Background

[0001] Cells may adjust their density during various processes such as cell cycle progression, differentiation, and/or disease state. Cell density can also be used to differentiate and separate cell type. Therefore, accurate measurements of the cell density distribution for a population may provide information about a patient's disease state, and/or inform separation of particular cells from a general population for various purposes. Additionally, changes in cell density may also be an indicator of cellular processes that may otherwise be undetectable by mass and/or volume measurements. In some instances, intrinsic cell-to-cell variation in density may be nearly 100-fold smaller than the mass or volume variation. As such, the ability to detect cell density in individual cells may be applicable in numerous applications, for example to monitor these cellular processes with precision.

Brief Description of the Drawings

[0002] Features of the present disclosure are illustrated by way of example and not limited in the following figure(s), in which like numerals indicate like elements.

[0003] Fig. 1 depicts an example of an image-based cell density measurement system configured with selected aspects of the present disclosure.

[0004] Fig. 2 depicts an example of an image-based cell density measurement system configured with a thermal inkjet pump and other selected aspects of the present disclosure.

[0005] Figs. 3 depicts an example of an image-based cell density measurement system configured with a partition and other selected aspects of the present disclosure.

[0006] Fig. 4 depicts an example of an imaging subsystem configured with selected aspects of the present disclosure.

[0007] Fig. 5 depicts an example method of determining a density of each of a plurality of cells configured with selected aspects of the present disclosure. Detailed Description

[0008] For simpiicity and illustrative purposes, the present disclosure is described by referring mainly to examples thereof. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. It will be readily apparent however, that the present disclosure may be practiced without limitation to these specific details. In other instances, some methods and structures have not been described in detail so as not to unnecessarily obscure the present disclosure. [0009] Additionally, it should be understood that the elements depicted in the accompanying figures may include additional components and that some of the components described in those figures may be removed and/or modified without departing from scopes of the elements disclosed herein. It should also be understood that the elements depicted in the figures may not be drawn to scale and thus, the elements may have different sizes and/or configurations other than as shown in the figures.

[0010] Some cell density measurements rely on buoyant force to transport cells to a location where the surrounding fluid density most closely equals the aggregate cell density. In other words, these processes measure an average cell density, rather than measuring the distribution of cell density with single cell resolution. Additionally, these processes may be performed with large volumes and involve additional processing, such as centrifugation. Other methods of measuring density at the single cell resolution may include use of microcantilevers; for example, as a cell passes through the microcantilever, it may alter the cantilever's mass and thus its vibrational frequency. This vibrational frequency may then be detected and correlated to the cell's density. However, this microcantilever process may be slow, which may result in low throughput.

[0011] Techniques, apparatus, and systems described herein are directed to determining a density of each of a plurality of cells using imaging. Generally, the system described herein measures cell densities with single cell resolution in such a way as to allow for high throughput. Examples described herein may utilize a microfluidic channel filled with a density gradient media, an imaging subsystem, and image analysis to determine individual cell density. In one implementation, a microfluidic channel includes transparent portion(s) to allow for imaging therethrough. The transparent portion(s) may be along an imaging axis, which in some cases, may be aligned with the gravity vector. The microfluidic channel may also include an intake that receives a density gradient media, which then flows through the microfluidic channel in a direction perpendicular to the imaging axis (and gravity, where applicable). As the media and cell mixture flows through the microfluidic channel, the density gradient media undergoes density stratification, creating a gradient in density through the media. The gradient may be aligned with the gravity vector.

[0012] The density gradient media and cells are mixed, for example through controlled release into the microfluidic channel by a pressure-controlled valve(s). Once mixed, the density gradient media induces a buoyancy force on the cells/media mixture. Each cell may then settle into a position within the density gradient where the cell's density is substantially equivalent to the density of the surrounding media. The density of the cell(s) is determined based on where within the density gradient media the cell is located. The density of the media may be predetermined based on vertical position, for example, through information provided by the media manufacturer. As such, each cell's density may be determined by its position as determined from imaging and image analysis. Such a system may allow for high throughput analysis of individual cell density by simultaneously and/or parallel imaging of cells at all positions in the microfluidic channel. The imaging system, discussed in detail herein, can then output a distribution of the cells density, shape, and other morphological parameters.

