Method for digital assay of a target and device using the method

Abstract

Apparatuses for digital assays of targets and methods of using the apparatuses according to example embodiments of the present disclosure are provided. Methods of digital assays of targets according to example embodiments of the present disclosure include acquiring images of a plurality of microdroplets, predicting at least one region based on the images of the plurality of microdroplets using a prediction model based on an artificial neural network, the prediction model configured to segment the at least one region between positive microdroplets, negative microdroplets, and atypical microdroplets with the images of the plurality of microdroplets as input, determining a quantity of the plurality of microdroplets based on the at least one region, and providing quantitative data of the target based on the quantity of the plurality of microdroplets.

Description

[0001] Cross-references to related applications

[0002] This application claims priority to Korean Patent Application No. 10-2021-0159677, filed with the Korean Intellectual Property Office on November 18, 2021, the disclosure of which is incorporated herein by reference. Technical Field

[0003] This disclosure relates to an apparatus for digital measurement of a target and a method of using the apparatus. Background Technology

[0004] Digital assays involve segmenting a target into thousands of microstructures, examining the signal based on the target's presence or absence, and then quantifying the signal using a Poisson process. This digital assay offers the advantages of higher accuracy and sensitivity than other analytical methods and allows for absolute quantification without the need for standard samples.

[0005] Examples of digital assays include digital polymerase chain reaction (dPCR), digital enzyme-linked immunosorbent assay (digital ELISA), digital adjacent linkage assay (digital PLA), and digital loop-mediated isothermal amplification assay (digital LAMP). These digital assays are being used in a variety of diagnostics.

[0006] For example, digital PCR is widely used for research, development, or diagnostic purposes in the fields of life sciences, genetic engineering, and medicine.

[0007] Specifically, PCR is a molecular biology technique that replicates DNA from isolated biological samples. It can be used for a variety of tasks, such as the diagnosis of infectious diseases, the detection of hereditary diseases, the identification of genetic fingerprints, gene cloning, paternity testing, genotyping, gene sequencing, and DNA computation.

[0008] Specifically, digital polymerase chain reaction (dPCR) is a gene detection method that divides existing samples into droplets with nanoliter volumes and observes changes in the fluorescence of targets within them. This digital PCR has very high sensitivity and enables absolute quantitative analysis, making it highly applicable in gene analysis, biomarker development, and gene sequencing.

[0009] Meanwhile, digital assays, including digital PCR, have problems such as difficulty in determining the threshold for tiny droplets in optical detection, and the need for optical coherence and the use of separate fluorescence readers based on geometric optics.

[0010] Therefore, there has been a need to develop a system for digital measurement of new targets that can overcome the limitations of related technologies and analyze targets in samples with high precision.

[0011] The background of this disclosure is provided to facilitate understanding of the disclosure. It should not be construed as an admission that the matters described in the background of this disclosure are prior art. Summary of the Invention

[0012] As a way to overcome the limitations of digital measurement methods in related technologies, adaptive thresholding methods have emerged. More specifically, adaptive thresholding methods involve segmenting an image and then obtaining the histogram of the segmented images, thus achieving higher analytical accuracy than methods that collectively determine the threshold. However, according to adaptive thresholding methods, the user needs to determine an optimal threshold for each image, which can introduce user bias. Furthermore, limitations may still exist in providing highly reliable results due to background variations, signal variations, noise spikes, etc.

[0013] The inventors of this disclosure have noted that image-based quantitative analysis of droplets can solve the aforementioned problems.

[0014] More specifically, the inventors of this disclosure have attempted to develop a method for applying an artificial intelligence-based predictive model that can very reliably and accurately analyze randomly distributed positive and negative microstructures (droplets) without having fixed units and without user bias, such as droplet-based numerical determinations.

[0015] Specifically, the inventors of this disclosure have attempted to apply deep learning capable of segmenting droplet images to create models for segmenting positive and negative droplets in an image.

[0016] Furthermore, the inventors of this disclosure have noted that the Hough transform can be used to count droplets more accurately, as the transform finds the diameter and radius of the circle to count the number of circles, in order to count the number of droplets in the segmented image.

[0017] As a result, the inventors of this invention have applied predictive models to develop a new quantitative analysis system that can not only distinguish and count positive droplets, but also distinguish and count negative droplets, as well as other atypical droplets that differ from droplets with typical shape and size, and which do not include the target gene in the droplet image (e.g., fluorescence image).

[0018] At this point, the inventors of this invention can design a new quantitative analysis system to output the absolute quantitative value of the target gene based on droplet images to achieve automatic quantitative analysis without user intervention.

[0019] The inventors of this invention provide a novel quantitative analysis system that aims to detect droplets with target genes with high precision, unaffected by factors such as atypical shape, size, low fluorescence intensity, and background variations.

[0020] Furthermore, the inventors of this disclosure expect the diagnostic technology to be highly advanced.

