Methods and systems for cytometry

Abstract

To provide methods and systems for ghost cytometry which can be used to produce an image of an object without using a spatially resolving detector.SOLUTION: Spatial information acquired from motion of cells relative to a patterned optical structure can be compressively converted into signals arriving sequentially at a single pixel detector. Combinatorial use of a temporal waveform with an intensity distribution of a random or pseudo-random pattern allows for computational reconstruction of cell morphology. Machine learning methods can be applied directly to compressed waveforms without image reconstruction for enabling efficient image-free morphology-based cytometry. Image-free ghost cytometry (GC) can achieve accurate and high throughput cell classification as well as selective sorting based on cell morphology not accompanied by a specific biomarker, which have been difficult using a conventional flow cytometer.SELECTED DRAWING: Figure 1A

Description

Technical Field

[0001] Cross-reference

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 62 / 684,612, filed Jun. 13, 2018, entitled "METHODS AND SYSTEMS FOR CYTOMETRY"; U.S. Provisional Patent Application No. 62 / 701,395, filed Jul. 20, 2018, entitled "METHODS AND SYSTEMS FOR CYTOMETRY"; U.S. Provisional Patent Application No. 62 / 804,560, filed Feb. 12, 2019, entitled "METHODS AND SYSTEMS FOR CYTOMETRY"; and U.S. Provisional Patent Application No. 62 / 848,478,478, filed May 15, 2019, entitled "METHODS AND SYSTEMS FOR CYTOMETRY", each of which is hereby incorporated by reference in its entirety for all purposes.

Background Art

[0002]

[0002] Flow cytometry is a technique that can be used for cell counting, cell sorting, biomarker detection, and protein engineering. Flow cytometry can be performed by suspending cells in a fluid stream and passing the cells through an electronic detection device. A flow cytometer can enable simultaneous multi-parameter analysis of the physical and chemical properties of particles.

Summary of the Invention

Problems to be Solved by the Invention

[0003]

[0003] The present disclosure provides methods and systems for morphology-based cell classification and sorting using a process called ghost cytometry (GC). The methods and systems of the present disclosure can be used to achieve morphology-based cell classification and sorting with high accuracy and throughput without acquiring images.

Means for Solving the Problems

[0004]

[0004] In one aspect, the present disclosure provides a method for particle processing or analysis, comprising: (a) obtaining spatial information during movement of particles relative to a patterned optical structure; (b) compressively transforming the spatial information into signals sequentially arriving at a detector; (c) using the signals to identify at least a subset of the particles with an arbitrary accuracy of at least 70%; and (d) sorting at least the subset of the particles identified in (c) into one or more groups at a rate of at least 10 particles per second. In some embodiments, (a) comprises: (i) directing light from a light source through a randomly or pseudo-randomly patterned optical structure; (ii) directing light from the patterned optical structure at the particles; and (iii) directing light from the particles at the detector. In some embodiments, (a) comprises: (i) directing light from a light source at the particles; (ii) directing light from the particles through a patterned optical structure; and (iii) directing light from the patterned optical structure at the detector. In some embodiments, the light comprises ultraviolet or visible light. In some embodiments, (b) comprises applying a two-step iterative shrinkage / thresholding (TwIST) procedure. In some embodiments, (c) comprises computationally reconstructing at least in part the morphology of the particles by using a combination of temporal waveforms including one or more intensity distributions imparted by the patterned optical structure. In some embodiments, the particles comprise one or more biological particles. In some embodiments, the particles comprise one or more cells. In some embodiments, the particles comprise one or more rare cells. In some embodiments, the particles comprise one or more cancer cells. In some embodiments, the particles comprise one or more circulating tumor cells. In some embodiments, (c) comprises applying one or more machine learning classifiers to the compressed waveforms corresponding to the signals to identify at least a subset of the particles.In some embodiments, one or more machine learning classifiers are selected from the group consisting of support vector machines, random forests, artificial neural networks, convolutional neural networks, deep learning, ultra-deep learning, gradient boosting, AdaBoost, decision trees, linear regression, and logistic regression. In some embodiments, the particles are analyzed without image reconstruction. In some embodiments, the detector includes a single pixel detector. In some embodiments, the single pixel detector comprises a photomultiplier tube. In some embodiments, the method further comprises the step of reconstructing one or more images of the particles. In some embodiments, the one or more images include fluorescence images. In some embodiments, the method further comprises the step of reconstructing a plurality of images of the particles, each of the plurality of images including a different wavelength or wavelength range. In some embodiments, the one or more images are free of blur artifacts. In some embodiments, the particles move at a speed of at least 1 m / s relative to the patterned optical structure. In some embodiments, the method further comprises the step of sorting the particles into one or more groups of selected particles based on the morphology of the particles. In some embodiments, the speed is at least 100 particles per second. In some embodiments, the speed is at least 1,000 particles per second. In some embodiments, (d) includes the step of collecting one or more groups to generate a concentrated particle mixture. In some embodiments, the one or more groups have a purity of at least 70%. In some embodiments, the method further comprises the step of subjecting one or more particles of the one or more groups to one or more assays. In some embodiments, the one or more assays are selected from the group consisting of lysis, nucleic acid extraction, nucleic acid amplification, nucleic acid sequencing, and protein sequencing. In some embodiments, the method includes subjecting the particles to hydrodynamic flow focusing prior to (a). In some embodiments, the method includes collecting a partial transmission speckle pattern of the particles as the particles move relative to the patterned optical structure.In some embodiments, the patterned optical structure includes a regular patterned optical structure. In some embodiments, the patterned optical structure includes a disordered patterned optical structure. In some embodiments, the disordered optical structure includes an aperiodic patterned optical structure. In some embodiments, the disordered optical structure includes an optically structured pattern that is patterned randomly or pseudo-randomly. In some embodiments, the optical structure includes a static optical structure.

[0005]

[0005] In another aspect, the present disclosure provides an image-free method for classifying or sorting particles based at least in part on the morphology of the particles without using non-natural labels with at least 70% accuracy. In some embodiments, the particles include one or more biological particles. In some embodiments, the particles include one or more cells. In some embodiments, the particles include one or more rare cells. In some embodiments, the particles include one or more cancer cells. In some embodiments, the particles include one or more circulating tumor cells.

[0006]

[0006] In another aspect, the present disclosure provides a system for particle processing or analysis comprising a fluid flow path configured to direct particles, a detector in sensing communication with at least a portion of the fluid flow path, and one or more computer processors operably coupled to the detector, the one or more computer processors being programmed to: (a) acquire spatial information during movement of the particles relative to an optically structured pattern that is patterned randomly or pseudo-randomly; (b) compressively transform the spatial information into signals arriving sequentially at the detector; (c) use the signals to identify at least a subset of the particles with at least 70% accuracy; and (d) individually or collectively sort at least a subset of the particles identified in (c) into one or more groups at a rate of at least 10 particles per second. In some embodiments, the detector is a single pixel detector.