[0013] In one example, the imaging system may include an illumination device and an imaging sensor. In one example, the imaging sensor may image cells at a selected focal length. In such implementations, the density of the cells in focus at the selected focal length may be determined. In another example, the imaging system may utilize an image analysis process to reconstruct the three-dimensional volume of the channel and segment each cell from the volume in order to identify its three-dimensional position. In some implementations, for example, light-field microscopy may allow three- dimensional imaging to be obtained; for example, sub-second large volumetric imaging with approximately 1 pm spatial resolution may be obtained using light- field microscopy techniques. In some implementations, the imaging system may further include a lens; however, this is not intended to be limiting, as a lensless setup may also be used. As an example, (lensless) holographic imaging may be used, and may enable three-dimensional reconstruction of samples at extended depth of field by digitally refocusing to different reconstruction planes within a sample volume.

[0014] Image analysis may determine a density for each of the individual cells based on an image that captures the position of individual cells within the density gradient media. Such imaging analysis may distinguish different cell types from each other based on the cells’ morphology. This may allow individual cell density to be measured simultaneously in a mixed population of cell types.

[0015] In some implementations, the microfluidic channel may additional include partition(s) for separating the cells by density. In some such implementations, these partitions may be used to guide the collection of the cells into collection container(s) by density and/or by cell type. In other implementations, the cells may be collected into well plates, if desired. Other sizes of plates, flasks, and wells may also be used for collecting the cells.

[0016] Figure 1 is a schematic of an example an image-based cell density measurement system 100 that may be used to capture and analyze images of a plurality of cells and determine from the images an individual cell density. In some implementations, the system may contain density gradient media 102 in a first reservoir 104 and a plurality of cells 106 in a second reservoir 108. In an example, the cells 106 may be suspended in a neutral buffer in the second reservoir 108 while the density gradient 102 is in the first reservoir.

[0017] Depending on the cell type, the density of solution may be, for example, within the range of 0.8-1.2 g/ml, and may be adjusted by the percentage of density gradient stock in solution; however, this is not to be understood as limiting. An example of commercially available density gradient solutions is Ficoll-Paque PLUS (Cytiva®, Marlborough, MA), which is a common gradient for separating peripheral blood mononuclear cells (PBMCs) from serum and red blood cells. Other examples of commercially available density gradient solutions that may be used to separate peripheral blood mononuclear cells (PBMCs) from serum and red blood cells include, but are not limited to, Histopaque® and Accuspin™ (Sigma-Aldrich®, St. Louis, MO) may also be used peripheral blood mononuclear cells (PBMCs) from peripheral blood. Still other examples of density gradient solution, for example for viruses and other cells include, but are not limited to, HistoDenz™ (Sigma-Aldrich®, St. Louis, MO), OptiPrep™ (Sigma-Aldrich®, St. Louis, MO), and Percoll™ (Cytiva®, Marlborough, MA).

[0018] in some implementations, a pressure-controlled valve 110 may be coupled to each of the first reservoir 104, second reservoir 108, and a microfluidic channel 112. The pressure-controlled valve 110 may control the flow of the density gradient media 102 and plurality of cells 106 into the microfluidic channel 112, where the density gradient media 102 and a plurality of cells 106 may be mixed, in other implementations, the density gradient media 102 and a plurality of cells 106 may be mixed prior to entering the microfiuidic channel 112.

[0019] The density gradient media 102 and a plurality of cells 106, regardless of the location of their mixing, may flow through the microfluidic channel 112 in a direction perpendicular to a gravity vector (indicated by the “g” and arrow in Figure 1 ). The gradient media 102 and a plurality of cells 106 may flow through a length of microfluidic channel 112 allowing the density gradient media 102 to undergo density stratification with the gradient vector aligned with the gravity vector. The density gradient media 102 induces a buoyancy force on the cells 106 mixed into the media, which allows a cell 106 to find a position in the density gradient of the media where its density matches the density of the surrounding media. The density gradient media, when flowing through the channel undergoes density stratification, creating a gradient in density, with the gradient vector aligned with the gravity vector. The media may induce a buoyancy force on the ceils mixed into the media, until a ceil finds a position in the density gradient where its density matches the density of the surrounding media. The density of the media may be a known function of the vertical position (e.g. provided by the media manufacturer). This function may be used to calculate an individual cell's, of the plurality of cells, density from its vertical position as determined by imaging and image analysis.