[0021] Specifically, the inventors of this invention have recognized that by using an artificial intelligence-based predictive model to adjust the focusing precision of droplet-based digital PCR with analytical limitations according to a method based on related technologies, signals can be analyzed with high precision based on differences in fluorescence intensity and edge ambiguity.

[0022] Therefore, the objective of this invention is to provide a method for digital measurement of a target and an apparatus for digital measurement of a target using the method, which is configured to receive images of droplets and perform quantitative analysis of the target using a prediction model based on an artificial neural network algorithm.

[0023] The purpose of this disclosure is not limited to the above-mentioned purposes, and other purposes not mentioned above will be clearly understood by those skilled in the art through the following description.

[0024] To address the aforementioned problems, according to one aspect of this disclosure, a method for digital measurement of a target is provided.

[0025] The method includes acquiring images of multiple droplets; using an artificial neural network-based prediction model to predict at least one region based on the images of the multiple droplets, the prediction model being configured to segment at least one region between positive droplets, negative droplets, and atypical droplets with the images of the multiple droplets as input; determining the number of multiple droplets based on the at least one region; and providing quantitative data on the target based on the number of multiple droplets.

[0026] According to the features of this disclosure, acquiring an image of multiple droplets may include allowing the multiple droplets to flow into a chamber with a height of 1.1 to 1.9 times the diameter of the droplets, such that the multiple droplets exist as a monolayer; and acquiring a monolayer image of the multiple droplets. In this case, predicting at least one region may include using a prediction model to predict at least one region based on the monolayer image of the multiple droplets.

[0027] According to another feature of this disclosure, at least one region includes a positive droplet region, a negative droplet region, and an atypical droplet region. A positive droplet is defined as a droplet that includes a target and fluorescent material, a negative droplet is defined as a droplet that includes only fluorescent material or an empty droplet, and an atypical droplet is defined as a droplet that includes a target and fluorescent material and whose shape is different from that of a positive droplet.

[0028] According to another feature of the invention, determining the number of multiple droplets includes: determining the number of positive and negative droplets based on positive and negative droplet regions, excluding atypical droplet regions. Furthermore, providing quantitative data on the target further includes determining the target concentration based on the number of positive and negative droplets.

[0029] According to another feature of the invention, determining the number of positive and negative droplets includes: using Hough transform based on the shape of the positive droplet region, the negative droplet region, and the atypical droplet region to remove the atypical droplet region; and determining the number of positive and negative droplets.

[0030] According to another feature of this disclosure, acquiring images of multiple droplets includes acquiring images of multiple droplets composed of multiple sets, and predicting at least one region includes: using a prediction model to predict at least one region of the images of the multiple droplets for each set of the multiple sets. Furthermore, providing quantitative data on the target includes determining the total copy number of the target based on the number of multiple droplets predicted for each set of the multiple sets.

[0031] According to another feature of this disclosure, predicting at least one region may further include using a prediction model to predict at least one region of multiple droplets selected from one of a plurality of sets. Furthermore, providing quantitative data on the target may further include determining the copy number of the target based on the number of multiple droplets predicted for a set.

[0032] According to another feature of the invention, the prediction model is further configured to perform quantitative analysis of the target based on at least one region, and providing quantitative data of the target may further include using the prediction model to determine quantitative data of the target based on the number of multiple droplets.

[0033] According to another feature of this disclosure, the prediction model can be further configured to segment at least one region and a background region using an image of multiple droplets as input. The method may further include, after acquiring the image, using the prediction model to segment at least one region and a background region, determining the number of multiple droplets based on the at least one region and the background region, and providing quantitative data of the target based on the number of multiple droplets.

[0034] According to yet another feature of this disclosure, the image can be a fluorescence image.

[0035] The image is a series of focused images of multiple droplets, and the prediction model can be configured to further segment at least one region using the multiple focused images as input.

[0036] To achieve the above objectives, according to another aspect of this disclosure, an apparatus for digital measurement of a target is provided.

[0037] The device includes a light source that illuminates at least one surface of a chamber in which microdroplets including a target are disposed; an image sensor configured to provide an image of the microdroplets in the chamber; and a processor operatively connected to the image sensor. The processor is configured to predict at least one region based on the images of multiple microdroplets using an artificial neural network-based prediction model. This prediction model is configured to segment at least one region between positive, negative, and atypical microdroplets using the images of the microdroplets as input, determine the number of multiple microdroplets based on the at least one region, and determine quantitative data of the target based on the number of multiple microdroplets.

[0038] According to the features of this disclosure, the height of the chamber is 1.1 to 1.9 times the diameter of the plurality of microdroplets, and the processor is further configured to allow the plurality of microdroplets to flow into the chamber, such that the plurality of microdroplets exist as a monolayer. At this point, the image sensor is further configured to acquire a monolayer image of the plurality of microdroplets, and the processor can be further configured to use a prediction model to predict at least one region based on the monolayer image of the plurality of microdroplets.

