Imaging system and method of use

Abstract

To provide an imaging system and a controller that generate and / or analyze a variety of objects, such as biological cells to determine clonality.SOLUTION: An imaging system comprises: an imaging device; and a controller in operable connection to the imaging device. The controller performs the steps of: generating a plurality of chronological images of an image area; identifying a target object within the image area; using a prior image of the image area, and cropping the prior image to generate a cropped image area sized to the perimeter of the target object image area; generating a location region of the cropped image area within the image area of the most recent image; analyzing the location region of the most recent image; and inductively iterating the steps for the plurality of chronological images to determine clonality of the target object.SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] Cross - Reference to Related Applications This application claims the benefit of priority under 35 U.S.C. § 119(e) to U.S. Provisional Patent Application No. 62 / 910,951, filed October 4, 2019; U.S. Provisional Patent Application No. 62 / 971,017, filed February 6, 2020; and U.S. Provisional Patent Application No. 63 / 051,310, filed July 13, 2020; the entire contents of these provisional applications are incorporated herein by reference.

[0002] Field of the Invention The present invention generally relates to imaging, and more specifically to systems and methods for generating an image of a target object and analyzing the target object within the generated image.

Background Art

[0003] Background Information Monoclonality is established by the isolation and subsequent growth of single cells derived from a parental population, which is often considered an essential step in developing high - quality cell lines. This procedure aims to minimize or eliminate genomic and phenotypic heterogeneity in an attempt to maximize the homogeneity of the cell line. For example, a newly genomically engineered cell population may contain a mixture of cells with diverse alleles, ploidies, and epigenetic characteristics. Thus, a homogeneous cell line can only be re - established by ensuring that all cells within the parental population are derived from a single ancestral cell isolated downstream of events that introduce variation. This step is called single - cell cloning.

[0004] One example of a cell culture process where monocloning is often considered critical is that of human induced pluripotent stem cells (iPSCs). This cell type, with its ability to self-replicate indefinitely and differentiate through any lineage, is highly promising for modeling disease states in vitro, enabling non-invasive genetic association studies, particularly those concerning drug responses. Such endeavors inevitably require large-scale cohorts at the population level. Therefore, the throughput of cell line derivation is the greatest limiting factor in unlocking the vast potential of iPSC technology in areas including functional genomics and precision medicine. The reprogramming process of iPSCs places significant stress on the cells, resulting in a highly heterogeneous population with respect to variables such as residual viral vector loads and resulting chromosomal abnormalities, thus necessitating monocloning. While fully automated methods for iPSC generation have been described, the need for monocloning workflows in iPSC generation remains, especially when using viral vectors for iPSC vectors. Historically, this stage has been a major bottleneck in automated, high-throughput iPSC derivation, making this cell type a promising case study for investigating monoclonalization methodologies.

[0005] Single-cell isolation is typically achieved via fluorescence-activated cell sorting (FACS), a form of flow cytometry. While this process allows for rapid sorting of individual cells, several factors can lead to undesirable outcomes. Sorted cells may not survive, leaving empty wells; or malfunctions in the sorting process may result in more than one cell being mistakenly transferred to the target well, leading to polyclonality. Furthermore, for any given cell type, various morphological or physiological changes can occur during development that alter the quality of the cell line. For example, in the case of stem cells (SCs), there are several known morphological markers that indicate loss of pluripotency, a defect widely seen in newly reprogrammed iPSCs. As a result of these factors, the presence, clonality, and quality of cell aggregates in wells presumed to be monoclonal must be validated retrospectively.

[0006] Currently, the only way to validate monoclonality is to manually examine microscopic images taken at regular intervals to track colony growth after sorting. Doing so is extremely time-consuming, with technicians often spending several hours a day classifying wells based on the presence, clonality, and morphology of colonies. More importantly, however, relying on human judgment introduces a significant source of bias and technical variability, especially when such protocols are distributed among multiple researchers and research groups. As a result of this lack of standardization, monocloning cannot be reliably upscaled without exacerbating technical variability in cell lines. All of these factors make monocloning a highly desirable target for automation that would allow for the infinite scalability of colony selection protocols and their large-scale distribution while minimizing technical variability.

[0007] Deep learning based on the use of convolutional neural networks (CNNs) has enabled significant advancements in computer vision over the past few years, becoming an invaluable tool for automating the analysis of various types of biomedical images. These techniques have already been applied to numerous processes in SC research, including the automation of differentiation inference and function prediction in iPSC-derived cell types. However, CNNs have never been used to automatically identify clonality during monocloning protocols for any cell type.

[0008] Deep learning models often rival or surpass the image analysis performance of human researchers in domain-specific tasks. Dedicated neural network architectures exist for specific tasks such as image classification and segmentation. Specifically, detection networks trained to detect and localize each instance of a given object class within an image clearly offer a promising opportunity for automated monoclonality verification that ultimately relies on counting individual cells. Embodiments of detection networks in other scientific endeavors have been shown to be very successful. These typically adhere to standardized procedures for training and inference; such procedures involve annotating images with object bounding boxes for training; followed by fitting the labeled data through predefined network architectures such as region-based convolutional neural networks (RCNNs) and you only look once (YOLO).

[0009] Due to several key nuances inherent in monocloning, the task is resistant to automation through standardized and widely adopted deep learning practices. For example, identifying monoclonal wells requires counting individual initiating cells. These typically occupy <0.01% of the well's field of view, and are often too small for human researchers to see without manually zooming in on the image to pinpoint the cell's exact location. Grayscale imaging typically presents significant noise, exacerbating this difficulty. Debris particles are very often subjectively indistinguishable from initiating cells, and researchers frequently rely on information in later images, such as growth, to determine whether a particular particle is a cell or a non-living artifact.

[0010] Regardless of the above, clonality verification inevitably depends on the interaction between images taken at different points in time. For example, the step of counting individual cells in the day 0 image to validate that the sorting process was successful and that exactly one initiating cell was isolated provides no information about whether the cell subsequently survived, proliferated, or retained the desired morphological traits. Conversely, the step of validating that only a single colony is visible at the time of examination is insufficient to confirm monoclonality, because a single polyclonal cell mass may arise from multiple initiating cells, even if it superficially resembles a monoclonal colony. In short, as far as human researchers can assess, there is no case in which a single image can contain all the information necessary to infer the clonality of a well. For this reason, it is not feasible to simply construct conventional training sets consisting of images and their corresponding semantic labels.

[0011] iPSCs are an attractive cell source for therapeutic applications, medical research, and pharmaceutical testing. However, to meet these and other needs, there has long been a need in the art for automated systems to rapidly generate and isolate reproducible iPSC cell lines under standard conditions. [Overview of the project]

[0012] This disclosure provides a system and method for image analysis based on a computational workflow, hereafter referred to herein as "Monoqlo" or the system of the present invention, which integrates a trained neural network. While the system and method are applicable to the generation and analysis of many types of images, in one aspect they are useful for identifying and analyzing living cells, such as determining cellular characteristics such as physical attributes, clonality, karyotype, phenotype, anomalous properties, and disease status.

[0013] Accordingly, in one embodiment, the present invention provides an imaging system, which includes an imaging device and a controller functionally connected to the imaging device, the controller being operable for generating an image via the imaging device and for analyzing the generated image via a processor. In various aspects, the processor includes functionality for performing one or more of the following operations: i) generating a plurality of time-series images of an image area via the imaging device; ii) identifying a target object within the image area of ​​the most recent image among the plurality of time-series images; iii) generating a target object image area within the image area of ​​the most recent image containing the identified target object, wherein the target object area has a perimeter within the image area of ​​the most recent image; iv) using a preceding image of the image area and cropping the preceding image to generate a cropped image area sized to the perimeter of the target object image area; v) generating a location region of the cropped image area within the image area of ​​the most recent image; and optionally, vi) analyzing the location region of the most recent image.

[0014] In another embodiment, the present invention provides a method for performing image analysis. The method includes the steps of identifying a target object in an image using the system of the present invention, and optionally analyzing it. In some aspects, the step of analyzing the target object includes classifying the target object based on its attributes, such as physical characteristics of the target object, including size and / or shape. In some aspects, the target object is a cell or cell colony, and the physical attributes are morphological characteristics of the cell, such as size and / or shape. In some aspects, the attributes are cellular characteristics such as clonality, karyotype, phenotype, anomalousity, and / or disease state.

[0015] In yet another embodiment, the present invention provides an automated system for generating iPSCs or differentiated cells from iPSCs or SCs. The system includes: a) an induction unit for automated reprogramming of iPSCs or differentiation of SCs or iPSCs, the induction unit being operable to bring cells into contact with reprogramming factors or differentiation factors; b) an imaging system operable to identify iPSCs or differentiated cells, the imaging system including a non-transient computer-readable medium having instructions for identifying monoclonal or polyclonal cell populations; and optionally, c) a sorting unit for isolating the identified cells. In several aspects, one or more CNNs are used to process images taken by the imaging system of cells generated in a) and cultured for a certain duration, thereby creating a set of cell images to identify monoclonal or polyclonal cell populations.

[0016] In another embodiment, the present invention provides an automated method for generating iPSCs or differentiated cells from iPSCs or SCs. The method comprises the steps of: a) generating iPSCs or differentiated cells from SCs or iPSCs; b) identifying iPSCs or differentiated cells using an imaging system comprising a non-transient computer-readable medium having instructions for identifying monoclonal or polyclonal cell populations; and optionally, c) isolating monoclonal or polyclonal cells via a sorting unit. In several aspects, one or more CNNs are used to process images taken by the imaging system of cells generated in a) and cultured for a certain duration, thereby creating a set of cell images to identify monoclonal or polyclonal cell populations.