[0007]

[0007] In another aspect, the present disclosure provides a non-transitory computer-readable medium including machine-executable code that, when executed by one or more computer processors, implements a method for particle analysis, the method comprising: (a) obtaining spatial information during movement of particles relative to an optically structured pattern that is randomly patterned; (b) compression-transforming the spatial information into signals that sequentially reach a single pixel detector; (c) using the signals to identify at least a subset of the particles with at least 70% accuracy; and (d) sorting at least a subset of the particles identified in (c) into one or more groups at a rate of at least 10 particles per second.

[0008]

[0008] In another aspect, the present disclosure provides a method for cell processing or analysis, the method comprising: (a) obtaining spatial information during movement of cells relative to a patterned optical structure; (b) compression-transforming the spatial information into signals that sequentially reach a detector; (c) using the signals to identify at least a subset of the cells as being cancerous; and (d) sorting at least a subset of the cells identified in (c) into one or more groups of cancerous cells and one or more groups of non-cancerous cells.

[0009]

[0009] In another aspect, the present disclosure provides a method for cell processing or analysis, the method comprising: (a) obtaining spatial information during movement of cells relative to a patterned optical structure; (b) compression-transforming the spatial information into signals that sequentially reach a detector; (c) using the signals to identify at least a subset of the cells as being therapeutic; and (d) sorting at least a subset of the cells identified in (c) into one or more groups of therapeutic cells and one or more groups of non-therapeutic cells.

[0010] In another aspect, the present disclosure provides a method for identifying one or more target cells from a plurality of cells, the method comprising: (a) obtaining spatial information during the movement of the plurality of cells relative to a patterned optical structure; and (b) inputting the spatial information into a trained machine learning algorithm to identify one or more target cells from the plurality of cells.

[0011] In another aspect, the present disclosure provides a method for cell processing, the method comprising: (a) obtaining spatial information of a plurality of cells; and (b) separating or isolating a subset of the plurality of cells from the plurality of cells at a rate of at least 1,000 cells per second using at least the spatial information.

[0012] In another aspect, the present disclosure provides a method for processing one or more target cells from a plurality of cells, the method comprising: (a) obtaining spatial information during movement of the plurality of cells relative to a patterned optical structure; (b) using the spatial information to identify one or more target cells from the plurality of cells; and (c) separating or isolating one or more target cells from the plurality of cells at a rate of at least 10 cells per second, based at least in part on the one or more target cells identified in (b). In some embodiments, (a) comprises: (i) directing light from a light source through the patterned optical structure; (ii) directing light from the patterned optical structure towards the plurality of cells; and (iii) directing light from the plurality of cells towards a detector. In some embodiments, (a) comprises: (i) directing light from a light source towards the plurality of cells; (ii) directing light from the plurality of cells through the patterned optical structure; and (iii) directing light from the patterned optical structure towards a detector. In some embodiments, the patterned optical structure comprises a disordered patterned optical structure. In some embodiments, (c) comprises computationally reconstructing the cell morphology at least in part by using a combination of one or more temporal waveforms including one or more intensity distributions imparted by the patterned optical structure. In some embodiments, the target cells include one or more cancer cells or circulating tumor cells. In some embodiments, the target cells include one or more therapeutic cells. In some embodiments, the target cells include one or more members selected from the group consisting of stem cells, mesenchymal stem cells, induced pluripotent stem cells, embryonic stem cells, cells differentiated from induced pluripotent stem cells, cells differentiated from embryonic stem cells, genetically engineered cells, blood cells, red blood cells, white blood cells, T cells, B cells, natural killer cells, chimeric antigen receptor T cells, chimeric antigen receptor natural killer cells, cancer cells, and blast cells. In some embodiments, (b) comprises applying one or more machine learning classifiers to a compressed waveform corresponding to the spatial information to identify the one or more target cells.In some embodiments, one or more machine learning classifiers achieve one or more of sensitivity, specificity, and at least 70% accuracy. In some embodiments, one or more machine learning classifiers are selected from the group consisting of support vector machines, random forests, artificial neural networks, convolutional neural networks, deep learning, ultra-deep learning, gradient boosting, AdaBoost, decision trees, linear regression, and logistic regression. In some embodiments, the plurality of cells are processed without image reconstruction. In some embodiments, the detector includes a single pixel detector. In some embodiments, the single pixel detector comprises a photomultiplier tube. In some embodiments, the method further comprises the step of reconstructing one or more images of the plurality of cells. In some embodiments, the method further comprises the step of reconstructing a plurality of images of the plurality of cells, each of the plurality of images including a different wavelength or wavelength range. In some embodiments, one or more of the images are free of blur artifacts. In some embodiments, the plurality of cells move at a speed of at least 1 m / s relative to the patterned optical structure. In some embodiments, (c) includes (i) sorting the plurality of cells into one or more groups of sorted cells based on the results of analyzing the plurality of cells, and (ii) collecting one or more target cells from the one or more groups of sorted cells. In some embodiments, (c) includes sorting the plurality of cells into one or more groups of sorted cells based on the morphology of the plurality of cells. In some embodiments, the sorting is achieved at a rate of at least 10 cells / second. In some embodiments, the method further comprises collecting one or more groups of sorted cells to generate a concentrated cell mixture. In some embodiments, the one or more groups of sorted cells have a purity of at least 70%. In some embodiments, the method further comprises subjecting one or more cells of the one or more groups of sorted cells to one or more assays.In some embodiments, one or more assays are selected from the group consisting of lysis, nucleic acid extraction, nucleic acid amplification, nucleic acid sequencing, and protein sequencing. In some embodiments, the method further comprises, prior to (a), subjecting the cells to hydrodynamic flow focusing. In some embodiments, the method further comprises collecting a partial transmission speckle pattern of the plurality of cells as the plurality of cells move relative to a patterned optical structure. In some embodiments, the spatial information corresponds to features, characteristics, or information regarding the plurality of cells. In some embodiments, the spatial information corresponds one-to-one with features, characteristics, or information regarding the plurality of cells. In some embodiments, the features, characteristics, or information regarding the plurality of cells include one or more members selected from the group consisting of: metabolic state, growth state, differentiation state, maturation state, expression of marker proteins, expression of marker genes, cell morphology, organelle morphology, organelle location, organelle size or extent, cytoplasm morphology, cytoplasm location, cytoplasm size or extent, nucleus morphology, nucleus location, nucleus size or extent, mitochondria morphology, mitochondria arrangement, mitochondria size or extent, lysosome morphology, lysosome arrangement, lysosome size or extent, distribution of intracellular molecules, distribution of intracellular peptides, polypeptides or proteins, distribution of intracellular nucleic acids, distribution of intracellular saccharides or polysaccharides, and distribution of intracellular lipids.

[0013]

[0013] Another aspect of the present disclosure provides a non-transitory computer-readable medium including machine-executable code that, when executed by one or more computer processors, performs any of the methods described above or elsewhere in this specification.