[0039] According to another feature of this disclosure, a region includes a positive droplet region, a negative droplet region, and an atypical droplet region. A positive droplet is defined as a droplet that includes a target and fluorescent material, a negative droplet is defined as a droplet that includes only fluorescent material or an empty droplet, and an atypical droplet is defined as a droplet that contains a target and fluorescent material and whose shape is different from that of a positive droplet.

[0040] According to another feature of this disclosure, the processor can be further configured to determine the number of positive and negative droplets based on the positive and negative droplet regions, excluding atypical droplet regions, and to determine the concentration of the target based on the number of positive and negative droplets.

[0041] According to another feature of this disclosure, the processor can be further configured to use Hough transform based on the shape of positive droplet regions, negative droplet regions, and atypical droplet regions to remove atypical droplet regions; and to determine the number of positive and negative droplets.

[0042] According to yet another feature of this disclosure, the image sensor can be further configured to acquire images of multiple droplets comprising multiple sets. In this case, the processor can be further configured to use a prediction model to predict at least one region of the images of the multiple droplets for each of the multiple sets, and to determine the total copy number of the target based on the number of the multiple droplets.

[0043] According to yet another feature of this disclosure, the processor can be further configured to use a prediction model to predict at least one region of a plurality of droplets for one of a plurality of sets, and to estimate the copy number of a target based on at least one region of a set.

[0044] According to another feature of this disclosure, the prediction model can be further configured to perform quantitative analysis of the target based on at least one region. In this case, the processor can be further configured to use the prediction model to determine quantitative data of the target based on the number of multiple droplets.

[0045] According to another feature of this disclosure, the prediction model can be further configured to segment at least one region and a background region using an image of multiple droplets as input. In this case, the processor can be further configured to use the prediction model to segment the at least one region and the background region, determine the number of multiple droplets based on the at least one region and the background region, and provide quantitative data of the target based on the number of multiple droplets.

[0046] Further details of the exemplary embodiments are included in the detailed description and accompanying drawings.

[0047] The present invention can provide a novel target quantification system based on droplet imaging, which can detect targets in droplets without the need for units for fixing droplets and units for adjusting the spacing between droplets.

[0048] More specifically, according to this disclosure, it is possible to detect positive droplets including targets without using units (e.g., arrays) for immobilizing the generated droplets, and to perform quantitative analysis of genes with improved accuracy.

[0049] Furthermore, according to this disclosure, the target can be detected more easily than in serial counting-based readout methods, where interval adjustment is essential when detecting the optical signal of a droplet moving in the channel.

[0050] Specifically, according to the present invention, a target quantification system is provided, which is configured to perform quantitative analysis based on microdroplet images using artificial intelligence algorithms, thereby enabling precise quantitative analysis of microdroplets and, consequently, targets within the microdroplets.

[0051] Furthermore, according to this disclosure, images of droplets (e.g., fluorescence images) can be provided individually, including not only positive droplets, but also negative droplets that do not contain a target, as well as atypical droplets whose shape or size differs from that of typical droplets.

[0052] Therefore, it is possible to detect target-containing droplets with high precision without being affected by factors such as atypical shape, size, low fluorescence intensity, and background changes.

[0053] Furthermore, according to this disclosure, a prediction model trained to segment droplet regions based on images of droplets with various focal points is used, enabling the segmentation of droplet regions and the determination of quantitative data with high accuracy even when the input image has blurred focal points.

[0054] Furthermore, this disclosure can provide absolute quantitative values ​​of targets based on droplet images to automatically perform quantitative analysis without user intervention, which can contribute to significant advancements in diagnostic technology.

[0055] Specifically, according to this disclosure, positive signals are clearly identified to detect very small amounts of virus and to easily distinguish patients with asymptomatic viral infections.

[0056] The effects of this disclosure are not limited to those illustrated above, and this specification includes many more effects. Attached Figure Description

[0057] The above and other aspects, features and advantages of this disclosure will become clearer from the following detailed description taken in conjunction with the accompanying drawings, in which:

[0058] Figure 1 The structure and configuration of an apparatus for digital measurement of a target are shown for various exemplary embodiments of this disclosure;

[0059] Figures 2A to 2C The process of performing digital measurement of a target using an apparatus for digital measurement of a target according to various exemplary embodiments of the present disclosure is illustrated;

[0060] Figure 3A and 3B The illustration shows learning data of a predictive model for a device for digital measurement of a target according to various exemplary embodiments of the present disclosure; and

[0061] Figure 4A , 4B Images 5A and 5B show evaluation results of predictive models for devices for digital measurement of targets according to various exemplary embodiments of the present disclosure. Detailed Implementation

[0062] The advantages of this disclosure and methods for achieving these advantages and features will become clear by referring to the exemplary embodiments described below in detail and in conjunction with the accompanying drawings. However, this disclosure is not limited to the exemplary embodiments disclosed herein, but will be implemented in various forms. The exemplary embodiments are provided by way of example only to enable those skilled in the art to fully understand the content and scope of this disclosure. Therefore, this disclosure will be limited only by the scope of the appended claims.