[0017] In another embodiment, the present invention provides a non-temporal computer-readable medium having instructions for identifying monoclonal or polyclonal cell populations. In various aspects, the non-temporal computer-readable medium is electronically linked to an imaging system.

[0018] In yet another embodiment, the present invention provides a method for determining the clonality of a cell population. The method comprises the steps of: a) culturing cells for a certain duration to generate a cell population; and b) analyzing the cell population for the duration using an imaging system electronically coupled to a non-temporal computer-readable medium of the present invention, thereby determining whether the cell population is monoclonal or polyclonal.

[0019] The present invention also provides an automated system for analyzing cells or cell populations. The system includes: a) a cell culture unit for culturing cells or cell populations; b) an imaging system operable for analyzing cells or cell populations, comprising a non-transient computer-readable medium having instructions for identifying morphological features of cells or for identifying monoclonal or polyclonal cell populations; and optionally, c) a sorting unit for isolating cells of interest from the cell culture unit.

[0020] In yet another embodiment, the present invention provides an automated method for analyzing cells or cell populations. The method comprises the steps of: a) culturing cells or cell populations; b) analyzing cells or cell populations using an imaging system comprising a non-transient computer-readable medium having instructions for trained identification of morphological features of cells or identification of monoclonal or polyclonal cell populations; and optionally, c) isolating cells of interest from the cultured cells.

[0021] In another embodiment, the present invention provides a method comprising the steps of: a) culturing cells in a sample well; and b) analyzing the cells using an imaging system of the present invention, wherein the target object is a cell.

[0022] In yet another embodiment, the present invention provides an automated method for generating iPSCs or differentiated cells from iPSCs or SCs. The method comprises the steps of: a) generating iPSCs or differentiated cells from SCs or iPSCs; b) identifying iPSCs or differentiated cells using the imaging system of the present invention, wherein a controller identifies monoclonal or polyclonal cell populations; and optionally, c) isolating monoclonal or polyclonal cells via a sorting unit.

[0023] In another aspect, the present invention provides a method for determining the clonality of a cell population. The method includes the following steps: a) culturing cells for a certain duration to generate a cell population; and b) analyzing the cell population over the duration using the imaging system of the present invention, wherein a controller identifies a monoclonal or polyclonal cell population, thereby determining whether the cell population is monoclonal or polyclonal.

[0024] In yet another aspect, the present invention provides an automated system for analyzing a cell or a cell population. The system includes the following: a) a cell culture unit for culturing the cell or the cell population; b) the imaging system of the present invention, operable by a controller to analyze the cell or the cell population by identifying morphological characteristics of the cells or identifying a monoclonal or polyclonal cell population; and optionally, c) a sorting unit for isolating cells of interest from the cell culture unit.