[0014]

[0014] Another aspect of the present disclosure provides a system comprising one or more computer processors and a computer memory coupled thereto. The computer memory includes machine-executable code that, when executed by one or more computer processors, performs any of the methods described above or elsewhere in this specification.

[0015]

[0015] Additional aspects and advantages of the present disclosure will become readily apparent to those skilled in the art from the following detailed description, which illustrates only exemplary embodiments of the present disclosure. As will be understood, the present disclosure is capable of other and different embodiments, and some of the details thereof are capable of modification in various obvious respects, all without departing from the present disclosure. Accordingly, the drawings and description are to be regarded as illustrative in nature and not as restrictive.

[0016] Incorporation by reference

[0016] All publications, patents, and patent applications mentioned in this specification are hereby incorporated by reference into this specification to the same extent as if each individual publication, patent, or patent application was specifically and individually indicated to be incorporated by reference.

[0017]

[0017] The novel features of the invention are particularly set forth in the appended claims. A better understanding of the features and advantages of the present invention will be obtained from the following detailed description, which illustrates exemplary embodiments in which the principles of the present invention are utilized, and from the appended drawings (also referred to herein as "Figure" and "FIG.").

Brief description of the drawings

[0018]

Figure 1A

[0018] A schematic diagram of an optical compressive sensing process used in a ghost cytometry (GC) process is shown.

Figure 1B

[0019] A flowchart of an example of a particle analysis method is shown.

Figure 1C

[0020] A flowchart of an example of an image-free optical method for classifying or sorting particles is shown.

Figure 2A

[0021] An example of an optical setup for motion-based compressive fluorescence imaging with a structured illumination (SI) element is shown.

Figure 2B

[0022] An example of an optical setup for motion-based compressed fluorescence imaging with a structured detection (SD) element is shown.

Figure 2C

[0023] An example of a time waveform related to moving fluorescent beads obtained using an optical system including an SI element is shown.

Figure 2D

[0024] An example of a two-dimensional (2D) fluorescence image of moving fluorescent beads computationally reconstructed from the time waveform obtained by an optical system including an SI element is shown.

Figure 2E

[0025] An example of a time waveform related to moving fluorescent beads obtained using an optical system including an SD element is shown.

Figure 2F

[0026] An example of a 2D fluorescence image of moving fluorescent beads computationally reconstructed from the time waveform obtained by an optical system including an SD element is shown.

Figure 2G

[0027] An example of a fluorescence image of moving fluorescent beads obtained using an array pixel camera is shown.

Figure 3A

[0028] An optical setup for multi-color motion-based compressed fluorescence imaging with an SD element is shown.

Figure 3B

[0029] An example of a time waveform related to moving fluorescent cells obtained using an optical setup for multi-color motion-based compressed fluorescence imaging is shown.

Figure 3C

[0030] An example of a 2D fluorescence image of moving fluorescent cells computationally reconstructed from the time waveform obtained by an optical setup for multi-color motion-based compressed fluorescence imaging is shown.

Figure 3D

[0031] An example of multi-color sub-millisecond fluorescence imaging of cells flowing at a throughput speed exceeding 10,000 cells / second is shown.

Figure 4A

[0032] A procedure for training a classifier model applied to a GC signal is shown.

Figure 4B

[0033] An example of a procedure for testing a classifier model is shown.

Figure 4C

[0034] Examples of the concentration ratios of MCF-7 cells in various samples are shown.

Figure 4D

[0035] An example of the correlation between the true positive rate and the false positive rate related to a classifier model is shown.

Figure 4E

[0036] An example of the correlation between the true positive rate and the false positive rate related to a classifier model based on the ability to classify model cancer (MCF-7) cells against a complex mixture of peripheral blood mononuclear cells is shown.

Figure 5A

[0037] An example of a microfluidic device for use with the systems and methods of the present disclosure is shown.

Figure 5B

[0038] Accurate isolation of MIA PaCa-2 cells against morphologically similar MCF-7 cells using the systems and methods of the present disclosure is shown.

Figure 5C

[0039] An example of the accurate isolation of model cancer (MIA PaCa-2) cells against a complex mixture of peripheral blood mononuclear cells (PBMC) using the systems and methods of the present disclosure is shown.

Figure 6

[0040] An example of an optical system for estimating the intensity distribution of an optical encoder pattern such as an SI element or an SD element is shown.

Figure 7

[0041] Examples of the optical setup and calibrated intensity distribution of an optical encoder used to demonstrate GC-based multicolor fluorescence imaging of cells are shown.

Figure 8

[0042] Examples of the optical setup and calibrated intensity distribution of an optical encoder used for ultra-fast multicolor fluorescence imaging and image-free imaging cytometry by GC are shown.

Figure 9

[0043] Examples of the geometric shapes of an example of a flow cell for hydrodynamic three-dimensional (3D) focusing of a fluid for use with the systems and methods described herein are shown. 『Figure 9B』

[0044] Examples of loosely and tightly focused flows are shown. 『Figure 9C』

[0045] Examples of the accuracy of a machine learning classifier for classifying cells using a GC signal acquired at a flow rate of 20 μL / min, when the classifier is trained using the GC signal acquired at a flow rate of 20 μL / min, are shown. 『Figure 9D』

[0046] Examples of the accuracy of a machine learning classifier for classifying cells using a GC signal acquired at a flow rate of 30 μL / min, when the classifier is trained using the GC signal acquired at a flow rate of 30 μL / min, are shown. 『Figure 9E』

[0047] Examples of the accuracy of a machine learning classifier for classifying cells using a GC signal acquired at a flow rate of 30 μL / min, when the classifier is trained using the GC signal acquired at a flow rate of 20 μL / min, are shown. 『Figure 9F』

[0048] Examples of the accuracy of a machine learning classifier for classifying cells using a GC signal acquired at a flow rate of 20 μL / min, when the classifier is trained using GC signals acquired at flow rates of 20 μL / min and 30 μL / min, are shown.

Figure 10A

[0049] An example of a histogram of the total blue fluorescence intensity of mCF-7 cells is shown.

Figure 10B

[0050] An example of a histogram of the total blue fluorescence intensity of mIA PaCa-2 cells is shown.

Figure 10C

[0051] An example of a histogram of the total blue fluorescence intensity of an example of a mixture of mCF-7 and MIA PaCa-2 cells is shown.

Figure 10D

[0052] An example of a histogram of the total green fluorescence intensity of an example of a mixture of mCF-7 and MIA PaCa-2 cells is shown.

Figure 11

[0053] An example of using a support vector machine (SVM)-based machine learning classifier to confirm consistent classification accuracy over a wide range of concentration ratios is shown.

Figure 12A

[0054] An example of a cell sorting chip including a flow focusing segment and a cell sorting segment fabricated using soft lithography technology is shown.

Figure 12B

[0055] An example of a microscopic image of a flow focusing segment is shown.

Figure 12C

[0056] An example of a microscopic image of a cell sorting segment is shown.