[0020] In some implementations, the microfluidic channel 112 may include a transparent portion 114, for example to allow an image to be taken of the contents of the microfluidic channel 112. For example, in the implementation illustrated in Figure 1 , the microfluidic channel 112 has two transparent walls 114 ₁ and 114 ₂ , which may be substantially perpendicular to the gravity vector. In other implementations, the microfluidic channel 112 includes a length of transparent tubing.

[0021] The image-based cell density measurement system 100 may include an imaging subsystem 116, described in detail herein, in order to image the plurality of cells. A controller 118 may be used for analyzing the image data. The controller 118 receives data from the image subsystem 116. The controller 118 may temporarily store the data in a local memory for analysis. In some implementations the controller 118 may be physically a part of the imaging subsystem 116; in other implementations the controller 118 may be physically separate from the imaging subsystem 116 and be connected to the imaging subsystem either via wires or wirelessly.

[0022] The imaging subsystem 116 may include an image sensor 120 disposed proximate the microfluidic channel 112, for example aimed at the transparent portion, to acquire image data of the density gradient media 102 and the cells 106 in the microfluidic channel 112. In some implementations, the imaging subsystem 116 may capture images and image data at a speed fast enough so as to record the information while cells 106 _1-n are moving or flowing through the microfluidic channel 112. Additionally, in some implementations, the imaging subsystem 116 may capture images and image data with enough resolution so as to be able to detect individual cells. After passing through the imaging subsystem 116, the density gradient media 102 and the cells 106 may flow into a collection container 124.

[0023] In some implementations, the imaging subsystem is set up using light- field microscopy, for example as illustrated in Figure 1. Light-field microscopy may allow three-dimensional imaging to be obtained, for example, using sub- second large volumetric imaging with approximately 1 pm spatial resolution. In some implementations, imaging subsystem 116 may further include a microlens array 126, which may include a plurality of micro lenses (or "lenslets") that may each render a focused image of a portion of the imaged area. These microlens arrays 126 may, in some instances, eliminate the initial focusing of the lens and instead allow for the image to be focused using software in the post-processing steps. In other implementations, the imaging subsystem may be set up using (lensless) holographic imaging, which may enable three-dimensional (3D) reconstruction of samples at extended depth of field (DOF) by digitally refocusing to different reconstruction planes within a sample volume.

[0024] A variety of image-based cell counting techniques may be used with the image-based cell density measurement system 100 described herein, and the selected technique may depend on the imaging modality (e.g. brightfield, fluorescent, phase-contrast, etc.). For example, images with a uniform background, such as fluorescent channels, image processing methods such as global thresholding method (e.g. Otsu's method), local thresholding method (e.g. sliding window approaches), model-based method (e.g. assuming intensity distribution within a cell is Gaussian) may be used. Where the image(s) have a complex background, such as brightfield, phase-contrast, or there are cell clusters, a learning-based technique may be used.

[0025] In some instances, counting and/or categorizing cells in a three- dimensional volume with accuracy at an individual cell level may include, for example, three-dimensional object/cell segmentation and/or detection, in many instances, these three-dimensional techniques may be extended from methods developed and used for two-dimensional images.

[0026] Generally, the segmentation technique may classify all the voxels and generate a label volume where different types of cells have different labels, such as different intensity or color. For example, one deep learning-based technique that can be used for three-dimensional object segmentation may be a modified U-Net, which is a three-dimensional convolutional neural network (CNN) consisting of two paths: (1) a down-sampling path; and (2) up-sampling path, where each of the up-sampling and down-sampling includes some number of layers, such as five. Each layer may include, for instance, two three- dimensional convolutions, batch normalization, and a leaky rectified linear unit activation (Leaky ReLU). This two-stage process of down-sampling (e.g., achieved through max-pooling) and up-sampling may enable the network to extract and learn features that can be used in the segmentation and identification processes. Concatenation may be used to transfer information between the down-sampling path and corresponding up-sampling path.