[0063] The shapes, dimensions, ratios, angles, quantities, etc., shown in the accompanying drawings to describe exemplary embodiments of this disclosure are merely examples, and this disclosure is not limited thereto. Furthermore, in the following description, detailed explanations of known related technologies may be omitted to avoid unnecessarily obscuring the subject matter of this disclosure. Terms such as “comprising,” “having,” and “consisting of” as used herein are generally intended to allow for the addition of additional components, unless these terms are used in conjunction with the term “only.” Unless otherwise expressly stated, any reference to the singular is capable of including the plural.

[0064] Even if not explicitly stated, components are interpreted as including a normal error range.

[0065] The features of the various embodiments of this disclosure may be combined or integrated with each other in part or in whole, and may be technically interlocked and operated in various ways as understood by those skilled in the art, and the embodiments may be performed independently or in relation to each other.

[0066] To clearly explain this specification, the terms used in this specification will be defined below.

[0067] The term "target" used in this instruction manual can refer to a specific DNA or RNA. Ideally, a target could be the RNA of a specific virus, but it is not limited to this.

[0068] As used in this disclosure, the term "droplet" refers to a droplet used for digital PCR and may include a target (or non-target) to be amplified, fluorescent material, or a sample for PCR.

[0069] At this point, droplets can be generated by contacting the sample with an oil that is immiscible.

[0070] The term "image of a droplet" as used in this specification can refer to an image of multiple droplets, which may or may not include a target or fluorescent material as a monolayer. In this case, the image of a droplet can be a fluorescence image of a droplet that expresses fluorescence by amplifying a target, but is not limited to this.

[0071] According to another feature of this disclosure, the image of the droplet can be an image with multiple focal points for a chamber region in which the droplet is present.

[0072] Meanwhile, the images of droplets can include positive droplets, negative droplets, or atypical droplets.

[0073] The term "positive droplet" as used in this specification can refer to a droplet containing both a target and fluorescent material, and "negative droplet" can refer to a droplet containing only fluorescent material or an empty droplet. Furthermore, "atypical droplet" can refer to a droplet whose shape differs from that of a positive or negative droplet. More specifically, atypical droplet can refer to a droplet containing both a target and fluorescent material but whose shape differs from that of a positive droplet.

[0074] At this point, positive droplets and / or atypical droplets can express fluorescence by amplifying the target.

[0075] Furthermore, according to the features of this disclosure, an image of the droplets can be obtained from the chamber containing the droplets on which PCR is performed. In other words, the image of the droplets can be an image of the chamber in which the droplets are present.

[0076] Meanwhile, the height of the chamber can be similar to the height of the channel in which the droplets are generated or the diameter of the droplets. Therefore, within the chamber, the droplets can be arranged on a single layer.

[0077] As used in this disclosure, the term "single-layer image" refers to an image of droplets aligned on a single layer in a chamber and can also refer to an image captured from the upper part of the chamber.

[0078] At this point, the height of the chamber can be 1.1 to 1.9 times the diameter of the multiple droplets.

[0079] As used in this disclosure, the term "predictive model" can refer to a model trained to segment droplets expressing fluorescence in an image by taking an image of the droplets as input.

[0080] More specifically, the prediction model can be a model trained to segment and classify at least one region among positive droplets, negative droplets, atypical droplets, or background regions without droplets, using fluorescent images of droplets as input.

[0081] In this case, the prediction model can be a model trained to segment the droplet region by taking multiple focused images of the chamber region containing droplets as input.

[0082] In other words, when the input image has an indistinct focus, the prediction model can segment the droplet region with high accuracy.

[0083] According to the features of this disclosure, the prediction model can be further trained to output quantitative data (e.g., copy number or concentration) of the target based on the segmentation results.

[0084] At this point, the prediction model can be a ResNet deep neural network (DNN) based model, but is not limited to it. For example, the prediction model can be a SegNet network, VGG-16, deep convolutional neural network (DCNN), convolutional neural network (CNN), recurrent neural network (RNN), restricted Boltzmann machine (RBM), deep belief network (DBN), single detector (SSD) model, or a U-net based prediction model.

[0085] In the following text, reference will be made to Figure 1A detailed description is provided of apparatus and configuration for digital measurement of targets according to various exemplary embodiments of the present disclosure.

[0086] In this context, digital PCR will be described as an example in digital assay methods, but it is not limited to this.

[0087] Figure 1 The structure and configuration of an apparatus for digital measurement of a target are shown for various exemplary embodiments of this disclosure.

[0088] refer to Figure 1 The apparatus 1000 for digital determination of a target according to an exemplary embodiment of the present disclosure may consist of: a droplet generating unit 110 for generating droplets; a chamber 120 in which PCR is performed on the generated droplets and droplets including target amplification completed thereon; a valve 130 for controlling the movement of droplets; a light source 310; an image sensor 400 for providing images of the droplets; and a processor 500 configured to communicate therewith.