Figure 12D

[0057] An example of a process flow for a real-time classification and electrical control system for use with the systems and methods of the present disclosure is shown.

Figure 13

[0058] An example of a computer system programmed or otherwise configured to implement the methods and systems of the present disclosure is shown.

Figure 14

[0059] An example of a label-free cell sorter using GC is shown.

Figure 15

[0060] An example of kinematic drive compression GC for a label-free cell sorter is shown.

Figure 16A

[0061] An example of sorting of Jurkat and MIA PaCa-2 cells using a label-free cell sorter is shown.

Figure 16B

[0062] Examples of histograms of Jurkat and MIA PaCa-2 cell populations before and after cell sorting are shown.

Figure 17A

[0063] An example of a scatter plot of the fluorescence intensities of propidium iodide (PI) and annexin V for classifying induced pluripotent stem cells (iPSCs) as live, early apoptotic, or dead is shown.

Figure 17B

[0064] An example of a scatter plot of forward scatter (FSC) and side scatter (SSC) for iPSCs is shown.

Figure 17C

[0065] Examples of receiver operating characteristic (ROC) curves and SVM score histograms for the classification of live and dead cells are shown.

Figure 17D

[0066] Examples of ROC curves and SVM score histograms for the classification of live and early apoptotic cells are shown.

Figure 17E

[0067] Examples of ROC curves and SVM score histograms for the classification of dead cells and early apoptotic cells are shown.

Figure 18A

[0068] Examples of scatter plots of calcein AM and rBC2LCN-635 intensities for mixtures of neural progenitor cells (NPCs) and iPSCs used to train a machine learning classifier are shown.

Figure 18B

[0069] Examples of ROC curves and SVM score histograms for the classification of NPCs and iPSCs are shown.

Figure 18C

[0070] Examples of scatter plots of calcein AM and rBC2LCN-635 intensities for mixtures of hepatoblasts and iPSCs used to train a machine learning classifier are shown.

Figure 18D

[0071] Examples of ROC curves and SVM score histograms for the classification of hepatoblasts and iPSCs are shown.

Figure 19A

[0072] Examples of histograms of fab FITC intensity of T cells are shown.

Figure 19B

[0073] Examples of ROC curves and SVM score histograms for classifying different states in the cell cycle are shown.

Figure 19C

[0074] Examples of histograms of 2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose (2-NBDG) intensity of raji cells are shown.

Figure 19D

[0075] Examples of ROC curves and SVM score histograms for classifying cells with high and low levels of glucose are shown.

Figure 20A

[0076] A training dataset containing a population of neutrophils is shown.

Figure 20B

[0077] Examples of label-free optical signals corresponding to neutrophils and non-neutrophils are shown.

Figure 20C

[0078] An example of an SVM score histogram for the classification of eosinophils and non-eosinophils is shown.

Figure 20D

[0079] A training dataset including a population of eosinophils is shown.

Figure 20E

[0080] Examples of label-free optical signals corresponding to eosinophils and non-eosinophils are shown.

Figure 20F

[0081] An example of an SVM score histogram for the classification of eosinophils and non-eosinophils is shown.

Figure 20G

[0082] A training dataset including a population of basophils is shown.

Figure 20H

[0083] Examples of label-free optical signals corresponding to basophils and non-basophils are shown.

Figure 20I

[0084] An example of an SVM score histogram for the classification of basophils and non-basophils is shown.

Figure 21A

[0085] An example of an SVM score histogram for the classification of oral cells and HeLa cells is shown.

Figure 21B

[0086] An example of an SVM score histogram for the classification of BM1 cells and K562 cells is shown.

Figure 22A

[0087] Examples of signals caused by the first type of particles, signals caused by the second type of particles, and signals caused by the simultaneous presence of the first and second types of particles are shown.

Figure 22B

[0088] An example of a confusion matrix for distinguishing different types of particles simultaneously present in the detection region of the present disclosure is shown.

Mode for Carrying Out the Invention

[0019]

[0089] Although various embodiments of the present invention are shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art can conceive of numerous variations, modifications, and substitutions without departing from the present invention. It should be understood that various alternative forms of the embodiments of the present invention described herein can be used.

[0020]

[0090] When the terms "at least", "greater than", or "above" precede a first numerical value in a series of two or more numerical values, the terms "at least", "greater than", or "above" always apply to each of the numerical values in that series. For example, 1, 2, or 3 or more is equivalent to 1 or more, 2 or more, or 3 or more.

[0021]

[0091] Whenever the terms "below", "less than", "under", or "up to" precede a first numerical value in a series of two or more numerical values, the terms "below", "less than", "under", or "up to" apply to each of the numerical values in that series. For example, 3, 2, or 1 or less is equivalent to 3 or less, 2 or less, or 1 or less.

[0022]

[0092] As used herein, the term "optical spatial information" generally refers to information derived from an optical sensing process related to the spatial arrangement of parts or components of an object. For example, "optical spatial information" can refer to the spatial arrangement or organization of intracellular cell components (such as organelles, membranes, or cytoplasm), the spatial arrangement or organization of cells within a tissue, the spatial arrangement or organization of microstructures within a material, and the like.

[0023]

[0093] As used herein, the terms "compressed sensing," "compressed sampling," "sparse sensing," and "sparse sampling" generally refer to signal processing techniques for efficiently acquiring and reconstructing signals containing optical spatial information by reconstructing the optical spatial information from fewer samples of the signal than required by the Nyquist-Shannon sampling theorem. Compressed sampling can utilize the sparsity of optical spatial information in several domains, such as the time domain, time-frequency domain, spatial domain, spatial-frequency domain, wavelet transform domain, discrete cosine transform domain, or discrete sine transform domain.

[0024]

[0094] As used herein, the term "random" generally refers to a process lacking a pattern or predictability, or the product of such a process. A random process may not have a recognizable order and / or may not follow an easily understandable pattern. A random process can include a sequence of events whose outcomes do not follow a deterministic pattern. A random process can include a sequence of events whose outcomes follow a probabilistic or stochastic pattern.

[0025]

[0095] As used herein, the term "pseudo-random" generally refers to a deterministic or partially deterministic process that generates a nearly random product, or the product of such a process.

[0026]

[0096] The present disclosure provides methods and systems for biological analysis. Such methods and systems can be used to analyze particles such as cells, or biological materials such as proteins or nucleic acids.