[0025] In another aspect, the present invention provides an automated method for analyzing a cell or a cell population. The method includes the following steps: a) culturing the cell or the cell population; b) analyzing the cell or the cell population using the imaging system of the present invention, wherein a controller is operable to analyze the cell or the cell population by identifying morphological characteristics of the cells or identifying a monoclonal or polyclonal cell population; and optionally, c) isolating cells of interest from the cultured cells. [The present invention 1001] An imaging system comprising the following: (a) an imaging device, and (b) a controller functionally connected to the imaging device, wherein the controller is operable to generate an image via the imaging device and analyze the generated image via a processor, the processor being (i) Generating a plurality of time-lapse images of an image area via the imaging device; (ii) Identifying a target object within the image area of the most recent image among the plurality of time-lapse images; (iii) Generating a target object image area within the image area of the most recent image including the identified target object, wherein the target object area has an outer periphery within the image area of the most recent image; (iv) Using a previous image of the image area, cropping the previous image to generate a cropped image area sized to match the size of the outer periphery of the target object image area; (v) Generating a location area of the cropped image area within the image area of the most recent image; and (vi) Analyzing the location area of the most recent image including functionality for performing a controller. [Invention 1002] The system of Invention 1001, wherein (i) to (vi) are repeated for each consecutive image of the plurality of time-lapse images. [Invention 1003] The system of Invention 1002, wherein the plurality of time-lapse images include 10, 100, 1,000, 10,000, more than 100,000 or more images. [Invention 1004] The system of Invention 1002, wherein (i) to (vi) are repeated when only one target object is identified within the image area. [Invention 1005] The system of Invention 1001, further including identifying a target object within the location area of the most recent image. [Invention 1006] The system of Invention 1005, further including analyzing the target object. [Invention 1007] The system of Invention 1006, wherein analyzing the target object includes classifying the target object based on an attribute of the target object. [Invention 1008] A system according to the present invention 1007, wherein the attributes are physical characteristics of the target object. [Invention 1009] A system of the present invention 1008, wherein the physical characteristic is size or shape. [Invention 1010] The system of the present invention 1001, wherein (i) to (vi) are performed via one or more convolutional neural networks (CNNs). [Invention 1011] A system according to the present invention 1001, wherein the target object is a cell or a cell colony. [Invention 1012] The system of the present invention 1011, wherein the cells of the cell colony are monoclonal. [Invention 1013] A method for performing image analysis, comprising the steps of identifying a target object in an image using any of the systems described in 1001 to 1012 of the present invention, and optionally analyzing the object therein. [Invention 1014] The method of the present invention 1013, wherein the step of analyzing a target object includes classifying the target object based on the attributes of the target object. [Invention 1015] The method of the present invention 1014, wherein the attribute is a physical characteristic of the target object. [Invention 1016] The method of the present invention 1015, wherein the physical feature is size or shape. [Invention 1017] The method of the present invention 1015, wherein the target object is a cell or cell colony, and the physical attribute is a morphological feature of a cell. [Invention 1018] Automated systems for generating induced pluripotent stem cells (iPSCs) or differentiated cells from iPSCs or stem cells (SCs), including the following: (a) An induction unit for automatically reprogramming iPSCs or differentiating SCs or iPSCs, the induction unit being operable to bring cells into contact with reprogramming factors or differentiation factors; (b) an imaging system operable for identifying iPSCs or differentiated cells, comprising a non-transient computer-readable medium having instructions for identifying monoclonal or polyclonal cell populations; and optionally, (c) Sorting unit for isolating identified cells. [Invention 1019] The system of the present invention 1018, wherein one or more CNNs are used to process images of cells generated in (a) and cultured for a certain duration, which are captured by an imaging system, thereby creating a set of images of the cells to identify monoclonal or polyclonal cell populations. [Invention 1020] A system according to the present invention 1019, wherein images are processed in a time-series manner. [Invention 1021] The system of the present invention 1020, in which each image is assigned a distributional timestamp. [Invention 1022] The system of the present invention 1021 further includes the step of categorizing a set of images based on the morphological characteristics of cells. [Invention 1023] The system of the present invention 1022 further includes the step of classifying cells as polyclonal or monoclonal based on categorization. [Invention 1024] The system of the present invention 1024 further comprises the step of isolating cells classified as monoclonal via a sorting unit. [Invention 1025] The system of the present invention 1024, wherein sorting is optionally performed via cell dispensing technology. [Invention 1026] The system of the present invention 1019, wherein the image set includes 1, 10, 100, 1,000, 10,000, 15,000, 20,000, 25,000, 30,000, 50,000, or more than 100,000 images. [Invention 1027] An automated method for generating iPSCs or differentiated cells from iPSCs or SCs, including the following steps: (a) The stage of generating iPSCs or generating differentiated cells from SCs or iPSCs; (b) Identifying the iPSC or differentiated cells using an imaging system that includes a non-transient computer-readable medium having instructions for identifying monoclonal or polyclonal cell populations; and optionally, (c) The step of isolating the monoclonal or polyclonal cells via a sorting unit. [Invention 1028] The method of the present invention 1027, wherein one or more CNNs are used to process images of cells generated in (a) and cultured for a certain duration, captured by an imaging system, thereby creating a set of images of the cells to identify monoclonal or polyclonal cell populations. [Invention 1029] The method of the present invention 1028, wherein images are processed in a time-series manner. [Invention 1030] The method of the present invention 1029, wherein each image is assigned a distributional timestamp. [Invention 1031] The method of the present invention 1030, further comprising the step of categorizing a set of images based on the morphological characteristics of cells. [Invention 1032] The method of the present invention 1031, further comprising the step of classifying cells as polyclonal or monoclonal based on categorization. [Invention 1033] The method of the present invention 1032 further comprises the step of isolating cells classified as monoclonal via a sorting unit. [Invention 1034] The method of the present invention 1027, wherein sorting is optionally performed via cell dispensing technology. [Invention 1035] The method of the present invention 1027, wherein the image set includes more than 10,000, 15,000, 20,000, 25,000, or 30,000 images. [Invention 1036] A non-temporary, computer-readable medium containing instructions for identifying monoclonal or polyclonal cell populations. [Invention 1037] A non-temporary computer-readable medium according to the present invention 1036, electronically connected to an imaging system. [Invention 1038] The non-temporary computer-readable medium of the present invention 1037 comprises a step of generating a set of images of cells cultured over a certain duration via an imaging system, wherein the set has a plurality of individual images. [Invention 1039] A non-temporary computer-readable medium according to Invention 1038, wherein the image set includes more than 10,000, 15,000, 20,000, 25,000, or 30,000 images. [Invention 1040] A non-temporary computer-readable medium according to the present invention 1038, wherein individual images are captured in a chronological manner and assigned chronological timestamps. [Invention 1041] A non-temporal computer-readable medium according to the present invention 1040, wherein the instructions define a step of processing a set of images in chronological order using one or more CNNs. [Invention 1042] A non-temporary computer-readable medium according to the present invention 1041, wherein the instructions define a step of categorizing a processed set of images based on the morphological characteristics of cells. [Invention 1043] A non-temporary computer-readable medium according to the present invention 1042, wherein the instructions define a step of classifying cells as polyclonal or monoclonal based on categorization. [Invention 1044] A non-transient computer-readable medium according to Invention 1043, wherein the instructions define the steps for isolating cells classified as monoclonal or polyclonal. [Invention 1045] A method for determining the clonality of a cell population, including the following steps: (a) The step of culturing cells for a certain period of time to generate a cell population; and (b) Analyzing the cell population over the duration using an imaging system electronically coupled to a non-temporary computer-readable medium according to any of the invention items 1036 to 1044, thereby determining whether the cell population is monoclonal or polyclonal. [Invention 1046] Automated systems for analyzing cells or cell populations, including the following: (a) A cell culture unit for culturing cells or a population of cells; (b) an imaging system operable for analyzing the cells or cell populations, comprising a non-transient computer-readable medium having instructions for identifying the morphological features of cells or for identifying monoclonal or polyclonal cell populations; and optionally, (c) A sorting unit for isolating cells of interest from the cell culture unit. [Invention 1047] The system of the present invention 1046, wherein one or more CNNs are used to process images of cells cultured in (a) and cultured for a certain duration, thereby creating a set of time-series images of the cells over time, and identifying monoclonal or polyclonal cell populations. [Invention 1048] The system of the present invention 1046, wherein one or more CNNs are used to process images of cells cultured in (a) and cultured for a certain duration, which are captured by an imaging system, thereby creating a set of time-series images of the cells over time, and the morphological features of the cells are identified and analyzed. [Invention 1049] An automated method for analyzing cells or cell populations, including the following steps: (a) The step of culturing cells or a population of cells; (b) the step of analyzing the cells or cell population using an imaging system which includes a non-temporary computer-readable medium having instructions for trained identification of the morphological features of cells or identification of monoclonal or polyclonal cell populations; and optionally, (c) The step of isolating cells of interest from the cultured cells. [Invention 1050] The method of the present invention 1049, wherein one or more CNNs are used to process images of cells cultured in (a) and cultured for a certain duration using an imaging system, thereby creating a set of time-series images of the cells over time to identify monoclonal or polyclonal cell populations. [Invention 1051] The method of the present invention 1049, wherein one or more CNNs are used to process images of cells cultured in (a) and cultured for a certain duration using an imaging system, thereby creating a set of time-series images of the cells over time, and identifying and analyzing their morphological features. [Invention 1052] (a) The step of culturing cells in a sample well; and (b) A step of analyzing the cell using any imaging system 1001 to 1012 of the present invention, wherein the target object is the cell. A method that includes this. [Invention 1053] An automated method for generating iPSCs or differentiated cells from iPSCs or SCs, including the following steps: (a) The stage of generating iPSCs or generating differentiated cells from SCs or iPSCs; (b) A step of identifying the iPSC or differentiated cells using any imaging system 1001 to 1012 of the present invention, wherein the controller identifies a monoclonal or polyclonal population of cells; and optionally, (c) The step of isolating the monoclonal or polyclonal cells via a sorting unit. [Invention 1054] A method for determining the clonality of a cell population, including the following steps: (a) The step of culturing cells for a certain period of time to generate a cell population; and (b) A step of analyzing the cell population over a duration using an imaging system according to any of the present invention 1001 to 1012, wherein a controller identifies monoclonal or polyclonal cell populations, thereby determining whether the cell population is monoclonal or polyclonal. [Invention 1055] (a) A cell culture unit for culturing cells or a population of cells; (b) any imaging system according to item 1001 to 1012 of the present invention, wherein the controller is operable to analyze the cell or cell population by identifying the morphological characteristics of the cell or by identifying a monoclonal or polyclonal cell population; and optionally, (c) Sorting unit for isolating cells of interest from the cell culture unit. An automated system for analyzing cells or cell populations, including [specific data / features]. [Invention 1056] An automated method for analyzing cells or cell populations, including the following steps: (a) The step of culturing cells or a population of cells; (b) A step of analyzing the cells or cell population using any imaging system of the present invention 1001 to 1012, wherein the controller is operable to analyze the cells or cell population by identifying the morphological characteristics of the cells or by identifying monoclonal or polyclonal cell populations; and optionally, (c) The step of isolating cells of interest from the cultured cells. [Invention 1057] The system of the present invention 1012 further includes a cell isolation module for isolating monoclonal cells. [Invention 1058] The system of the present invention 1057 further comprises a protein isolation module for isolating proteins produced by monoclonal cells. [Invention 1059] The system of the present invention 1025, wherein the cell dispensing technology is FACS. [Invention 1060] The method of the present invention 1034, wherein the cell dispensing technique is FACS. [Invention 1061] Automated systems for analyzing cells or cell populations, including the following: (a) A cell culture unit for culturing cells or a population of cells; (b) an imaging system operable for analyzing the cells or cell populations, comprising a non-transient computer-readable medium having instructions for identifying the characteristics of the cells or cell populations; and optionally, (c) A sorting unit for isolating cells of interest from the cell culture unit. [Invention 1062] The system of the present invention 1061, wherein one or more CNNs are used to process images of cells cultured in (a) and cultured for a certain duration, which are captured by an imaging system, thereby creating a set of time-series images of the cells over time and identifying their characteristics. [Invention 1063] A system according to either invention 1061 or 1062, wherein the characteristics are morphological features, clonality, karyotype, phenotype, anomalousity, and / or disease state. [Invention 1064] An automated method for analyzing cells or cell populations, including the following steps: (a) The step of culturing cells or a population of cells; (b) the step of analyzing the cells or cell population using an imaging system which includes a non-temporary computer-readable medium having instructions for performing trained identification of the characteristics of the cells or cell population; and optionally, (c) The step of isolating cells of interest from the cultured cells. [Invention 1065] The method of the present invention 1064, wherein one or more CNNs are used to process images of cells cultured in (a) and cultured for a certain duration using an imaging system, thereby creating a set of time-series images of the cells over time and identifying their characteristics. [Invention 1066] A method of the present invention 1064 or 1065, wherein the characteristics are morphological features, clonality, karyotype, phenotype, abnormality, and / or disease state. [Invention 1067] (a) The step of culturing cells in a sample well; and (b) A step of analyzing the cell using any imaging system 1001 to 1012 of the present invention, wherein the target object is the cell. A method that includes this. [Invention 1068] The method of the present invention 1068, wherein the analysis step includes using one or more CNNs to process images of cells cultured over a certain period of time, captured by an imaging system, thereby creating a set of time-series images of the cells over time and determining the characteristics of the identified cells. [Invention 1069] A method of the present invention 1067 or 1068, wherein the characteristics are morphological features, clonality, karyotype, phenotype, abnormality, and / or disease state. [Brief explanation of the drawing]

[0026] [Figure 1A] The images show a portion of four CNN modules used in one embodiment of the present invention. The images also show simple schematic diagrams of two neural network architectures used for detection and classification tasks. [Figure 1B] The images show a portion of four CNN modules used in one embodiment of the present invention. The functionality of each of the three detection modules, along with representative target data and outputs, is shown. [Figure 1C]The images show portions of four CNN modules used in one embodiment of the present invention. These are examples of four target morphological classes used in training the morphological classification network of the present invention. [Figure 2] This is a series of images illustrating a schematic daily automated workflow for generating data for training and real-time use in one aspect of the present invention. After cell deposition via FACs, the cells are grown for N days, with well-level imaging occurring nightly. N is a variable that depends on decisions regarding cell growth rate and passage timing. [Figure 3] This is a schematic diagram illustrating a rough outline of the design and algorithmic logic used in one embodiment of the present invention. The arrows indicate the processing order in the reverse time-series analysis of this algorithm, starting from the most recent scan. If a colony is detected, the region around the colony is extracted from the previous day's scan, and the image is passed to the local detection model. The process is repeated, progressively reducing the field of view being analyzed. If multiple colonies are detected in a scan, the well is declared polyclonal, and subsequent scans are not analyzed. Upon reaching the earliest "day 0" scan, the resulting image is passed to the local detection model. Based on the number of cells detected, the clonality of the well is finally declared. [Figure 4]Figure 4A is an image showing the validation results generated by the present invention. Illustrated is the well-level clonality identification performance of the present invention framework on real-world fabricated run data. The outer color represents the clonality of ground-trussed wells, the meaning of which is shown in the legend; the inner color represents the clonality identified by the present invention, and therefore wells with double coloring represent errors. Figure 4B is an image showing the validation results generated by the present invention. Illustrated is the class-specific clonality identification performance of the present invention on a manually curated and class-balanced test dataset. Figure 4C is an image showing the validation results generated by the present invention. Illustrated is an overview of the clonality performance of the present invention when the analysis is limited to morphologically healthy monoclonal wells selected by biologists for further passage. [Figure 5] Figure 5A is a graph that provides an overview of the training and performance of a classification model. It shows the trajectory of the accuracy of training and validation of the classification CNN, plotted against epochs. Figure 5B is a graph that provides an overview of the training and performance of a classification model. It shows the confusion matrix when a fully trained classification CNN is tested with a holdout validation set. [Figure 6] This graph shows the relationship between the width of the colony bounding box predicted by the global detection model of the present invention and the true width measured by biologists using an image overlay of a scale bar. [Figure 7A] This image shows an example of non-biological artifacts that cause false colony detection by the global detection model of the present invention. The image represents the entirety of the image report generated by the present invention. [Figure 7B] This image shows an example of non-biological artifacts that cause false colony detection by the global detection model of the present invention. The image is a zoomed-in version of the same image report shown in Figure 7A. [Figure 8] This image shows an example of a colony overlapping report using the local detection model of the present invention, where only a single colony remains after ground trusing. [Figure 9] This image shows an example of a colony overlapping report using the local detection model of the present invention, where only a single colony remains after ground trusing. [Figure 10] This image shows an example of a colony overlapping report using the local detection model of the present invention, where only a single colony remains after ground trusing. [Figure 11] This image illustrates the concept of "colony splitting," where a seemingly single colony was revealed through reverse time-series analysis to have originated from multiple colonies that eventually merged. [Figure 12] This is a series of graphs representing a gating strategy used during FACS sorting single cloning of iPSCs in one embodiment of the present invention. [Modes for carrying out the invention]

[0027] Detailed description of the invention The present invention is based on innovative systems and methods for image analysis. Before describing the composition and methods of the present invention, it should be understood that the present invention is not limited to the specific systems, methods, and / or experimental conditions described herein; for such systems, methods, and conditions may vary. Since the present invention is limited only to the appended claims, it should also be understood that the technical terms used herein are for the purpose of describing specific aspects only and are not intended to limit the present invention.