[0089] At this point, the apparatus 1000 for digital measurement of a target according to an exemplary embodiment of the present disclosure may further include a temperature regulating unit 200 for controlling the temperature of the chamber 120 and a reflector 320 for switching the direction of the light source.

[0090] More specifically, light is irradiated onto the microdroplets in the chamber 120 via a light source 310 and a reflector 320, and an image sensor 400 can acquire images of multiple microdroplets in the chamber 120. The light source 310 can be a fluorescent lamp used to express the color of the fluorescent material, but is not limited to this. For example, fluorescence can be irradiated onto the chamber 120 via a fluorescent filter (not shown).

[0091] Meanwhile, the height of chamber 120 can be 1.1 to 1.999 times the diameter of the multiple microdroplets generated from microdroplet generation unit 110. In other words, the multiple microdroplets on chamber 120 can be arranged on a single layer.

[0092] Meanwhile, the image of the droplets, which is obtained as an image of the chamber on which PCR is performed, can be a monolayer image of droplets expressing fluorescence.

[0093] Images of multiple droplets acquired from image sensor 400 are transmitted to processor 500, which is configured to communicate with image sensor 400 and can perform digital determination of the target based on the images.

[0094] At this point, the processor 500 can make predictions based on an artificial neural network-based model.

[0095] More specifically, the processor 500 can be further configured to segment droplet types in a droplet image using a prediction model, count positive and negative droplets expressing a target, and determine the concentration of a target (specifically a target gene) based on the counted droplets. The prediction model is trained to detect regions of droplets by taking the image of the droplets as input.

[0096] Based on the aforementioned structural features, quantitative analysis can be performed solely on images of droplets on which PCR has been completed, without adjusting the droplet movement process to the channel-type detection unit or the droplet spacing. Specifically, this predictive model can be applied to improve the reliability of target testing.

[0097] Meanwhile, as described above, the apparatus 1000 for digital measurement of a target according to an exemplary embodiment of the present disclosure can be applied to various digital measurements other than digital PCR when its configuration does not include the temperature control unit 200.

[0098] In the following text, reference will be made to Figures 2A to 2C The process of performing digital measurement of a target using an apparatus for digital measurement of a target according to various exemplary embodiments of the present disclosure is described.

[0099] Figures 2A to 2C The process of performing digital measurement of a target using an apparatus for digital measurement of a target according to various exemplary embodiments of the present disclosure is illustrated.

[0100] First, refer to Figure 2A According to the target digital determination method of the exemplary embodiment of this disclosure, in step S210, the sample is induced to contact with oil to generate droplets, and in step S220, the temperature is controlled to perform a PCR reaction of the target in the droplets. Next, in step S230, images of multiple droplets are acquired, in step S240, at least one region of positive droplets, negative droplets, and atypical droplets is determined by a prediction model, and in step S250, the number of positive and negative droplets is determined. Finally, in step S250, quantitative data of the target gene are determined.

[0101] More specifically, see reference Figure 2B In step S210, which induces the sample to come into contact with oil, in droplet generation unit 110, the sample, including a target, fluorescent material, primers, and polymerase, comes into contact with oil to form multiple droplets encapsulated by oil.

[0102] As a result of step S210, which induces contact between the sample and the oil, multiple microdroplets, including the target, are generated.

[0103] According to the features of this disclosure, after step S210, which induces contact between the sample and the oil, a step can be performed in which multiple microdroplets flow into chamber 120 such that the multiple microdroplets exist as a monolayer.

[0104] At this point, the height of the chamber can be 1.1 to 1.99 times the diameter of the droplet, but is not limited to this.

[0105] In other words, multiple microdroplets can be placed on a single layer in chamber 120.

[0106] Meanwhile, multiple droplets can be at least one of positive droplets, negative droplets, and atypical droplets.

[0107] At this point, a positive droplet is defined as a droplet that includes both a target and fluorescent material, a negative droplet is defined as a droplet that includes only fluorescent material or an empty droplet, and an atypical droplet is defined as a droplet whose shape differs from that of a positive or negative droplet.

[0108] Next, in the temperature control step S220, temperature conditions are cycled for amplifying the target in the droplets, resulting in the presence of droplets including the amplified target in the chamber 120.

[0109] For more specific details, please refer to the following: Figure 2C As a result of step S220, which controls the temperature, positive droplets containing targets such as viral genes can express fluorescence by amplifying the genes and using fluorescent materials. Conversely, in the case of negative droplets that are empty droplets, no fluorescence is expressed even if PCR is performed.

[0110] Conversely, in the case of negative droplets that are empty droplets, fluorescence may not be expressed even if PCR is performed.

[0111] In addition, atypical droplets (not shown) whose shape differs from negative or positive droplets may express fluorescence or not, depending on whether they contain a target.

[0112] Next, refer to Figure 2A In step S230, images of multiple droplets on which PCR has been completed are acquired.

[0113] More specifically, see reference Figure 2B In step S230 of acquiring images of multiple droplets, images 612 of multiple droplets are acquired by the quantitative analysis device 100 according to various exemplary embodiments of the present disclosure.