[0027]

[0097] Particles analyzed and / or processed using the methods described herein may be biological particles such as cells. The cells can be of any type and can have any properties. The cells can have any size and volume. The cells can be derived from any possible source. For example, the cells can be human cells, animal cells, non-human primate cells, horse cells, pig cells, dog cells, cat cells, mouse cells, plant cells, bacterial cells, or other cells. The cells can be derived from tissues or organs. For example, the cells can be derived from the heart, arteries, veins, brain, nerves, lungs, spine, spinal cord, bone, connective tissue, trachea, esophagus, stomach, small intestine or large intestine, bladder, liver, kidney, spleen, urethra, ureter, prostate, vas deferens, penis, ovary, uterus, endometrium, fallopian tube, or vagina, or any tissue associated with any of the foregoing. The cells can be cancerous or suspected of being cancerous (e.g., tumor cells). For example, the cells can be derived from tumor tissue. The cells can contain natural or non-natural components. For example, the cells can contain nucleic acids such as deoxyribonucleic acid (DNA) or ribonucleic acid (RNA), proteins, or carbohydrates. The cells can contain one or more optically detectable elements such as one or more fluorophores. The fluorophore can be natural or non-natural to the cell. For example, the fluorophore can be a non-natural fluorophore introduced into the cell by one or more cell staining or labeling techniques, etc.

[0028]

[0098] In some cases, multiple particles can be analyzed using the methods described herein. The particles of the multiple particles can be derived from the same source or different sources. The particles can have the same or different properties. For example, the first particle can have a first size, the second particle can have a second size, and the first size and the second size are not the same. The methods and systems described herein can be used to characterize and / or identify multiple particles. For example, the methods and systems can characterize the first particle and the second particle, provide information regarding the size and morphology of the particles, and thereby identify the first particle as being different from the second particle.

[0029]

[0099] Imaging and analysis of many single cells hold the potential to substantially increase the understanding of heterogeneous systems involved in immunology (1), cancer (2), neuroscience (3), hematology (4), and development (5). Many important applications in these fields require the accurate and high-throughput isolation of specific populations of cells according to the information contained in high-content images. This poses several challenges. First, despite recent developments (6 - 10), it remains difficult to simultaneously meet the requirements of high sensitivity, multi-colorability, high shutter speed, high frame rate, continuous acquisition, and low cost. Second, ultra-fast and continuous image acquisition then requires computer-based image reconstruction and analysis that is costly both in terms of time and expense (11). Because of such challenges, fluorescence imaging-activated cell sorting (FiCS) has not been achieved prior to the methods and systems disclosed herein. Disclosed herein are methods and systems that can apply machine learning methods to compressed imaging signals measured using a single pixel detector to enable ultra-fast, high-sensitivity, and accurate morphology-based cell analysis. The methods and systems can be image-free (not requiring image generation), but the methods and systems can incorporate image generation if desired. The methods and systems may be further configured to sort particles in real time. Such methods and systems may be referred to as "ghost cytometry" (GC).

[0030]

[0100] Figure 1A shows a schematic diagram of an example of an optical compressive sensing process used in the GC process. Using the relative motion of an object across a patterned optical structure, the spatial information of the object can be compression-mapped into a time signal sequence. The patterned optical structure can be defined by a matrix H(x, y) that includes values of optical properties (such as transmittance, absorbance, reflectance, or indices thereof) in each region within the optical structure. As shown in Figure 1A, F1 and F2 can depict representative features (such as fluorescence features) within the object. Based on the motion of the object, the spatially varying optical properties of H(x, y) can be encoded as temporal modulations of the light intensity (such as fluorescence emission intensity) from fluorophores F1 and F2 (shown as "emission intensity from F1" and "emission intensity from F2" respectively in Figure 1A). Their sum g(t) can be recorded using a single-pixel detector as shown at the bottom of Figure 1A. In imaging mode, a two-dimensional (2D) image of the object can be computationally reconstructed by the combined use of the multiplexed time waveform g(t) and the intensity distribution H(x, y) of the optical structure. In image-free mode, by directly applying machine learning methods to the compressed time waveform, high-throughput, high-precision, image-free form-based cell classification can be achieved. The schematic diagram shown in Figure 1A is not to scale.

[0031]

[0101] In GC, when an object passes through a patterned optical structure, each arranged spot within the structure can be a sequentially different location of the object. For example, each spot arranged within the structure can sequentially excite fluorophores at different positions of the object. The encoded intensity from each position can be multiplexed and measured in a compressed and continuous manner as a single time waveform measured using a detector (such as a single-pixel detector like a photomultiplier tube (PMT)). Assuming the object is moving in a constant one-direction at a speed v, the signal acquisition can be mathematically described as follows.

[0032]

Number

[0033] Here, \(g(t)\) is the multiplexed time waveform, \(H(x,y)\) is the intensity distribution of the optical structure, and \(I(x,y)\) is the intensity distribution of the moving object. Note that \(H(x,y)\), which acts as the spatial encoding operator, can be static, and as a result, scanning or continuous light projection may not be required in GC. Binary random patterns for the optical structure are described herein as a simple implementation. In the measurement process of GC, the object may be convolved with the optical structure along the \(x\)-direction, and the resulting signal may be integrated along the \(y\)-direction. In the literature on compressive sensing, the randomized convolution can be regarded as an imaging modality (12). When Equation (1) is given as the forward model, the image reconstruction process will solve an inverse problem. The solution can be iteratively estimated by minimizing an objective function that can be calculated by the combined use of the multiplexed time waveform \(g(t)\) and the intensity distribution \(H(x,y)\) of the optical structure. For sparse events in the regularization region, the moving object can be reasonably estimated from the measured signal \(g(t)\) by adopting a compressive sensing algorithm such as two-stage iterative shrinkage / thresholding (TwIST) (13). Such a reconstruction process can share the concept with ghost imaging, in which a number of random optical patterns are sequentially projected onto an object, and after recording the resulting signal using a single-pixel detector, the original image is computationally restored (14 - 19). Ghost imaging has attracted considerable attention in the scientific community, but the sequential projection of optical patterns has slowed down ghost imaging and hindered its practical use. Even when compressive sensing is used to shorten the time required for light projection, this method is still slower than a conventional array pixel camera (18). In contrast, GC does not require any movement of the device, and the speed of image acquisition can increase with the movement of the object. The speed of image acquisition in GC can be limited only by the bandwidth of the single-pixel detector. Such bandwidth can be very high, such as at least about 1 megahertz (MHz), 10 MHz, 100 MHz, 1 gigahertz (GHz), or more.Thus, the use of motion in GC can convert low-speed ghost imaging into a practical ultra-high-speed continuous imaging procedure. For example, GC can achieve an acquisition speed that is at least 10,000 times faster than conventional fluorescence ghost imaging methods.

[0034]

[0102] In one aspect, the present disclosure provides a method for particle analysis. The method may include obtaining spatial information from the motion of particles relative to a patterned optical structure. Next, the spatial information may be compression-transformed into signals that sequentially reach a detector. Then, the signals can be used to analyze the particles.

[0035]

[0103] FIG. 1B shows a flowchart of a method 100 for particle analysis. In a first operation 110, the method 100 can include obtaining spatial information from the motion of particles relative to a patterned optical structure. The spatial information may include optical spatial information, which may be spatial information obtained by an optical sensing approach. Alternatively or in combination, the spatial information may comprise non-optical spatial information that can be obtained by a non-optical sensing approach such as impedance measurement.