[0028] In this specification and the appended claims, the singular forms “a,” “an,” and “the” also include references to the plural unless otherwise explicitly stated in the context. Thus, for example, a reference to “the system” includes one or more systems, and a reference to “the method” includes one or more methods and / or steps of the type described herein, which will be apparent to those skilled in the art upon reading this disclosure.

[0029] This disclosure provides an imaging system and method for analyzing imaged objects, utilizing a computational workflow that integrates multiple CNNs. In several aspects, the present invention is based on a system and computational design that overcomes known challenges by leveraging the chronological directionality inherent in the cell culture process. The system and computational methodology described herein, called Monoqlo, integrates multiple CNNs, each having its own "modular" functionality.

[0030] The present invention encompasses a highly scalable framework that enables the analysis of image datasets of more than 1,000, 10,000, 50,000, 100,000, 500,000, or 1,000,000 images within manageable timeframes of less than 10, 9, 8, 7, 6, 5, 4, 3, 2, or 1 hour. It will be recognized that the functionality described herein may be applied to any number of conventional imagers. As detailed in Example I, the research results described herein demonstrate the first application of machine learning to the identification of monoclonal cell lines from bright-field microscopy observations.

[0031] While this disclosure illustrates imaging and analysis of living cells, it will be understood that the systems and methods of the present invention are applicable to imaging and subsequent analysis of any target object.

[0032] Accordingly, in one embodiment, the present invention provides an imaging system, the system comprising an imaging device and a controller functionally connected to the imaging device; the controller is operable for generating images via the imaging device and for analyzing the generated images via a processor. In various aspects, the processor includes functionality for performing one or more of the following operations: i) generating a series of time-series images of an image area via the imaging device; ii) identifying a target object within the image area of ​​the most recent image among the series of time-series images; iii) generating a target object image area within the image area of ​​the most recent image containing the identified target object, wherein the target object area has a perimeter within the image area of ​​the most recent image; iv) using a preceding image of the image area and cropping the preceding image to generate a cropped image area sized to the perimeter of the target object image area; v) generating a location region of the cropped image area within the image area of ​​the most recent image; and optionally, vi) analyzing the location region of the most recent image.

[0033] The present invention further provides a method for performing image analysis using the system of the present invention. The method includes the steps of identifying a target object in an image using the system of the present invention, and optionally analyzing it.

[0034] In some situations, steps i) to vi) are repeated for each sequential image of the multiple time-series images. In some situations, steps i) to vi) are repeated when only one target object is identified within the image area.

[0035] As discussed herein, the present invention allows for the analysis of image datasets of various sizes within manageable timeframes. In some aspects, a dataset, for example, a set of images over time, may include 1, 10, 100, 1,000, 10,000, 100,000, 200,000, 300,000, 400,000, 500,000, 600,000, 700,000, 800,000, 900,000, or more than 1,000,000 images.

[0036] As will be further discussed herein, the functionality described herein may be applied to any number of conventional imaging devices. Thus, the generation of images for use with the system and methodology of the present invention may be implemented in various ways and may be analyzed and / or processed using the functionality described herein. In some aspects, the system of the present invention includes one or more imaging devices functionally coupled to a processor, and / or other robotic platform components such as a cell sorting unit, a cell culture unit, an optical analyzer or assembly, a cell reprogramming or differentiation unit, and a cryopreservation unit. The imaging devices used herein include any device or detector capable of capturing images, including but not limited to cameras, microscopes, CCD cameras, photodiodes, photomultiplier tubes, and laser scanners.

[0037] In various scenarios, the system includes functionality for identifying and analyzing target objects within the location area of ​​recent images.

[0038] In some cases, the stage of analyzing target objects involves classifying them based on their attributes. Such attributes may include physical characteristics of the target objects, such as size, shape, and / or color.

[0039] In some contexts, the target object is a cell or cell colony, and the attributes are physical attributes, including morphological features of the cell such as size and / or shape. In some contexts, the attributes are characteristics of the cell or cell colony, such as clonality, karyotype, phenotype, anomalous properties, and / or disease status.

[0040] In various aspects, the systems and methods of the present invention integrate a neural network which may be trained for the analysis and / or classification of specific types of target objects. An overview of the laboratory automation workflow for generating the data is shown in Figures 2 and 3, respectively. The algorithm processes images of image areas that typically contain target objects in a reverse-chronological manner. That is, for each image area, the algorithm begins by analyzing the most recently generated scan. In the inventors' case, this is an image that has been cropped to remove the black borders of the image while retaining the entire field of view of the image area. These images are passed to a global detection model; the output of the global detection model is a coordinate vector that defines the bounding box of the detected target object.

[0041] The algorithm then expands these coordinates until each dimension of the bounding box is twice that of the predicted target object, loads the next most recent image for that image area, and crops the image to form the resulting region. Because physical positioning is maintained between scans, earlier instantiations of the same target object will be approximately centered in the newly cropped image. This image is then passed to a local detection model; the local detection model reports the bounding box of the earlier target object, indicating its position in the original uncropped image when summed with the cropped coordinates. The algorithm inductively iterates this process until the resulting most recent image is the earliest scan ("day 0").

[0042] In several aspects, the training set is stratified based on timestamps and zoom and crop levels, and separate neural networks with their own unique "modular" functionality are trained. Firstly, the term "global detection" is assigned to the task of detecting the presence or absence of a target object within an image area. Secondly, the task of detecting a target object in cropped images of various image areas at various zoom magnifications is called "local detection." Thirdly, the task of counting individual target objects within a fully zoomed cropped image is called "single-cell detection." All three of the aforementioned tasks were aimed to be achieved through the use of a RetinaNet® detection architecture with focal loss (Lin et al., In Proceedings of the IEEE international conference on computer vision; pp. 2980-2988 (2017)). Furthermore, a model was deemed desirable for categorizing images extracted around the colony region into specific classes based on shape and / or size, such as morphological classes of cells; this is referred to herein as "morphological classification."

[0043] As illustrated in Example I, the systems and methodologies of the present invention identify the clonality of cells or cell populations, such as monoclonal or polyclonal cells or cell populations, in various contexts. Monocloning refers to the isolation and proliferation of single cells derived from a cultured population. This is typically done to minimize the technical variability of cell lines downstream of events that modify cells, such as reprogramming or gene editing, and to develop monoclonal antibodies. Due to the lack of automated and standardized methods for retrospectively assessing clonality, methods involved in monocloning cannot reliably upscale without exacerbating the technical variability of cell lines.

[0044] The present invention provides a deep learning workflow for automatically detecting the presence of colonies and identifying clonality from cell imaging. As discussed in Example I, the workflow of the present invention integrates multiple convolutional neural networks and, importantly, leverages the temporal anisotropy of the cell culture process. The systems and methodologies described herein provide a fully scalable and highly interpretable framework that can analyze industrial data volumes in less than an hour using general-purpose hardware. In several aspects, the present invention standardizes the monocloning process, enabling the colony selection protocol to be infinitely upscaled while minimizing technological variability.

[0045] Therefore, in another embodiment, the present invention provides a non-temporary computer-readable medium having instructions for identifying monoclonal or polyclonal cell populations. In various aspects, the non-temporary computer-readable medium is electronically linked to an imaging system.

[0046] In some cases, the instructions specify a step of generating a set of images of cells cultured over a certain duration via an imaging system, the set having multiple individual images. In some cases, the individual images are captured in a chronological manner and assigned chronological timestamps. In some cases, the instructions further specify a step of processing the image set in chronological order using one or more CNNs, as well as a step of categorizing the processed image set based on the morphological characteristics of the cells and further classifying the cells as polyclonal or monoclonal based on the categorization.

[0047] In a related embodiment, the present invention further provides a method for determining the clonality of a cell population. The method comprises the steps of: a) culturing cells for a certain duration to generate a cell population; and b) analyzing the cell population for the duration using an imaging system electronically coupled to a non-temporal computer-readable medium of the present invention, thereby determining whether the cell population is monoclonal or polyclonal.

[0048] The present invention also provides an automated system for analyzing cells or cell populations. The system includes: a) a cell culture unit for culturing cells or cell populations; b) an imaging system operable for analyzing cells or cell populations, comprising a non-transient computer-readable medium having instructions for identifying morphological features of cells or identifying monoclonal or polyclonal cell populations; and optionally, c) a sorting unit for isolating cells of interest from the cell culture unit. In several aspects, one or more CNNs are used to process images taken by the imaging system of cells generated in (a) and cultured for a certain duration, thereby creating a time-series image set of cells over time to identify monoclonal or polyclonal cell populations. In several aspects, one or more CNNs are used to process images taken by the imaging system of cells generated in (a) and cultured for a certain duration, thereby creating a time-series image set of cells over time to identify and analyze morphological features.