[0027] At the end of the down-sampling and up-sampling path, a three- dimensional convolution, batch normalization, and/or three-dimensional sigmoid activation function may be used to classify whether each voxel belongs to the foreground (/.e. presence of a nuclei) or background (Le. no nuclei present). The output of these functions may be a binary-valued volume where each voxel indicates where the neural network has detected the location of the nuclei, in some implementations, machine learning models such as CNNs that are used with these deep learning-based techniques(s) may also be trained to segment multiple types of cells (e.g. in the case of a cell mixture), for example based on cell morphology. In such instances, the image data may be used to determine each cell's morphology and/or the type of cell.

[0028] A detection technique may, generally, identify the location of cells and report bounding boxes outside individual cells. Where multiple types of cells may detected, it may also output the cell type for each detected cell. An example three-dimensional volume cell or object detection technique is to use a neural network, for example 3D RetinaNet, which is an object detection framework designed for two-dimensional natural images. RetinaNet may also be extended for use in three-dimensional object detection. Another example approach for three-dimensional volume objection detection may be to use two- dimensional image methods to handle the three-dimensional volume slice-by- slice, and then fuse the slice-by-slice results back to a three-dimensional model. [0029] The machine learning models described herein (e.g., CNNs, U-Net, RetinaNet) or others not specifically mentioned (e.g., AlexNet, very deep CNN, etc.) may be trained in various ways. Some machine learning models may undergo supervised training using labeled ground truth data, such as images with pixel-wise or bounding box annotations. For example, a labeled image may be processed using the machine learning model that generates an inference or prediction. This inference or prediction may be compared to the image's label to determine an error. This error may be used to train the machine learning model, e.g., using techniques such as back propagation and gradient descent. In some implementations, limited available labeled training data may be augmented with "synthetic" training data that includes, for instance, variations of labeled images with noise added, computer-generated imagery (e.g., based on canonical cell models), etc. Other machine learning models may undergo unsupervised training, e.g., via an unsupervised loss function, auto-encoders (e.g., de-noising auto-encoders), data augmentation (e.g., using elastic deformations) to create surrogate classes, and so forth. Alternatively, some machine learning models may be trained using transfer learning.

[0030] The various image processing techniques described herein are not to be construed as limiting, as any applicable image processing technique known in the art may be used.

[0031] Returning to Figure 1 , an example method for determining a density for an individual cell 106 _1-n may include acquiring image data of the density gradient media 102 and cell(s) 106 _1-n in the microfluidic channel 112 through an image sensor 120 (or sensor array, as illustrated). The image sensor 120 may, in some implementations, image cells at a selected focal length. In some instances, the image sensor 120 and the portion of the microfluidic channel 112 disposed proximate to the image sensor 120 may be illuminated by an illumination device 122. In some implementations, such as illustrated in Figure 1 , the image sensor 120 and the illumination device 122 may be aligned across the microfluidic channel 112 from each other. Controller 118, which receives the image data from the image subsystem 116, may be used for analyzing the image data.

[0032] The controller 118 may then, in some implementations, determine a three-dimensional volume of the microfluidic channel 112 using the image data. Using the techniques described herein, the controller 118 may then segment the image(s) to count and identify a three-dimensional position of each of the individual cell 106, based on the image data. In some instances, these segments (or slices or layers) may be taken vertically through the microfluidic channel 112. Based on the three-dimensional position of the individual cell 106, the density of the individual cell 106 may be determined. In some implementations, a density plot illustrating the number of cells in each segment layer may be generated in order to determine the distribution(s) of cell across the microfluidic channel 112.