[0036]

[0104] The particles may include one or more biological particles. The particles may include one or more cells.

[0037]

[0105] The particles may contain one or more rare cells. The rare cells may be present at a concentration of up to about 1 / 10, 1 / 20, 1 / 30, 1 / 40, 1 / 50, 1 / 60, 1 / 70, 1 / 80, 1 / 90, 1 / 100, 1 / 200, 1 / 300, 1 / 400, 1 / 500, 1 / 600, 1 / 700, 1 / 800, 1 / 900, 1 / 1,000, 1 / 2,000, 1 / 3,000, 1 / 4,000, 1 / 5,000, 1 / 6,000, 1 / 7,000, 1 / 8,000, 1 / 9,000, 1 / 10,000, 1 / 20,000, 1 / 30,000, 1 / 40,000, 1 / 50,000, 1 / 60,000, 1 / 70,000, 1 / 80,000, 1 / 90,000, 1 / 100,000, 1 / 200,000, 1 / 300,000, 1 / 400,000, 1 / 500,000, 1 / 600,000, 1 / 700,000, 1 / 800,000, 1 / 900,000, 1 / 1,000,000, 1 / 2,000,000, 1 / 3,000,000, 1 / 4,000,000, 1 / 5,000,000, 1 / 6,000,000, 1 / 7,000,000, 1 / 8,000,000, 1 / 9,000,000, 1 / 10,000,000, 1 / 20,000,000, 1 / 30,000,000, 1 / 40,000,000, 1 / 50,000,000, 1 / 60,000,000, 1 / 70,000,000, 1 / 80,000,000, 1 / 90,000,000, 1 / 100,000,000, 1 / 200,000,000, 1 / 300,000,000, 1 / 400,000,000, 1 / 500,000,000, 1 / 600,000,000, 1 / 700,000,000, 1 / 800,000,000, 1 / 900,000,000, 1 / 1,000,000,000, or more (compared to other cells under analysis).Rare cells may be present at a concentration of at least 1 in about 1,000,000,000, 1 in 900,000,000, 1 in 800,000,000, 1 in 700,000,000, 1 in 600,000,000, 1 in 500,000,000, 1 in 400,000,000, 1 in 300,000,000, 1 in 200,000,000, 1 in 100,000,000, 1 in 90,000,000, 1 in 80,000,000, 1 in 70,000,000, 1 in 60,000,000, 1 in 50,000,000, 1 in 40,000,000, 1 in 30,000,000, 1 in 20,000,000, 1 in 10,000,000, 1 in 9,000,000, 1 in 8,000,000, 1 in 7,000,000, 1 in 6,000,000, 1 in 5,000,000, 1 in 4,000,000, 1 in 3,000,000, 1 in 2,000,000, 1 in 1,000,000, 1 in 900,000, 1 in 800,000, 1 in 700,000, 1 in 600,000, 1 in 500,000, 1 in 400,000, 1 in 300,000, 1 in 200,000, 1 in 100,000, 1 in 90,000, 1 in 80,000, 1 in 70,000, 1 in 60,000, 1 in 50,000, 1 in 40,000, 1 in 30,000, 1 in 20,000, 1 in 10,000, 1 in 9,000, 1 in 8,000, 1 in 7,000, 1 in 6,000, 1 in 5,000, 1 in 4,000, 1 in 3,000, 1 in 2,000, 1 in 1,000, 1 in 900, 1 in 800, 1 in 700, 1 in 600, 1 in 500, 1 in 400, 1 in 300, 1 in 200, 1 in 100, 1 in 90, 1 in 80, 1 in 70, 1 in 60, 1 in 50, 1 in 40, 1 in 30, 1 in 20, 1 in 10, or more. Rare cells may be present at a concentration within a range defined by any two of the foregoing values.

[0038]

[0106] The particle(s) can include one or more cancer cells. The particle(s) can include one or more circulating tumor cells. The particle(s) can include any cell described herein. The cell(s) can be derived from any source described herein. The cell(s) can include any natural or non-natural component described herein.

[0039]

[0107] The patterned optical structure can include a plurality of regions. For example, the patterned optical structure can include at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 200, 300, 400, 500, 600, 700, 800, 900, 1,000, 2,000, 3,000, 4,000, 5,000, 6,000, 7,000, 8,000, 9,000, 10,000, 20,000, 30,000, 40,000, 50,000, 60,000, 70,000, 80,000, 90,000, 100,000, 200,000, 300,000, 400,000, 500,000, 600,000, 700,000, 800,000, 900,000, 1,000,000, or more regions. The patterned optical structure can include at most about 1,000,000, 900,000, 800,000, 700,000, 600,000, 500,000, 400,000, 300,000, 200,000, 100,000, 90,000, 80,000, 70,000, 60,000, 50,000, 40,000, 30,000, 20,000, 10,000, 9,000, 8,000, 7,000, 6,000, 5,000, 4,000, 3,000, 2,000, 1,000, 900, 800, 700, 600, 500, 400, 300, 200, 100, 90, 80, 70, 60, 50, 40, 30, 20, 10, 9, 8, 7, 6, 5, 4, 3, 2, or 1 region. The patterned optical structure can include some regions within a range defined by any two of the foregoing values.

[0040]

[0108] Each region of the plurality of regions can include one or more specific optical properties such as one or more transmittances, absorptances, or reflectances. For example, each region can have a transmittance, absorptance, or reflectance of at least about 0%, 1%, 2%, 3%, 4%, 5%, 60%, 7%, 8%, 9%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 91%, 92%, 93%, 94%, 95%, 96%, 97%, 98%, 99%, or 100%. Each region can have a transmittance, absorptance, or reflectance of up to about 100%, 99%, 98%, 97%, 96%, 95%, 94%, 93%, 92%, 91%, 90%, 80%, 70%, 60%, 50%, 40%, 30%, 20%, 10%, 9%, 8%, 7%, 6%, 5%, 4%, 3%, 2%, 1%, or 0%. Each region can have a transmittance, absorptance, or reflectance within a range defined by any two of the foregoing values.

[0041]

[0109] The optical properties can be different for each region. For example, any two regions can include the same or different transmittances, absorptances, or reflectances. The optical properties can be different for each region.

[0042]

[0110] The patterned optical structure can include a spatial light modulator (SLM), a digital micromirror device (DMD), a liquid crystal (LC) device, or a photomask.

[0043]

[0111] The patterned optical structure can be manufactured using microfabrication or nanofabrication techniques. For example, the patterned optical structure can be manufactured using solvent cleaning, piranha cleaning, RCA cleaning, ion implantation, ultraviolet photolithography, deep ultraviolet photolithography, extreme ultraviolet photolithography, electron beam lithography, nanoimprint lithography, wet chemical etching, dry chemical etching, plasma etching, reactive ion etching, deep reactive ion etching, electron beam milling, thermal annealing, thermal oxidation, thin film deposition, chemical vapor deposition, molecular organic chemical vapor deposition, low pressure chemical vapor deposition, plasma enhanced chemical vapor deposition, physical vapor deposition, sputtering, atomic layer deposition, molecular beam epitaxy, electrochemical vapor deposition, wafer bonding, wire bonding, flip chip bonding, thermosonic bonding, wafer dicing, or any one or more of other microfabrication or nanofabrication manufacturing techniques.