[0049] In yet another embodiment, the present invention provides an automated method for analyzing cells or cell populations. The method comprises the steps of: a) culturing cells or cell populations; b) analyzing cells or cell populations using an imaging system comprising a non-transient computer-readable medium having instructions for trained identification of morphological features of cells or identification of monoclonal or polyclonal cell populations; and optionally, c) isolating cells of interest from the cultured cells. In several aspects, one or more CNNs are used to process images taken by the imaging system of cells generated in (a) and cultured for a certain duration, thereby creating a time-series image set of cells over time to identify monoclonal or polyclonal cell populations. In several aspects, one or more CNNs are used to process images taken by the imaging system of cells generated in (a) and cultured for a certain duration, thereby creating a time-series image set of cells over time to identify and analyze morphological features.

[0050] As will be apparent from this disclosure, the present invention is useful in the generation of iPSCs or differentiated cells where the identification and / or classification of monoclonal cell populations is desirable. Thus, in one embodiment, the present invention provides an automated system for generating iPSCs or for generating differentiated cells from iPSCs or SCs. The system includes: a) an induction unit for automated reprogramming of iPSCs or differentiation of SCs or iPSCs, the induction unit being operable to bring cells into contact with reprogramming factors or differentiation factors; b) an imaging system operable to identify iPSCs or differentiated cells, the imaging system including a non-transient computer-readable medium having instructions for identifying monoclonal or polyclonal cell populations; and optionally, c) a sorting unit for isolating the identified cells.

[0051] In another embodiment, the present invention provides an automated method for generating iPSCs or differentiated cells from iPSCs or SCs. The method comprises the steps of: a) generating iPSCs or differentiated cells from SCs or iPSCs; b) identifying iPSCs or differentiated cells using an imaging system comprising a non-transient computer-readable medium having instructions for identifying monoclonal or polyclonal cell populations; and optionally, c) isolating monoclonal or polyclonal cells via a sorting unit.

[0052] In several aspects, one or more CNNs are used to process images of cells generated in a) and cultured for a certain duration, thereby creating a set of cell images to identify monoclonal or polyclonal cell populations.

[0053] In some cases, cell sorting is achieved by cell dispensing or cell sorting techniques, which may optionally include flow cytometry. For example, cells may be sorted using single-cell sorting, fluorescence-activated cell sorting (FACS), and / or magnetic-activated cell sorting (MACS).

[0054] As used herein, "adult" refers to organisms from the post-fetal period, for example, from the neonatal period to the terminal stage, and includes, for example, cells obtained from delivered placental tissue, amniotic fluid, and / or umbilical cord blood.

[0055] As used herein, the term “adult differentiated cell” encompasses a wide range of differentiated cell types obtained from adult organisms that are well-suited for iPSC production using the automated system described later. Preferably, the adult differentiated cell is a “fibroblast.” Fibroblasts, also known as “fibrocytes” in their less active form, are derived from the mesenchyme. Functions of fibroblasts include, for example, secreting precursors of extracellular matrix components, including collagen. Histologically, fibroblasts are highly branched cells, while fibrocytes are generally smaller and often described as spindle-shaped. Fibroblasts and fibrocytes derived from any tissue may be used as starting materials for the automated workflow system of the present invention.

[0056] As used herein, the terms "induced pluripotent stem cell" or iPSC refer to stem cells that have been induced or modified, for example through reprogramming, to become cells capable of differentiating into all three germ layers or cortical tissues: mesoderm, endoderm, and ectoderm. Since these iPSCs are not found in nature, the term "cell" does not apply to them.

[0057] As used herein, the terms “stem cell” or “undifferentiated cell” refer to cells in an undifferentiated or partially differentiated state that possess the characteristic of self-renewal and the developmental capacity to differentiate into multiple cell types, without any specific implication regarding developmental capacity (e.g., totipotent, pluripotent, multipotent). Stem cells can proliferate and produce more such stem cells while maintaining developmental capacity. Theoretically, self-renewal can occur through one of two main mechanisms. Stem cells can undergo asymmetrical differentiation, known as obligatory asymmetrical differentiation; in this case, one daughter cell retains the developmental capacity of the parent stem cell, while the other daughter cell expresses other specific functions, phenotypes, and / or developmental capacity that are somewhat different from the parent cell. Daughter cells can be induced to proliferate and produce offspring; these offspring then differentiate into one or more mature cell types, while also retaining one or more cells with the parent's developmental potential. Differentiated cells may originate from pluripotent cells, and the pluripotent cells themselves may also originate from pluripotent cells. Each of these pluripotent cells can be considered a stem cell, but the cell types that can arise from each of these stem cells, such as the range of developmental potential, can vary considerably. Alternatively, some stem cells in a population may undergo symmetrical division into two stem cells, known as stochastic differentiation; thus, some stem cells in the population as a whole are maintained, while other cells in the population produce only differentiated offspring. Therefore, the term “stem cell” refers to any subset of cells that, under certain circumstances, have the developmental potential to differentiate into more specialized or differentiated phenotypes, and under certain circumstances retain the ability to proliferate substantially without differentiation.In some embodiments, the term stem cell generally refers to naturally occurring parental cells; their offspring (progeny) are specialized through differentiation, often in different directions, acquiring entirely distinct characteristics, such as those that occur in the gradual diversification of embryonic cells and tissues. Some differentiated cells also possess the ability to produce cells with greater developmental potential. Such ability may occur spontaneously or may be artificially induced by treatment with various factors. Cells that begin as stem cells may progress to a differentiated phenotype, but can then be “reversed” to re-express the stem cell phenotype, which is often referred to by those skilled in the art as “dedifferentiation,” “reprogramming,” or “retrodifferentiation.”

[0058] The term “differentiated cell” encompasses any somatic cell that is not pluripotent in its native form; the term “pluripotency” is as defined herein. Therefore, the term “differentiated cell” also encompasses partially differentiated cells, such as pluripotent cells, partially reprogrammed stable and non-pluripotent cells, or partially differentiated cells produced using any of the compositions and methods described herein. In some embodiments, differentiated cells are stable intermediate cells, such as partially reprogrammed non-pluripotent cells. Transitioning differentiated cells (including partially reprogrammed stable and non-pluripotent cell intermediates) to pluripotency requires a reprogramming stimulus that outweighs the stimulus that would lead to a partial loss of differentiation characteristics when placed in culture. Reprogrammed cells, and in some embodiments partially reprogrammed cells, also possess the characteristic of being able to be passaged for a longer period without relative loss of growth capacity compared to parent cells, which have lower developmental capacity and generally only a limited number of divisions in culture. In some aspects, the term “differentiated cell” also refers to a more specialized cell type (e.g., with lower developmental potential) that has been derived from a less specialized cell type (e.g., from undifferentiated or reprogrammed cells) through the cell differentiation process.

[0059] As used herein, the term “reprogramming” refers to the process of reversing the developmental potential of a cell or a population of cells (e.g., somatic cells). In other words, reprogramming refers to the process of bringing a cell to a state of higher developmental potential, such as returning it to a less differentiated state. The cells being reprogrammed may be partially differentiated or fully differentiated before reprogramming. In some aspects of the aspects described herein, reprogramming includes completely or partially reversing the differentiation state to a differentiated state of a cell having a pluripotent state, such as increasing the cell’s developmental potential. In some aspects, reprogramming includes bringing somatic cells to a pluripotent state so that the cells have the developmental potential of embryonic stem cells, such as the phenotype of embryonic stem cells. In some aspects, reprogramming also includes partially reversing the differentiation state of a cell, such as a somatic cell or a unipotent cell, or partially increasing its developmental potential to make it multipotent. Reprogramming also includes partially reversing the differentiation state of a cell so that it is more likely to undergo complete reprogramming to a pluripotent state when subjected to additional operations as described herein. As a result of such manipulation, certain genes that, when expressed, contribute to the maintenance of reprogramming may be endogenously expressed by the cell or its offspring. In certain embodiments, the reprogramming of cells using synthetically modified RNA as described herein induces a pluripotent state in the cell (e.g., the cell becomes a pluripotent cell). In some embodiments, the reprogramming of cells (e.g., somatic cells) using synthetically modified RNA as described herein induces a pluripotent state or an embryonic stem cell phenotype in the cell. The resulting cells are referred to herein as “reprogrammed cells,” “somatic pluripotent cells,” and “RNA-induced somatic pluripotent cells.”The term “partially reprogrammed somatic cell” as used herein refers to a cell that has been reprogrammed from a cell with lower developmental potential by the methods disclosed herein, such that it is reprogrammed to a non-pluripotent and stable intermediate state rather than being fully reprogrammed to a pluripotent state. Such partially reprogrammed cells may have developmental potential lower than that of pluripotent cells but higher than that of multipotent cells; these terms are as defined herein. Partially reprogrammed cells may be able to differentiate into one or two of the three germ layers, for example, but not into all three germ layers.

