelial cell lines, and immune cell lines; 2. The method of claim 1, wherein the various culture conditions are different for each cell line.

5. The stem cell line includes at least one of mouse embryonic stem cells, mouse dedifferentiated stem cells, human embryonic stem cells, human dedifferentiated stem cells, human neural stem cells, human hair follicle stem cells, human mesenchymal stem cells, and human fibroblast cells; The epithelial cell line comprises human epidermal keratinocyte (HaCaT) cells; The immune cell line comprises a T cell, 5. The method of claim 4, wherein the human neural stem cells include human somatic cell-derived cell transformed neural stem cells or human brain-derived neural stem cells.

6. The mouse embryonic stem cells are Culture conditions including a leukaemia inhibitory factor (LIF) medium, which is a medium for maintaining the cell characteristics; Culture conditions including ITS (insulin-transferrin-selenium supplement) media, which is a media for inducing cell differentiation; and culture conditions in which LIF media is removed, The culture conditions of the mouse dedifferentiated stem cells are as follows: Culture conditions including PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; Culture conditions that excluded PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; and Any one or more of the culture conditions including ITS media, The human embryonic stem cells or the human dedifferentiated stem cells are Culture conditions including PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; Culture conditions with the removal of PD0325901, SB431542, thiazovivin, ascorbic acid and LIF media; and Any one or more of the culture conditions including ITS media, The human somatic cell-derived transformed neural stem cells are Culture conditions including DMEM / F12, N2, B27, bFGF, EGF, thiazovivin, valproic acid, purmorphamine, A8301, SB431542, CHIR99021, DZNep (Deazaneplanocin A), and 5-AZA (Azacitidine); Culture conditions including DMEM / F12, N2, B27, bFGF, and EGF, and Any one or more of the culture conditions including DMEM / F12 and ITS media, The human brain-derived neural stem cells are Culture conditions including basal medium, induced neural stem cell growth supplement and antibiotics; Culture conditions including basal medium and antibiotics; Culture conditions include one or more of a basal medium, an antibiotic, and an ITS media; The human hair follicle stem cells are Culture conditions included 10% FBS, Pen / Strep, L-glutamine, and streptomycin in DMEM media; and The culture conditions include one or more of the following: DMEM medium containing ITS media; The human mesenchymal stem cells are Culture conditions included 10% FBS, NEAA, and Pen / Strep in DMEM media; The culture conditions include one or more of the following: DMEM medium containing ITS media; The human fibroblasts Culture conditions included 10% FBS, Pen / Strep, and NEAA in DMEM media; The human epidermal keratinocyte (HaCaT) cells are Culture conditions included 10% FBS, Pen / Strep, L-glutamine, and streptomycin in DMEM media; The method of claim 5, wherein the T cells are cultured in RPMI 1640 medium containing Pen / Strep, β-mercaptoethanol, and L-glutamine.

7. The discrimination model is The method for cell discrimination using artificial intelligence according to claim 1, further comprising a data set for image learning, the data set including a training image set of 1,000, 1,500, and 2,000, a validation image set of 800, and a test image set of 100.

8. The method of claim 6, wherein the discrimination model adopts a training image set of 2,000.

9. an input section to which a cell image is input; a discrimination unit that discriminates whether the cell image corresponds to any of various cell types, various culture conditions, and culture times using a discrimination model based on deep learning; and an output unit for providing a determination result of the determination unit to a user terminal; The discrimination unit is Extracting a first feature within the cell image; extracting a second feature within the cell image; and determining at least one of a cell type, a culture condition, and a culture time for the cell image based on the extracted first and second features; The discrimination model is a first neural network for extracting the first features from the cell images; a second neural network for extracting the second features from the cell images; and a fully connected layer for determining one or more of a cell type, a culture condition, and a culture time for an input cell image based on the extracted first feature and the extracted second feature; The culture conditions include one or more of culture conditions including media for maintaining cell characteristics, culture conditions including media for inducing cell differentiation, and culture conditions that remove media for maintaining cell characteristics.