[0044]

[0112] The patterned optical structure may include a regularly patterned optical structure. The regularly patterned optical structure may include a periodic arrangement or other regular arrangement of regions of different optical properties. The patterned optical structure may include a disordered patterned optical structure. The disordered patterned optical structure can include a disordered arrangement of regions of different optical properties. The patterned optical structure may include an aperiodic patterned optical structure. The aperiodic patterned optical structure may include an aperiodic arrangement of regions of different optical properties. The patterned optical structure may include a randomly or pseudo-randomly patterned optical structure. The randomly or pseudo-randomly patterned optical structure can include a random or pseudo-random arrangement of regions of different optical properties.

[0045]

[0113] Disordered or aperiodic optical structures, such as an optical structure patterned randomly or pseudo-randomly, can enable the acquisition of optical spatial information with higher fidelity for an object under inspection (such as more accurate reconstruction of an image corresponding to the object), for example, by reducing attenuation across the frequency (or Fourier) domain. The signal measured using the systems and methods of the present disclosure can be regarded as a one-dimensional convolution of the patterned optical structure and the object under inspection. When the measurements are interpreted in the frequency domain, this can correspond to the multiplication of the Fourier transform of the patterned optical structure and the Fourier transform of the object under inspection, due to the convolution theorem. Thus, it can be beneficial to obtain the Fourier transform of a patterned optical structure that is close to a uniform distribution in order to measure the optical spatial information corresponding to the object under inspection without attenuation. The Fourier transform of a disordered or aperiodic function is a uniform distribution. Thus, disordered or aperiodic patterned optical structures, such as an optical structure patterned randomly or pseudo-randomly, can enable the acquisition of optical spatial information with higher fidelity for the object under inspection.

[0046]

[0114] In the second operation 120, method 100 can include compressing and converting spatial information into signals that sequentially reach a detector. The detector may comprise one or more single pixel detectors. The detector may comprise one or more multi-pixel detectors. The detector may include at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 200, 300, 400, 500, 600, 700, 800, 900, 1,000, 2,000, 3,000, 4,000, 5,000, 6,000, 7,000, 8,000, 9,000, 10,000, 20,000, 30,000, 40,000, 50,000, 60,000, 70,000, 80,000, 90,000, 100,000, 200,000, 300,000, 400,000, 500,000, 600,000, 700,000, 800,000, 900,000, 1,000,000, or more pixels. The detector may include at most about 1,000,000, 900,000, 800,000, 700,000, 600,000, 500,000, 400,000, 300,000, 200,000, 100,000, 90,000, 80,000, 70,000, 60,000, 50,000, 40,000, 30,000, 20,000, 10,000, 9,000, 8,000, 7,000, 6,000, 5,000, 4,000, 3,000, 2,000, 1,000, 900, 800, 700, 600, 500, 400, 300, 200, 100, 90, 80, 70, 60, 50, 40, 30, 20, 10, 9, 8, 7, 6, 5, 4, 3, 2, or 1 pixel. The detector may comprise some pixels within a range defined by any two of the foregoing values.

[0047]

[0115] The detector may comprise one or more photomultiplier tubes (PMTs), photodiodes, avalanche photodiodes, phototransistors, reverse-biased light-emitting diodes (LEDs), charge-coupled devices (CCDs), or complementary metal-oxide-semiconductor (CMOS) cameras.

[0048]

[0116] The detector can be configured to detect light. The light can include infrared (IR) light, visible light, or ultraviolet (UV) light. The light can include a plurality of wavelengths. The light can include one or more wavelengths of at least about 200 nm, 210 nm, 220 nm, 230 nm, 240 nm, 250 nm, 260 nm, 270 nm, 280 nm, 290 nm, 300 nm, 310 nm, 320 nm, 330 nm, 340 nm, 350 nm, 360 nm, 370 nm, 380 nm, 390 nm, 400 nm, 410 nm, 420 nm, 430 nm, 440 nm, 450 nm, 460 nm, 470 nm, 480 nm, 490 nm, 500 nm, 510 nm, 520 nm, 530 nm, 540 nm, 550 nm, 560 nm, 570 nm, 580 nm, 590 nm, 600 nm, 610 nm, 620 nm, 630 nm, 640 nm, 650 nm, 660 nm, 670 nm, 680 nm, 690 nm, 700 nm, 710 nm, 720 nm, 730 nm, 740 nm, 750 nm, 760 nm, 770 nm, 780 nm, 790 nm, 800 nm, 810 nm, 820 nm, 830 nm, 840 nm, 850 nm, 860 nm, 870 nm, 880 nm, 890 nm, 900 nm, 910 nm, 920 nm, 930 nm, 940 nm, 950 nm, 960 nm, 970 nm, 980 nm, 990 nm, 1,000 nm, or more.

[0049]

[0117] Light can include one or more wavelengths of at least about 1,000 nm, 990 nm, 980 nm, 970 nm, 960 nm, 950 nm, 940 nm, 930 nm, 920 nm, 910 nm, 900 nm, 890 nm, 880 nm, 870 nm, 860 nm, 850 nm, 840 nm, 830 nm, 820 nm, 810 nm, 800 nm, 790 nm, 780 nm, 770 nm, 760 nm, 750 nm, 740 nm, 730 nm, 720 nm, 710 nm, 700 nm, 690 nm, 680 nm, 670 nm, 660 nm, 650 nm, 640 nm, 630 nm, 620 nm, 610 nm, 600 nm, 590 nm, 580 nm, 570 nm, 560 nm, 550 nm, 540 nm, 530 nm, 520 nm, 510 nm, 500 nm, 490 nm, 480 nm, 470 nm, 460 nm, 450 nm, 440 nm, 430 nm, 420 nm, 410 nm, 400 nm, 390 nm, 380 nm, 370 nm, 360 nm, 350 nm, 340 nm, 330 nm, 320 nm, 310 nm, 300 nm, 290 nm, 280 nm, 270 nm, 260 nm, 250 nm, 240 nm, 230 nm, 220 nm, 210 nm, 200 nm, or less. The light may include one or more wavelengths within a range defined by any two of the foregoing values.