[0060] As used herein, the term “reprogramming factor” refers to developmental modifiers such as genes, proteins, RNA, DNA, or small molecules whose expression contributes to reprogramming cells, such as somatic cells, into a less differentiated or undifferentiated state, such as a pluripotent or partially pluripotent state; the term developmental modifier is as defined herein. Reprogramming factors may include transcription factors capable of reprogramming cells into a pluripotent state, such as SOX2, OCT3 / 4, KLF4, NANOG, LIN-28, and c-MYC, and also include any genes, proteins, RNA, or small molecules that can be substituted for one or more of these in methods of reprogramming cells in vitro. In some embodiments, exogenous expression of reprogramming factors using the synthetically modified RNA and methods described herein induces endogenous expression of one or more reprogramming factors to such an extent that exogenous expression of one or more reprogramming factors is no longer necessary to stably maintain cells in a reprogrammed or partially reprogrammed state.

[0061] As used herein, the term “differentiation factor” refers to developmental modifiers, such as proteins, RNAs, or small molecules, that induce cells to differentiate into a desired cell type, for example, a differentiation factor that reduces the developmental potential of a cell; the term “developmental modifier” is as defined herein. In some embodiments, the differentiation factor may, but may not, be a cell-type-specific polypeptide. Differentiation into a particular cell type may require the simultaneous and / or sequential expression of more than one differentiation factor. In some aspects described herein, the developmental potential of a cell or cell population is first increased through reprogramming or partial reprogramming with synthetically modified RNA as described herein, and then the cells or progeny produced by such reprogramming are exposed to or introduced with one or more synthetically modified RNAs encoding a differentiation factor, thereby inducing differentiation of the cell or progeny to have lower developmental potential.

[0062] In the context of cell-organ development, the terms “differentiate” or “differentiating” are relative terms referring to the developmental process in which a cell has progressed further along the developmental pathway than its immediately preceding precursor cell. Thus, in some embodiments, a reprogrammed cell as defined herein may differentiate into a differentiation-restricted precursor cell (such as a mesodermal stem cell), which may then differentiate further down the pathway into other types of precursor cells (such as tissue-specific precursors, such as cardiomyocyte precursors), and then differentiate into a final-stage differentiated cell that plays a characteristic role in a certain tissue type and may or may not retain the ability to proliferate further.

[0063] The present invention includes systems and processors for performing the steps of the methods disclosed herein, and is described in part in terms of functional components and various processing steps. Such functional components and processing steps may be implemented by any number of components, operations, and techniques configured to perform specified functions and achieve various results. For example, various biological samples, biomarkers, elements, materials, computers, data sources, storage systems and storage media, information acquisition techniques and processing, data processing criteria, statistical analysis, and regression analysis, which may perform diverse functions, may be used in the present invention.

[0064] Image analysis methods based on various aspects of the present invention may be implemented in any preferred manner, for example, by using a computer program running on a computer system. Exemplary analysis systems based on various aspects of the present invention may be implemented in association with a computer system, which may be a conventional computer system including, for example, a processor and random access memory, such as a remotely accessible application server, network server, personal computer, or workstation. The computer system also preferably includes additional memory devices or information storage systems, such as a mass storage system, and a user interface, such as a conventional monitor, keyboard, and tracking device. However, the computer system may include any preferred computer system and associated devices, and may be configured in any preferred manner. In one embodiment, the computer system includes a standalone system. In another embodiment, the computer system is part of a computer network including servers and databases.

[0065] The software required to receive, process, and analyze information may be implemented on a single device or on multiple devices. The software may be accessible via a network so that the storage and processing of information are performed remotely to the user. Analysis systems and their various elements based on various aspects of the present invention provide functions and operations to facilitate image analysis, such as data collection, processing, analysis, classification, and / or reporting. For example, in an embodiment of this disclosure, a computer system runs a computer program which may receive, explore, analyze, classify, and / or report information relating to an image, cell, or cell population. The computer program may include multiple modules that perform various functions or operations, such as a processing module for processing raw data and generating supplemental data, and an analysis module for analyzing the raw and supplemental data to generate a quantitative assessment of a target object.

[0066] In this specification and the appended claims, the singular forms “a,” “an,” and “the” include multiple references unless otherwise explicitly stated in the context. The terms “a” (or “an”), as well as “one or more” and “at least one,” are interchangeable.

[0067] Furthermore, “and / or” should be taken as specifically disclosing each of the two specified features or components, with or without the other. Thus, the term “and / or” used in phrases such as “A and / or B” is intended to include A and B, A or B, A (only), and B (only). Similarly, the term “and / or” used in phrases such as “A, B, and / or C” is intended to include A, B, and C; A, B, or C; A or B; A or C; B or C; A and B; A and C; B and C; A (only); B (only); and C (only).

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those widely understood by those skilled in the art relating to the present invention. For example, The Dictionary of Cell and Molecular Biology (5th ed. JM Lackie ed., 2013), Oxford Dictionary of Biochemistry and Molecular Biology (2nd ed. R. Cammack et al. eds., 2008), and The Concise Dictionary of Biomedicine and Molecular Biology, PS. Juo, (2nd ed. 2002) may provide general definitions of some terms used herein to those skilled in the art.

[0069] Units, prefixes, and symbols are expressed in the form recognized by the International System of Units (SI). Numerical ranges include the numbers that define the range. The headings provided herein are not intended to limit the various aspects or aspects of the invention as grasped by referring to this specification as a whole. Thus, the terms defined immediately thereafter are more fully defined by referring to this specification as a whole.

[0070] When a manner is described with the word "comprising," it also includes manners that are otherwise similar, described in terms of "consisting of" and / or "consisting essentially of."

[0071] The following embodiments are provided to further illustrate the advantages and features of the present invention, but are not intended to limit the scope of the invention. These embodiments are typical examples of what may be used, but other procedures, methodologies, or techniques known to those skilled in the art may be used instead. [Examples]

[0072] Example I Modular deep learning enables automated identification of monoclonal cell lines. This embodiment describes a system and computational method for leveraging the temporal anisotropy inherent in the cell culture process. The computational workflow integrates multiple CNNs, each possessing its own unique "modular" functionality.

[0073] The system and methodology of this invention provide a highly scalable framework that enabled the analysis of datasets containing tens of thousands of images in less than an hour. This research demonstrates the first application of machine learning to the identification of monoclonal cell lines from bright-field microscopy observations through a combination of automated stem cell culture and deep learning.

[0074] method Single cloning of hiPSC Destination plates (PerkinElmer #6005182) were pre-coated with 17 µg of Geltrex® LDEV-Free, hESC-Qualified, Reduced Growth Factor Basement Membrane Matrix® (ThermoFisher #A1413302), diluted in 50 µL of DMEM / F12 (ThermoFisher #A1413302), in a 37°C incubator for 1 hour. After incubation, 150 μL of d0 medium consisting of 1x DMEM / F12, 1.5x PSC Freedom® Supplement (ThermoFisher #A27336SA), 1.5x Antibiotic / Antimycotic (ThermoFisher #15240062), and 15% Clone® (Stemcell Technologies #05888) was added to 50 μL of Geltrex® + DMEM / F12 present in the wells and incubated for 1 hour in a 37°C incubator. HiPSC colonies maintained on Geltrex® (all ThermoFisher) in Freedom® PSC medium (FRD1) were dissociated with Accutase® (ThermoFisher #A1110501) for 5-10 minutes at 37°C. Accutase® was quenched with Sort® Buffer (MACS Buffer Miltenyi, containing 10% Clone®), and the cell suspension was pelleted by centrifugation at 130 RCF. The cells were stained with the following antibodies: SSEA4-647: 1:100; BD #560219, Tra-1-60-488: 1:100; BD #560173, CD56-V450: 1:100; BD #560360, CD13-PE: 1:100; BD #555394. After staining, the cells were rinsed by a second centrifugation and resuspended in Sort® Buffer + Propidium Iodide (PI, 1:5000, ThermoFisher #P3566).Next, cells were sorted into pre-prepared destination plates using a FACSARIA-IIu® Cell Sorter (BD Biosciences) with a sheath pressure of 23 psi and a 100 μm ceramic nozzle. The flow cytometry gating strategies used are summarized in Figure 12. For samples sorted using the WOLFSorter®, SYTOX AADvanced® Ready Flow® Reagent (ThermoFisher #R37173) was added to the Sort® Buffer instead of PI.

[0075] Image acquisition and labeling All images were obtained from a repository of historical data derived from the monocloning stage used during the iPSC production process at the NYSCF Global Stem Cell Array®. Previously used for manual verification of clonality, these images are automatically generated once every 24 hours from seeding until plate disposal. All scans generated by the Nexcelom® Celigo cytometer are bright-field images with a resolution of 1 μm per pixel, providing a 7544 x 7544 pixel image after stitching together 16 individual fields of view. The inventors annotated a total of 3,139 images with bounding boxes and object classes. Additionally, 2,224 unannotated images of empty wells were included in the training set as background-only images. During preliminary investigations, it was determined that doing so was crucial in reducing false detection rates. All annotations were generated in Pascal VOC format using LabelImg® software (Tzutalin, 2015). The dataset was augmented by applying random flip and rotation transformations to the images (e.g., Perez & Wang, 2017). The morphological criteria required to categorize each object class were specified by PhD-level biologists specializing in iPSC culture. Annotation was performed by PhD, MS, and BS-level technicians, and all annotations were independently verified by additional researchers.

[0076] Training machine learning models Without pre-trained weights, a RetinaNet® detection model was trained using a Keras RetinaNet® implementation (github.com / fizyr / keras-retinanet) with a ResNet50® convolutional backbone (He et al., In Proceedings of the IEEE conference on computer vision and pattern recognition; pp. 770-778 (2016)). Preprocessing involved subtracting the ImageNet® mean from the image and normalizing the pixel intensity values ​​to a range of 0 to 1. The inventors also implemented a custom algorithm to extract the thick, black edges around the wells from the image; this involved removing the outermost line on each edge of the image and repeating this until the maximum raw pixel intensity value for that line exceeded 70. Each CNN model was trained for 60 epochs, and the weights were saved after each epoch; this allowed us to select the checkpoint with the minimum validation loss as the final model for use in the Monoqlo framework.