10. The cell discrimination device using artificial intelligence according to claim 9, characterized in that the cell images are acquired by an imaging device at one or more of the following time periods after cell culture: 1 hour to 1 hour 30 minutes, 3 hours to 3 hours 30 minutes, 6 hours to 6 hours 30 minutes, 12 hours to 12 hours 30 minutes, and 24 hours to 24 hours 30 minutes.

11. The first neural network is implemented as a shallow convolutional neural network formed of one convolutional layer and one pooling layer, 10. The cell discrimination device according to claim 9, wherein the second neural network is implemented as a deep-structure convolutional neural network formed of four convolutional layers.

12. The various cell types include animal and human cells, including at least one of stem cell lines, human dermal fibroblasts, epithelial cell lines, and immune cell lines; 10. The cell discrimination device using artificial intelligence according to claim 9, wherein the various culture conditions are different for each cell line.

13. The stem cell line includes at least one of mouse embryonic stem cells, mouse dedifferentiated stem cells, human embryonic stem cells, human dedifferentiated stem cells, human neural stem cells, human hair follicle stem cells, human mesenchymal stem cells, and human fibroblast cells; The epithelial cell line comprises human epidermal keratinocyte (HaCaT) cells; The immune cell line comprises a T cell, The cell discrimination device using artificial intelligence according to claim 12, wherein the human neural stem cells include human somatic cell-derived cell transformed neural stem cells or human brain-derived neural stem cells.

14. The mouse embryonic stem cells are Culture conditions including a leukaemia inhibitory factor (LIF) medium, which is a medium for maintaining the cell characteristics; Culture conditions including ITS (insulin-transferrin-selenium supplement) media, which is a media for inducing cell differentiation; culture conditions in which LIF media is removed, The culture conditions of the mouse dedifferentiated stem cells are as follows: Culture conditions including PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; Culture conditions with the removal of PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; Any one or more of the culture conditions including ITS media, The human embryonic stem cells or the human dedifferentiated stem cells are Culture conditions including PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; Culture conditions with the removal of PD0325901, SB431542, thiazovivin, ascorbic acid, and LIF media; Any one or more of the culture conditions including ITS media, The human somatic cell-derived transformed neural stem cells are Culture conditions including DMEM / F12, N2, B27, bFGF, EGF, thiazovivin, valproic acid, purmorphamine, A8301, SB431542, CHIR99021, DZNep (Deazaneplanocin A) and 5-AZA (Azacidine); Culture conditions including DMEM / F12, N2, B27, bFGF, and EGF; DMEM / F12, and ITS media, The human brain-derived neural stem cells are Culture conditions including basal medium, induced neural stem cell growth supplement, and antibiotics; Culture conditions including basal medium and antibiotics; Culture conditions include one or more of a basal medium, an antibiotic, and an ITS media; The human hair follicle stem cells are Culture conditions included 10% FBS, Pen / Strep, L-glutamine, and streptomycin in DMEM media; The culture conditions include one or more of the following: DMEM medium containing ITS media; The human mesenchymal stem cells are Culture conditions included 10% FBS, NEAA, and Pen / Strep in DMEM media; The culture conditions include one or more of the following: DMEM medium containing ITS media; The human fibroblasts Culture conditions included 10% FBS, Pen / Strep, and NEAA in DMEM media; The human epidermal keratinocyte (HaCaT) cells are Culture conditions included 10% FBS, Pen / Strep, L-glutamine, and streptomycin in DMEM media; The cell discrimination device using artificial intelligence according to claim 13, wherein the T cells are cultured under conditions including Pen / Strep, β-mercaptoethanol, and L-glutamine in RPMI 1640 media.

15. The discrimination model is The cell discrimination device using artificial intelligence of claim 9, further comprising a data set for image learning, the data set including a training image set of 1,000, 1,500, and 2,000, a validation image set of 800, and a test image set of 100.

16. The cell discrimination device using artificial intelligence according to claim 15, wherein the discrimination model adopts a training image set of 2,000.

Images