[0114] At this point, the image 612 of multiple droplets is a fluorescence image of multiple droplets existing as a monolayer in the chamber, and may be a defocused image of droplets with irregular sizes or with various fluorescence intensities.

[0115] According to the features of this disclosure, in step S230 of acquiring images of multiple droplets, images of multiple droplets formed by multiple sets can be acquired.

[0116] For example, in a digital PCR test to confirm whether an entity is infected with a virus, the copy number of the gene amplified for absolute quantitative analysis of the target (virus) is determined. In this case, to acquire images of multiple droplets and confirm the copy number of each sample, multiple droplet sets can be provided. Therefore, images of multiple droplets corresponding to each of the multiple sets can be acquired. However, this is not the only possibility; images of multiple droplets corresponding to one set selected from the multiple sets can also be acquired.

[0117] Next, refer to Figure 2A In step S240, at least one region of negative droplets and atypical droplets in the image of multiple droplets can be predicted by a prediction model.

[0118] According to the features of this disclosure, in step S240 of predicting at least one region, a prediction model can be used to predict at least one region of negative droplets and atypical droplets for images of multiple droplets corresponding to each of a plurality of sets.

[0119] According to another feature of this disclosure, in step S240 of predicting at least one region, a prediction model can be used to predict at least one region of negative droplets and atypical droplets for an image of multiple droplets corresponding to one of a plurality of sets.

[0120] According to another feature of this disclosure, in step S240 of predicting at least one region, a background region, as well as at least one region of negative droplets and atypical droplets, can be predicted by a prediction model.

[0121] For more specific details, please refer to the following: Figure 2B In step S240, which predicts at least one region, images 612 of multiple droplets are input into a prediction model 510 based on an artificial neural network. Next, droplet regions and background regions of positive, negative, and atypical droplets can be segmented and output by the prediction model 510. At this point, referring to an image 614 in which at least one region is segmented, negative droplets are red due to the expression of fluorescent material, while positive and atypical droplets containing target genes and fluorescent material can be green. That is, even if an image 612 of multiple droplets is input that is out of focus, has uneven fluorescence intensity, or has irregular droplet size, the prediction model 510 can segment droplet regions with high accuracy.

[0122] Meanwhile, the prediction model 510 can segment images based on deep learning algorithms such as ResNet, SegNet, UNet, Faster R-CNN, FCN, or VoxNet, but is not limited to these.

[0123] Next, refer to again Figure 2A Step S250 is performed to determine the number of positive and negative droplets.

[0124] According to the features of the present invention, in step S250 of determining the number of positive and negative droplets, atypical droplet regions can be removed by Hough transform based on the shape of the positive droplet region, the negative droplet region, and the atypical droplet region.

[0125] For example, refer to together Figure 2B Within the droplet region segmented by the Hough transform 520, the number of positive and negative droplets after excluding atypical droplets with atypical shapes was determined. At this point, the Hough transform 520 can determine the number of positive and negative droplets by finding the diameter and radius of the droplets. As a result, the number of positive and negative droplets can be determined (616).

[0126] Meanwhile, the number of positive and negative droplets cannot necessarily be determined by Hough transform.

[0127] Refer again Figure 2A Step S260 is performed to provide quantitative data on the target gene.

[0128] According to the features of the present invention, in step S260 of providing quantitative data of the target, the copy number of the target can be determined based on the number of positive and negative droplets.

[0129] According to the features of this disclosure, in step S260 of providing quantitative data on the target, the total copy number of the target can be determined based on the number of positive and negative droplets predicted for each of the multiple sets.

[0130] According to another feature of this disclosure, in step S260 of providing quantitative data of the target, the total copy number (μl) of the target for each sample is determined to determine the concentration of the target.

[0131] For example, refer to together Figure 2B In step S260, which provides quantitative data for the target, the target concentration (copy number / μl) for each sample is determined based on the number of positive droplets and the number of negative droplets after excluding atypical droplets (618), and the target concentration can be provided as quantitative data.

[0132] Refer again Figure 2AAccording to another feature of this disclosure, in step S260 of providing quantitative data on the target, the copy number of the target can be estimated based on the number of droplets predicted for one of a plurality of sets.

[0133] For example, in step S260, which provides quantitative data on the target, the copy number of the entire target can be estimated based on the copy number of the target for a defined set without performing quantitative analysis on the entire droplet set.

[0134] According to another exemplary embodiment of this disclosure, in step S260, which provides quantitative data on the target, the copy number and concentration of the target can be determined by a prediction model. That is, the prediction model can be a model trained to take images of multiple droplets as input and output quantitative analysis results (e.g., copy number, target concentration, or number of positive droplets).

[0135] It is possible to provide accurate analytical results for a target through quantitative analysis methods according to various exemplary embodiments of the present disclosure as described above.

[0136] Specifically, this disclosure overcomes the limitations of related digital measurement methods, because analyzing the intensity of fluorescent material flowing into each droplet in the detection channel requires a long time, requires expensive photomultiplier tubes (PMTs), may require droplet reproduction, and may result in quantitative analysis errors due to optical coherence.