[0033] Figure 2 is a block diagram of another example image-based cell density measurement system 200 that may be used to capture and analyze images of a plurality of cells 206. The system 200 may include a pre-mixed mixture 208 of density gradient media and a plurality of cells in one container 210. Rather than controlling the flow of the mixture 208 of the density gradient media and plurality of cells downstream (e.g. using a pressure valve) as described with reference to Figure 1 , the system 200 of Figure 2 may utilize an upstream thermal inkjet nozzle(s) 204 to facilitate and/or control the flow of the mixture 208 of the density gradient media and plurality of cells through the microfluidic channel 212. In some implementations, the thermal inkjet nozzle(s) 204 may also eject single cells into a collection container 214, and the system 200 may further include multiple the thermal inkjet nozzle(s) 204 for sorting the cells 206 by density, type, morphology, etc.

[0034] The system 200, similar to the system of Figure 1 , may include an imaging subsystem 216, described in detail herein, in order to image the plurality of cells. As described with reference to Figure 1 , the image(s) and image data may then be analysed to determine a density for an individual cell in the density gradient media. [0035] Figure 3 is yet another example image-based cell density measurement system 300. Similar to the implementation described with reference to Figure 1 , the system 300 may contain a density gradient media 302 in a first reservoir 304, a plurality of cells 306 in a second reservoir 308, a pressure-controlled valve 310 coupled to each of the first reservoir 304, second reservoir 308, and a microfluidic channel 312. The pressure-controlled valve 310 may control the flow of the density gradient media 302 and plurality of cells 306 into the microfluidic channel 312, where the density gradient media 302 and a plurality of cells 306 may be mixed.

[0036] The density gradient media 302 and a plurality of cells 306 may flow through the microfluidic channel 312 in a direction perpendicular to a gravity vector (indicated with by the “g” and arrow). The mixture of media 302 and cells 306 may stratify by density along the gravity vector as they flow through a length of microfluidic channel 312. As described previously, the density of the media may be a known function of the vertical position (e.g. from the media manufacturer), and from this function an individual cell's density may be calculated from its vertical position based on imaging and image analysis (as described in detail with reference to Figure 1 ).

[0037] Also similar to the implementation described in Figure 1 , the microfluidic channel 312 may include a transparent portion(s) 314, through which an image sensor 320 of the image subsystem 316 acquires image data of the density gradient media 302 and the cells 306 in the microfluidic channel 112. A controller 318 may receive data from the image subsystem 316 and analyze the image data as described herein.

[0038] In some implementations, the system may include partition(s) 321 within the microfluidic channel 312. The partition(s) 321 may sort the individual cells into a collection container 322 that corresponds to the determined density. As the stratified cells 306 flow though the microfluidic channel 312 the partition(s) 321 physically divide the microfluidic channel into portions or regions so that the cells 306 may be sorted. Although two partitions 321 are illustrated in Figure 3, this is not to be understood as limiting. There may be as many (or as few) partitions as desired to sort the cells by density. Furthermore, the partitions 320 may be completely translucent, opaque, or any opacity in between. Additionally, while the collection containers 322 _1-n are illustrated as test tubes, this is also not to be understood as limiting. In other implementations, the collection containers 322 _1-n may be microcentrifuge tubes, well plates, or any other suitable container known in the art.

[0039] Figure 4 illustrates another implementation of an imaging subsystem 416. Similar to the previous implementations, the imaging subsystem 416 may include image sensor 420i- _n disposed proximate the microfluidic channel 412 to acquire image data of the density gradient media 402 and the cells 406 in the microfluidic channel 412. The imaging subsystem 416 may additionally include illumination devices 422 _1-n to illuminate the microfluidic channel 412 and image sensor(s) 420 _1-n . The use of multiple image sensors 420 _1-n and illumination devices 422 _1-n may allow for imaging across multiple wavelengths, which may be particularly useful trying to differentiate cell type using cells tagged with fluorescent antibodies.

[0040] Filters 424, 426 in fluorescent imaging may be used to separate different wavelengths of light in order to permit selective excitation of specific fluorophores and also allow for the subsequent separation of the emission and excitation light, which may result in high contrast imaging. In some implementations, the filters 424, 426 allow for the fluorescing structures (i.e. cells) to have high contrast against a very dark or black background. As an example, fluorescent antibody (FA) techniques may attach a fluorescent marker (fluorogen) to the constant region of an antibody, which results in a molecule that may to bind to a target (e.g. a particular cell or cell type). The correct absorption wavelength is needed in order to excite the fluorophore tag attached to the antibody and detect the fluorescence given off.