[0077] result Modularity of neural networks We modularized the task of automatically assigning cloning to four distinct functionalities made possible by deep learning (Figure 1). The decision to modularize was based on empirical inferences made during preliminary research. Specifically, it was initially thought that, consistent with the principles of transfer learning, the feature extraction capability of a CNN would be best optimized by combining all image types into a single training set. However, networks trained in this manner performed poorly and frequently failed to distinguish between object classes. In particular, they often reported object types that were not feasible in the given image, such as detecting fully developed colonies in images generated immediately after seeding. This suggests that a single model will not be able to achieve good performance across the diverse image scaling and object classes used during single cloning.

[0078] Instead, the training set is stratified based on timestamps and zoom and crop levels, and four separate neural networks, each with its own unique "modular" functionality, are trained. Firstly, the term "global detection" is assigned to the task of detecting the presence or absence of colonies in images showing the entire well. Secondly, the task of detecting colonies in cropped images of various well regions at various zoom magnifications is called "local detection." Thirdly, the task of counting individual cells in a fully zoomed cropped image is called "single-cell detection." All three of the aforementioned tasks were aimed to be achieved through the use of a RetinaNet® detection architecture with focal loss (Lin et al., In Proceedings of the IEEE international conference on computer vision; pp. 2980-2988 (2017)). Furthermore, in the only task in this endeavor that is entirely classification-based, a model for categorizing images cropped around colony regions into morphological classes was considered desirable; this is referred to herein as "morphological classification" (summarized in Figures 5A-5B). This modularization allows the inventors to leverage the temporal directionality of the cell culture process, for example, by limiting the detectable object classes to those that are realistically possible to exist in the image based on the scan date.

[0079] Workflow Design Overview We designed a computational workflow to integrate each of the trained neural networks; we call this Monoqlo. Figures 2 and 3 outline the laboratory automation workflow for generating data for use with Monoqlo, and the design of Monoqlo itself, respectively. The algorithm processes images in a reverse-chronological manner on a well-by-well basis. That is, for each physical well, the algorithm begins by analyzing the most recently generated scan. In our case, this is an image that has been cropped only to remove the black edges of the image, while retaining the entire field of view of the physical well. These images are passed to a global detection model; the output of the global detection model is a coordinate vector that defines the bounding boxes of the detected colonies.

[0080] The algorithm then expands these coordinates until each dimension of the bounding box is twice that of that of the predicted colony, loads the next most recent image for the same well, and crops the image to form the resulting region. Because the plate orientation and physical positioning are maintained between scans, earlier instantiations of the same colony will be approximately centered in the newly cropped image. This image is then passed to a local detection model; the local detection model reports the bounding boxes of the earlier colonies, indicating their position in the original uncropped image when summed with the cropped coordinates. The algorithm inductively iterates this process until the resulting most recent image is the earliest scan ("day 0") generated within hours of sorting. Because growth is non-radial and there are small positional shifts between scans, small deviations from a precisely centered state occur invariably day by day, so this incremental and iterative phase of the workflow has been found essential, along with the expansion of the cropped box dimensions. These deviations over several days of imaging result in a considerable offset in total. Therefore, simply excising and magnifying the area precisely in the center of a late-stage colony rarely yields a field of view in which the initiating cells are located.

[0081] Apart from counting individual initiating cells, if two or more clearly distinct cell clusters are observed, they are often presumed to originate from two or more cells from the same FACS sorting and can be inferred to be polyclonal. If a colony count >1 is reported at any point during the retrospectively iterative processing from either a global or local detection model, the algorithm declares the well polyclonal and stops further image processing for that well. Alternatively, if the workflow continues to detect only one colony until reaching the scan for day 0, the resulting image is magnified and cropped precisely around the ancestral cell. This image may then be passed to a single-cell detection model, which yields a count of initiating cells. This criterion allows the well to ultimately be declared monoclonal or polyclonal.

[0082] Time-series processing logic enables optimization. In this case, a given monocloning "run" typically contains 300–900 plate wells, and typically 2–6 runs are active at any given time. Well-by-well scans occur daily over 12–30 days, and therefore the average volume of each run during algorithmic processing is approximately 30,000 images. However, in the case of monocloning, the target labels correspond to individual wells rather than images. For this reason, a "well knockout" approach was used, in which if any one of several exclusion criteria is detected by the workflow, the algorithm excludes the entire well from the workflow and ignores all subsequent scans for that well. For example, if no object is detected in the most recent scan, the well is reported as empty during analysis, and its preceding characteristics are considered irrelevant. During the experiment, Monoqlo was run on 8 plates of 96 wells each. The average number of empty wells per plate during processing was found to be 73, ranging from 41 to 92. Therefore, for example, in a case where Monoqlo was applied to eight plates on day 15 of the monocloning process, the well knockout approach eliminated the need to process approximately 8,760 images (76.8%) out of a total of 11,400 images, simply by being empty. Wells found to be polyclonal at any stage of analysis were also excluded from further processing. In the same test run, an average of 11 polyclonal wells were found per plate, and polyclonality was declared after processing an average of 5.73 images. During real-time deployment, the exclusion criteria were further expanded to exclude wells found to exhibit morphological markers of differentiation. Given the enormous size of the datasets requiring daily analysis, our knockout approach offers a significant improvement in computation time.

[0083] The neural network learns to detect colonies and classify their morphology. The inventors began by evaluating the learning trajectory of each CNN in its respective task and benchmarking its prediction performance. In the case of the object detection network, the inventors' initial assessment metric was the change in the value of the loss function when tested on a holdout validation dataset; this dataset represents 20% of the inventors' total image set. Accurate accuracy metrics are not automatically generated by the learning algorithm during the training of such networks, because the model may correctly detect an object even if the coordinates of the labeled bounding box and the predicted bounding box do not exactly match. As an alternative, the performance of these networks was manually evaluated by visually comparing the labels and predictions in the validation images in which each bounding box was drawn. From these comparisons, detection performance was quantified based on the following two metrics: 1) the percentage of labeled objects that were correctly predicted and classified, and 2) the number of false positives (when the model detected an object where it did not exist) as a ratio to the total number of image analyses. The inventors' model validation results are summarized in Figure 4. Furthermore, the true colony width, measured by biologists using an image scale bar, was highly predictable by the X dimension of the bounding box (Pearson's r(266) = 0.917, p < 2.2e-16) (Figure 6).

[0084] Modularized deep learning workflows identify clonality. The effectiveness of Monoqlo as an integrated, modular workflow was benchmarked by first testing its accuracy on a manually curated and class-balanced validation set, and then retrospectively evaluating its clonality identification performance (independent of morphology) on an unfiltered, raw dataset derived from a real-world monoclonal run. The curated test set contained 100 wells each from three classes: empty, monoclonal, and polyclonal, randomly selected from the historical records of manually classified wells. The imaging date at which processing for each well began was randomly generated from the range of 8 to 18 days. Real-world scenario validation was performed on a monoclonal run (DMR0001) containing a total of 768 wells over a 19-day timeframe, resulting in a data volume of 18,240 images. Manual image review found that 561 of these wells were empty; that is, they did not contain any indicators of living cells, regardless of whether they contained remnants of dead colonies, non-living debris, or other artifacts. Monoqlo correctly excluded 556 of these wells (99.1%). The remaining 5 empty wells were reported as monoclonal; these appeared to be false positives by the global detection model due to unidentified non-biological artifacts that superficially resembled cell colonies (Figure 7). Consequently, Monoqlo identified 194 non-empty wells. This included 115 declarations of monoclonality, of which 2 wells and 5 wells were found to be "polyclonal" and "empty" according to ground truth classifications, respectively; the remaining 108 wells (93.9%) were consistent with ground truth. Furthermore, 61 wells were reported as polyclonal; of these, 57 (93.4%) were confirmed by ground truthing, and 4 were found to be monoclonal. The results of both validations are summarized in Figure 5.

[0085] Custom-built programmatic solutions improve deep learning workflows. We identified several situations in which shortcomings of our trained CNN could be robustly corrected using simple programmatic logic, even when they were thought to lead to erroneous results. Perhaps most notably, we found that when only a single colony exists in ground truth (Figure 8), the detection CNN often tends to report multiple overlapping colonies within the image region. This could be partially mitigated by adjusting the size and distribution of the anchor boxes. However, this is cumbersome, can only be done before training the model, and provides only an incomplete solution. Instead, our algorithm combines the overlapping boxes and treats the resulting box as a single object. In the case of colony detection, this never results in a loss of polyclonality identification. For illustration, consider the concept of "colony splitting" that arises due to Monoqlo's inverse time-series approach. Colonies that overlap each other on day N are spatially isolated on day NK and have grown into a combined mass on day N+K; here, K is a variable time quantity that depends on the growth rate and the original separation distance (Figures 9-11). Therefore, the detection of overlapping objects can be safely considered as a single object by our algorithm; if it represents multiple colonies, it will later be detected as completely isolated from each other in earlier images and thus declared to be polyclonal.