[0137] Evaluation: Performance evaluation of prediction models based on various exemplary embodiments of this disclosure

[0138] In the following description, the evaluation results of the predictive models applied to various exemplary embodiments of the present disclosure will be described with reference to Figures 3, 4 and 5.

[0139] Figure 3A and 3B The diagram illustrates learning data for predictive models of devices for digital measurement of targets according to various exemplary embodiments of the present disclosure. Figures 4, 5A, and 5B show evaluation results of predictive models of devices for digital measurement of targets according to various exemplary embodiments of the present disclosure.

[0140] First, refer to Figure 3A and 3B The fluorescence image is shown. Figure 3A ) and masked images ( Figure 3B ), which is used as learning data for predictive models used in various exemplary embodiments of this disclosure.

[0141] More specifically, the learning data are fluorescence images captured in the state of multiple droplets scattered on a monolayer, and may include positive droplets containing both a target and fluorescent material, and negative droplets containing only fluorescent material or empty droplets. Furthermore, the fluorescence images may include atypical droplets whose shapes differ from those of positive or negative droplets, and may also include background regions that are not droplets.

[0142] In other words, the predictive model can be trained to segment regions of droplets by taking the learning data as input, such as positive droplet regions, negative droplet regions, atypical droplet regions, and further background regions.

[0143] Furthermore, the predictive model can be trained to segment droplet regions using multiple focused images as learning data, such as positive droplet regions, negative droplet regions, atypical droplet regions, and further background regions.

[0144] Therefore, even with out-of-focus images, droplet regions can be segmented with high precision.

[0145] Meanwhile, the prediction model used for various exemplary embodiments of this disclosure may be ResNet101, but is not limited thereto.

[0146] Next, refer to Figure 4A Multiple focused images are shown, including A-1 (focus position 0μm), A-2 (focus position 50μm), A-3 (focus position 100μm) and A-4 (focus position 150μm).

[0147] Let's refer to each other. Figure 4B The results show the predictions of four focused images using a combination of prediction models and Hough transform (segmentation + Hough) for various exemplary embodiments of this disclosure, as well as the results of analyzing four focused images by Hough transform and manual counting alone.

[0148] More specifically, as the fluorescence intensity of positive droplets becomes blurred with increasing defocus, the counting results obtained by Hough transform alone appear to deviate significantly from the manual counting values.

[0149] In contrast, the counting results using a combination of the predictive model and the Hough transform (segmentation + Hough) of the various exemplary embodiments of this disclosure show a high degree of matching with the manually counted values, even when out of focus.

[0150] This result demonstrates that a prediction model trained to segment droplet regions (especially positive droplet regions) using multiple focused images as learning data exhibits excellent segmentation performance.

[0151] Next, refer to Figure 5AThe image shows fluorescence images taken after digital PCR of input DNA with the following four concentrations: 3.94 × 10⁻⁶. 1 Copy number / μL, 3.94 × 10 2 Copy number / μL, 3.94 × 10 3 Copy number / μL and 7.88×10 3 Copy number / μL. At this point, the higher the concentration, the greater the number of positive droplets.

[0152] refer to Figure 5B The diagram illustrates concentration predictions for 10 fluorescence images with four concentrations using a combination of prediction models and Hough transforms (segmentation + Hough) employed in various exemplary embodiments of this disclosure, as well as concentration measurements of the 10 fluorescence images with four concentrations using a Bio-Rad reader (Bio-Rad droplet reader). These results are compared to ideal values ​​based on the input DNA concentration.

[0153] More specifically, when estimating the four concentrations using only the Hough transform, the results deviated significantly from the ideal values. Furthermore, the DNA concentration estimates using the Bio-Rad droplet reader also deviated significantly from the ideal values.

[0154] In contrast, the concentration predictions using a combination of the prediction model and the Hough transform (ResNet 101+Hough) of the various exemplary embodiments of this disclosure are closest to the ideal values.

[0155] These results demonstrate that a digital measurement system for targets based on prediction models and Hough transforms, configured to segment droplet regions, particularly positive droplet regions, and using Hough transforms to remove unnecessary counting information, can provide highly accurate quantitative data for targets.

[0156] In other words, the digital determination method for targets based on prediction models and / or Hough transforms according to various exemplary embodiments of this disclosure is capable of detecting positive droplets with high sensitivity and providing accurate quantitative analysis results.

[0157] Specifically, according to this disclosure, in various digital analyses, positive droplets including the target can be detected without using units for fixing the generated droplets or adjusting the droplet spacing, and quantitative analysis of the target can be performed with improved accuracy.