[0041] The imaging subsystem 416 may include both excitation (or illumination) filter(s) 424 _1-n and emission (or collection) filter(s) 426 _1-n . The excitation filter(s) 424 _1-n may allow specific wavelengths of light (excitation wavelengths) to pass through the filter towards the tagged cells 406. The excitation filter(s) 424 _1-n work by only allowing a specific (typically narrow) band of wavelengths to pass through; this wavelength band may be around the peak fluorophore excitation wavelength. The emission filters(s) 426 _1-n attenuate the excitation wavelengths while simultaneously allowing a selected emission wavelength to pass through the emission filter(s) 426 _1-n to pass towards image sensor 420 _1-n . The emission filter(s) 426 _1-n may allow the desirable fluorescence (e.g. the fluorescence emitted from the cell 406) to reach the image sensor 420 _1-n while simultaneously blocking as much of the unwanted traces of excitation light (e.g. from the illumination device 422 _1-n ) as possible. Similar to the excitation filter 424 _1-n , the emission filter 426 _1-n also allows a narrow band of wavelengths to pass through it, for example, the peak fluorophore emission wavelength.

[0042] In some implementations, the use of fluorescent tagging on the cell(s), for example through the use of fluorescent antibody marker(s), may allow for a determination of cell type(s) to be made simultaneously with the determination of cell density.

[0043] The systems 100, 200, 300 described herein, for example and as described with reference to Figures 1-4, may be able to provide high throughput imaging and analysis. This high throughput may be achieved by simultaneously flowing many cells through the system (e.g. in parallel) and simultaneously imaging them at all vertical positions in the microfluidic channel 112, 212, 312. Using this simultaneous imaging and analysis thereof, the system may be able to output a distribution of the cells’ density, type, and/or other morphological parameters.

[0044] Figure 5 illustrates a flowchart of an example method 500 determining a density of each of a plurality of cells in accordance with the present disclosure. It is to be understood that, in some instances, the method may include additional operations than those illustrated in Figure 5, may perform operations of Figure 5 in a different order and/or in parallel, and/or may omit various operations of Figure 5.

[0045] In some implementations, the method 500 may include, at block 505, mixing of the density gradient media and the plurality of cells upstream of the imaging system. In some such implementations, this mixing may occur in the microfluidic channel. In other implementations, the mixing may occur prior to entry into the microfluidic channel.

[0046] At block 510, an imaging system, such as the imaging system(s) 116, 216, 316, 416 described with reference to Figures 1-4, may capture image(s) and data associated therewith (collectively referred to herein as image data) of a density gradient media and the plurality of cells flowing through a microfluidic channel. At block 515, a density for an individual cell of the plurality of cells may be determined based on the image data.

[0047] In some implementations, determining the density for the individual cell may include: at block 520, constructing a three-dimensional volume of the microfluidic channel based on the image data; and at block 525, segmenting the individual cell of the plurality of cells, based on the image data, in order to identify a three-dimensional position of each of the individual cells. Based on the three-dimensional position of the individual position (e.g. within the density gradient media), the density of the individual cell may be determined (block 515).

[0048] At block 530, the plurality of cells may be sorted, for example into a plurality of collection containers, where each collection container corresponds to a different density, cell type, and/or cell morphology.

[0049] Although described specifically throughout the entirety of the instant disclosure, representative examples of the present disclosure have utility over a wide range of applications, and the above discussion is not intended and should not be construed to be limiting, but is offered as an illustrative discussion of aspects of the disclosure.

[0050] What has been described and illustrated herein is an example of the disclosure along with some of its variations. The terms, descriptions and figures used herein are set forth by way of illustration and are not meant as limitations. Many variations are possible within the scope of the disclosure, which is intended to be defined by the following claims -- and their equivalents - in which all terms are meant in their broadest reasonable sense unless otherwise indicated.