[0086] Consideration This research represents the first successful attempt to automate clonality identification using a deep learning object detection approach. This is expected to have the potential to remove significant limitations on scalability in several cell culture domains. This includes the case of iPSC derivation of the present invention, where monocloning is considered essential for two reasons. First, in the case of viral reprogramming, the residual load of the Sendai virus vector used to deliver transcription factors to the inner cells during reprogramming varies greatly between cells. Second, reprogramming treatments often lead to severe chromosomal abnormalities, possibly due to stress-induced mitotic disruption. Both of these factors result in significant phenotypic variation, leading to unpredictable and highly heterogeneous cell lines; thus, the need for monocloning arises, which has historically been a bottleneck in iPSC production. Although it has been suggested that physical monocloning treatment may impose further physiological stress on cells, single-cell cloning remains critical in several use cases. Given the extent to which cohort size determines the feasibility of population studies, removing this bottleneck is a key step in fully unlocking the vast research potential of iPSCs, as this study demonstrates.

[0087] Perhaps more importantly, beyond its initial derivation, a tremendous amount of effort has been put into optimizing the editing efficiency of CRISPR-Cas9 and other forms of genome engineering in iPSCs, which holds enormous potential for functional annotation of gene variants, disease modeling, and validation of polymorphisms identified in gene association studies. Due to the genomic heterogeneity resulting from the editing process, the newly edited population must be monocloned to ensure that all cells possess the same genotype. While the inventors focused on iPSCs in this study, the same applies to gene editing in all cell types. Therefore, the genome engineering pipeline appears to be another significant case where the Monoqlo framework can resolve major bottlenecks in disease research and therapeutic development.

[0088] As long as the cells are imageable and form clonal discrete clumps, this algorithm could potentially be adapted to any cell type. A key example is antibody development, one of the most common cases where monoclonalization is used due to the epitope specificity of monoclonal antibodies. Many of the cell types frequently used in antibody development have been successfully detected in microscopy imaging with CNNs. Since monoclonal antibodies are a central component in numerous drug discovery efforts, the Monoqlo framework may have the potential to provide a valuable tool for the entire pharmaceutical industry.

[0089] This research will further increase the number of applications of deep learning in iPSC process automation. Immunostaining and fluorescence microscopy imaging are far more costly in terms of financial investment and investigation time, and there is considerable interest in optimizing CNNs, particularly for use in conjunction with bright-field microscopy, in efforts to alleviate these costs. For example, previous attempts have successfully trained deep learning models to predict fluorescence labels from bright-field images alone. The results of this research further demonstrate the predictive power of deep learning in various analytical tasks using simple microscope images that do not require fluorescence labeling.

[0090] It has already been shown that standard CNN architectures such as Resnet50™ can be trained to distinguish between differentiated and undifferentiated stem cells in culture, even at an early stage. The classification CNN of the present invention differs from those previously reported in that it does not take a binary "differentiated versus undifferentiated" approach, but rather further stratifies the training classes. In real-world cell culture scenarios, there is considerable variation in the morphology of iPSC colonies due to factors other than pluripotency, but the above approach helped to increase the robustness of the inventors' algorithm when applied to real-world scenarios. In addition, the network was trained on images cropped around clear, single colonies, rather than field images containing numerous randomly seeded cell aggregates. In this sense, the inventors' training data is similar to that used when vector-based CNNs are used to distinguish between "healthy" and "unhealthy" colonies. However, this approach requires a considerable amount of custom pre-processing steps and, importantly, requires manual cropping of precise colony regions, thus limiting its usefulness in automated real-world scenarios. The inventors automated the segmentation stage by using a classification network in conjunction with a colony detection model; this enables fully autonomous deployment in laboratory automation scenarios.

[0091] Shortcomings of this approach were also acknowledged. For example, in cases where two or more initiating cells appear precisely adjacent to each other in the earliest available scan, the clonal state of the well must be considered ambiguous. This is because it is impossible to determine whether the cells were sorted independently from the source plate or whether single cells were successfully sorted during isolation and subsequently divided. However, it should be noted that there is a time lag between cell seeding and attachment to the substrate, during which cells cannot be imaged. For this reason, the timing window of the first scan is critical. Certain other approaches have also been attempted to address this ambiguity through the application of fluorescence microscopy observation to nuclear stained images; nuclear stained images allow for nuclear segmentation and help resolve the spatial distribution of individual cells. However, this does not completely eliminate ambiguity; physically adjacent cells, even if clearly distinct, can still certainly have a polyclonal origin. A limited number of viable approaches to address this ambiguity have been suggested. Researchers could hope to simply assume that wells containing multiple cells in the earliest scans are polyclonal. Otherwise, the ambiguity could only be resolved by generating images within minutes of seeding. However, optical focusing problems are unavoidable due to the time lag required for cells to adhere. Therefore, initiating cells are often invisible, making it impossible to reliably verify monoclonality.

[0092] The 100% detection rate of colonies that reached a sufficient size for successive generations in this study suggests the suitability of Monoqlo for development as a reliable and fully autonomous system.

[0093] Monoqlo is expected to help researchers address several important unanswered questions regarding the predictive potential of deep neural networks in iPSC research. Numerous studies have shown that deep learning approaches can sometimes discriminate biological groups within images even when morphological phenotypes were previously unknown or suspected but undetectable even to trained researchers. For example, CNNs have been shown to predict factors such as cardiovascular disease risk, gender, and smoking status from individual retinal images—none of which were previously thought to manifest morphologically within the retina. Furthermore, in the case of iPSCs, deep neural networks have been successfully trained to predict donor identity from imaging of clinical-grade iPSC-derived retinal pigment epithelium. These findings suggest that unidentified predictive markers are likely present in the morphology of iPSC colonies. For instance, they may be able to predict, with greater accuracy than randomness, whether currently undifferentiated colonies will spontaneously differentiate at an early stage. Given the considerable cost of continuing to culture cells that may eventually become unusable, the successful training of such models would bring significant advantages to iPSC derivation. Other candidate targets for prediction based on CNN classification or regression include Sendai virus loading, future QC pass / fail status, and relative differentiation affinity to specific germ layers.

[0094] Training such models always requires a large training volume. The Monoqlo framework allows for the algorithmic segmentation and extraction of colonies from raw datasets, in addition to automatically removing images from empty wells, which typically constitute the majority of images. In many cases, researchers may also be able to batch label images based on classifications they themselves have assigned to recent images of a given colony or well. By applying a classification network that identifies differentiation, Monoqlo can retrospectively assign labels such as "will differentiate" or "won't differentiate" to earlier instantiations of colonies. This can reduce the need for the very laborious task of manually reviewing and labeling unfiltered image sets, which allows for the partially or fully autonomous generation of large training volumes for future models. Thus, our algorithm provides a valuable tool for generating custom datasets for future investigations into the use of deep learning in iPSC research.

[0095] In summary, we demonstrated a framework that automates the verification of monoclonality in bright-field microscopy observations using a modular design and deep learning algorithm, requiring relatively little labeling. We further extended the functionality of this workflow to colony morphology classification, demonstrating the potential for autonomous monitoring of monoclonal cell line development and clonal selection in automated workflows. Monoqlo represents a crucial step in enabling the widespread adoption of high-throughput cell line generation and editing workflows. This could advance current technologies toward the goal of unrestricted upscaling and dissemination of pluripotent stem cells for biomedical research applications, eliminating significant bottlenecks in the specific cases of iPSC derivation and genome editing. Furthermore, in contrast to relying solely on machine learning models to tackle all aspects of a given task, this work serves as a useful example highlighting the advantages of combining the vast capabilities of now-recognized convolutional neural networks with human-designed algorithmic solutions.

[0096] While the present invention has been described with reference to the above embodiments, it will be understood that modifications and variations are included within the spirit and scope of the invention. Accordingly, the present invention is limited only by the appended claims.

Claims

1. The imaging system includes the following: (a) Imaging device, and (b) A controller functionally connected to the imaging device, The controller is operable to generate an image via the imaging device and to analyze the generated image via the processor to determine the clonality of the target object. The processor, (i) A step of generating multiple time-series images of an image area via the imaging device; (ii) The step of identifying the target object within the image area of ​​the most recent image among the multiple time-series images; (iii) A step of generating a target object image area within the image area of ​​the nearest image containing the identified target object, wherein the target object image area has a perimeter within the image area of ​​the nearest image; (iv) Using the target object image area, extract the next most recent image; (v) the step of identifying the target object in the cropped image; and (vi) The step of determining the clonality of the target object by inductively repeating (iv) and (v) until the next most recent image is the earliest image of the plurality of time-series images. Includes functionality for performing controller.

2. The system according to claim 1, wherein the plurality of time-series images include more than 10, 100, 1,000, 10,000, 100,000 or more images.

3. The system according to claim 2, wherein (iv) is performed only when only one target object is identified within the image area.

4. The system according to claim 1, wherein the identification of the target object in the cropped image in (v) includes the step of detecting the target object in a region corresponding to the target object image area from the most recently analyzed image.

5. The system according to claim 4, wherein the step of determining the clonality of a target object includes determining whether more than one target object is present in one or more of a plurality of time-series images.

6. The system according to claim 1, wherein the step of identifying a target object includes classifying the target object based on the attributes of the target object.

7. The system according to claim 6, wherein the attribute is a physical characteristic of the target object.

8. The system according to claim 7, wherein the physical feature is size or shape.

9. The system according to claim 1, wherein (ii) to (v) are performed via one or more convolutional neural networks (CNNs).

10. The system according to claim 1, wherein the target object is a cell or a cell colony.

11. A computer-implemented method comprising the step of performing image analysis, which includes identifying a target object in an image using a system according to any one of claims 1 to 10, and optionally analyzing it.

12. The method according to claim 11, wherein analyzing a target object includes classifying the target object based on the attributes of the target object.

13. The method according to claim 12, wherein the attribute is a physical feature of the target object.

14. The method according to claim 13, wherein the physical feature is size or shape.

15. The method according to claim 13, wherein the target object is a cell or cell colony, and the physical attribute is a morphological feature of a cell.

Images