[0158] Although exemplary embodiments of the present disclosure have been described in detail with reference to the accompanying drawings, the present disclosure is not limited thereto and can be implemented in many different forms without departing from the technical concept of the present disclosure. Therefore, the various exemplary embodiments disclosed herein are not intended to limit the technical spirit of the invention, but are described in accordance with the true scope and spirit indicated by the following claims, and the scope of the technical spirit of the invention is not limited to the exemplary embodiments. Therefore, it should be understood that the above embodiments are intended to be illustrative in all senses and not restrictive. The scope of protection of the present invention should be determined by the appended claims, and all technical concepts within the scope of the appended claims should be understood to fall within the scope of this disclosure.

Claims

1. A method for digital determination of a target, the method comprising: Acquire multiple focused images of multiple droplets; Using an artificial neural network-based prediction model, at least one region is predicted based on the multiple focused images of the multiple droplets. The prediction model is configured to segment at least one region and a background region between positive droplets, negative droplets, and atypical droplets, using the images of the multiple droplets as input. The number of the plurality of droplets is determined based on the at least one region and the background region; as well as Quantitative data on the target are provided based on the number of the multiple droplets. The at least one region includes a positive droplet region, a negative droplet region, and an atypical droplet region. The positive droplet is defined as a droplet comprising the target and the fluorescent material. The negative droplets are defined as droplets or empty droplets consisting only of the fluorescent material, and The atypical droplets are defined as droplets comprising the target and the fluorescent material and having a shape different from that of the positive droplets. Determining the number of the plurality of droplets includes: Excluding the atypical droplet regions, the number of positive and negative droplets is determined based on the positive and negative droplet regions. The quantitative data provided for the target further include: The concentration of the target is determined based on the number of positive and negative droplets.

2. The method for digital measurement of a target according to claim 1, wherein acquiring images of multiple droplets includes: The plurality of microdroplets are allowed to flow into a chamber with a height of 1.1 to 1.9 times the diameter of the microdroplets, so that the plurality of microdroplets exist as a monolayer; as well as Obtain single-layer images of the multiple microdroplets, and The predicted at least one region includes: The prediction model is used to predict at least one region based on the single-layer image of the plurality of droplets.

3. The method for digital determination of a target according to claim 1, wherein determining the number of the positive droplets and the negative droplets comprises: Using Hough transform, the atypical droplet region is removed based on the shapes of the positive droplet region, the negative droplet region, and the atypical droplet region; as well as Determine the number of the positive and negative droplets.

4. The method for digital measurement of a target according to claim 1, wherein acquiring images of multiple droplets includes: Obtain an image of the plurality of droplets, which consist of an aggregate of multiple droplets, and The predicted at least one region includes: The prediction model is used to predict at least one region of the image of the plurality of droplets for each set of droplets in the plurality of droplets, and The quantitative data provided for the target include: The total copy number of the target is determined based on the predicted number of droplets for each droplet in the set of droplets.

5. The method for digital determination of a target according to claim 4, wherein predicting at least one region further comprises: The prediction model is used to predict at least one region of the plurality of droplets selected from a set of droplets. The quantitative data provided for the target further include: The copy number of the target is estimated based on the number of the multiple droplets predicted for the set of the single droplet.

6. The method for digital determination of a target according to claim 1, wherein the prediction model is further configured to quantitatively analyze the target based on the at least one region, and The quantitative data provided for the target further include: The prediction model is used to determine the quantitative data of the target based on the number of the plurality of droplets.

7. The method for digital determination of a target according to claim 1, wherein the image is a fluorescence image.

8. An apparatus for digital measurement of a target, the apparatus comprising: A light source is configured to irradiate light onto at least one surface of a chamber in which microdroplets including a target are disposed; An image sensor is configured to provide multiple focused images of multiple microdroplets in a chamber; as well as A processor operably connected to the image sensor, The processor is configured to predict at least one region based on multiple focused images of the plurality of droplets using an artificial neural network-based prediction model. The prediction model is configured to segment the at least one region and a background region between positive droplets, negative droplets, and atypical droplets, using the images of the droplets as input; determine the number of the plurality of droplets based on the at least one region and the background region; and determine quantitative data of the target based on the number of the plurality of droplets. The at least one region includes a positive droplet region, a negative droplet region, and an atypical droplet region. The positive droplet is defined as a droplet comprising the target and the fluorescent material. The negative droplets are defined as droplets or empty droplets consisting only of the fluorescent material, and The atypical droplets are defined as droplets comprising the target and the fluorescent material and having a shape different from that of the positive droplets. Determining the number of the plurality of droplets includes: excluding the atypical droplet regions, determining the number of positive droplets and negative droplets based on the positive droplet regions and the negative droplet regions, wherein providing quantitative data of the target further includes determining the concentration of the target based on the number of positive droplets and negative droplets.

9. The apparatus for digital measurement of a target according to claim 8, wherein, The height of the chamber is 1.1 to 1.9 times the diameter of the plurality of microdroplets; The processor is further configured to flow the plurality of microdroplets into the chamber, such that the plurality of microdroplets exist in a monolayer form; The image sensor is further configured to acquire single-layer images of the plurality of microdroplets; and The processor is further configured to use the prediction model to predict the at least one region based on the single-layer image of the plurality of droplets.

Images