[0050]

[0118] Light can have a bandwidth of at least about 0.001 nm, 0.002 nm, 0.003 nm, 0.004 nm, 0.005 nm, 0.006 nm, 0.007 nm, 0.008 nm, 0.009 nm, 0.01 nm, 0.02 nm, 0.03 nm, 0.04 nm, 0.05 nm, 0.06 nm, 0.07 nm, 0.08 nm, 0.09 nm, 0.1 nm, 0.2 nm, 0.3 nm, 0.4 nm, 0.5 nm, 0.6 nm, 0.7 nm, 0.8 nm, 0.9 nm, 1 nm, 2 nm, 3 nm, 4 nm, 5 nm, 6 nm, 7 nm, 8 nm, 9 nm, 10 nm, 20 nm, 30 nm, 40 nm, 50 nm, 60 nm, 70 nm, 80 nm, 90 nm, 100 nm, or more. Light can have a bandwidth of at most about 100 nm, 90 nm, 80 nm, 70 nm, 60 nm, 50 nm, 40 nm, 30 nm, 20 nm, 10 nm, 9 nm, 8 nm, 7 nm, 6 nm, 5 nm, 4 nm, 3 nm, 2 nm, 1 nm, 0.9 nm, 0.8 nm, 0.7 nm, 0.6 nm, 0.5 nm, 0.4 nm, 0.3 nm, 0.2 nm, 0.1 nm, 0.09 nm, 0.08 nm, 0.07 nm, 0.06 nm, 0.05 nm, 0.04 nm, 0.03 nm, 0.02 nm, 0.01 nm, 0.009 nm, 0.008 nm, 0.007 nm, 0.006 nm, 0.005 nm, 0.004 nm, 0.003 nm, 0.002 nm, 0.001 nm, or less. Light can have a bandwidth within a range defined by any two of the aforementioned values.

[0051]

[0119] The detector may have a refresh rate of at least about 1 MHz, 2 mHz, 3 mHz, 4 mHz, 5 mHz, 6 mHz, 7 mHz, 8 mHz, 9 mHz, 10 mHz, 20 mHz, 30 mHz, 40 mHz, 50 mHz, 60 mHz, 70 mHz, 80 mHz, 90 mHz, 100 mHz, 200 mHz, 300 mHz, 400 mHz, 500 mHz, 600 mHz, 700 mHz, 800 mHz, 900 mHz, 1 GHz, 2 GHz, 3 GHz, 4 GHz, 5 GHz, 6 GHz, 7 GHz, 8 GHz, 9 GHz, 10 GHz, 20 GHz, 30 GHz, 40 GHz, 50 GHz, 60 GHz, 70 GHz, 80 GHz, 90 GHz, 100 GHz, 200 GHz, 300 GHz, 400 GHz, 500 GHz, 600 GHz, 700 GHz, 800 GHz, 900 GHz, 1,000 GHz, or more.

[0052]

[0120] The detector may have a refresh rate of at most about 1,000 GHz, 900 GHz, 800 GHz, 700 GHz, 600 GHz, 500 GHz, 400 GHz, 300 GHz, 200 GHz, 100 GHz, 90 GHz, 80 GHz, 70 GHz, 60 GHz, 50 GHz, 40 GHz, 30 GHz, 20 GHz, 10 GHz, 9 GHz, 8 GHz, 7 GHz, 6 GHz, 5 GHz, 4 GHz, 3 GHz, 2 GHz, 1 GHz, 900 mHz, 80 mHz, 700 mHz, 600 mHz, 500 mHz, 400 mHz, 300 mHz, 200 mHz, 100 mHz, 90 mHz, 80 mHz, 70 mHz, 60 mHz, 50 mHz, 40 mHz, 30 mHz, 20 mHz, 10 mHz, 9 mHz, 8 mHz, 7 mHz, 6 mHz, 5 mHz, 4 mHz, 3 mHz, 2 mHz, 1 MHz, or less. The detector may have a refresh rate within a range defined by any two of the foregoing values.

[0053]

[0121] The second operation 120 may include applying a two-stage iterative shrinkage / thresholding (TwIST) procedure.

[0054]

[0122] In a third operation 130, method 100 can include identifying at least a subset of particles using signals. The third operation 130 can include computationally reconstructing the morphology of the particles at least in part through use of a combination of one or more temporal waveforms including one or more light intensity distributions imparted by a patterned optical structure, as described herein (e.g., with respect to FIG. 1A).

[0055]

[0123] The first operation 110 can include (i) directing light from a light source through a patterned optical structure, (ii) directing light from the patterned optical structure to the particles, and (iii) directing light from the particles to a detector. In such a scenario, the patterned optical structure can impart an optical encoding to the light from the light source prior to the interaction of the light and the particles. Such a scenario is sometimes referred to as structured illumination (SI).

[0056]

[0124] Alternatively or in combination, the first operation 110 can include (i) directing light from a light source to the particles, (ii) directing light from the particles through a patterned optical structure, and (iii) directing light from the patterned optical structure to a detector. In such a scenario, the patterned optical structure can impart an optical encoding to the light from the light source after the interaction of the light and the particles. Such a scenario is sometimes referred to as structured detection (SD).

[0057]

[0125] The light source can include a laser. For example, the light source can include a continuous-wave laser. The light source can include a pulsed laser. The light source can include a gas laser such as a helium-neon (HeNe) laser, an argon (Ar) laser, a krypton (Kr) laser, a xenon (Xe) ion laser, a nitrogen (N2) laser, a carbon dioxide (CO2) laser, a carbon monoxide (CO) laser, a transverse excitation atmosphere (TEA) laser, or an excimer laser. For example, the light source can include an argon dimer (Ar2) excimer laser, a krypton dimer (Kr2) excimer laser, a fluorine dimer (F2) excimer laser, a xenon dimer (Xe2) excimer laser, an argon fluoride (ArF) excimer laser, a krypton chloride (KrCl) excimer laser, a krypton fluoride (KrF) excimer laser, a xenon bromide (XeBr) excimer laser, a xenon chloride (XeCl) excimer laser, or a xenon fluoride (XeF) excimer laser. The light source can include a dye laser.

[0058]

[0126] The light source can include a metal-vapor laser such as a helium-cadmium (HeCd) metal-vapor laser, a helium-mercury (HeHg) metal-vapor laser, a helium-selenium (HeSe) metal-vapor laser, a helium-silver (HeAg) metal-vapor laser, a strontium (Sr) metal-vapor laser, a neon-copper (NeCu) metal-vapor laser, a copper (Cu) metal-vapor laser, a gold (Au) metal-vapor laser, a manganese (Mn) metal-vapor, or a manganese chloride (MnCl2) metal-vapor laser.

[0059]

[0127] The light source can include solid lasers such as ruby lasers, metal-doped crystal lasers, or metal-doped fiber lasers. For example, the light source can be a neodymium-doped yttrium aluminum garnet (Nd:YAG) laser, a neodymium / chromium-doped yttrium aluminum garnet (Nd / Cr:YAG) laser, an erbium-doped yttrium aluminum garnet (Er:YAG) laser, a neodymium-doped yttrium lithium fluoride (Nd:YLF) laser, a neodymium-doped yttrium orthovanadate (Nd:YVO4) laser, a neodymium-doped yttrium calcium oxoborate (Nd:YCOB) laser, a neodymium glass (Nd:glass) laser, a titanium sapphire (Ti:sapphire) laser, a thulium-doped yttrium aluminum garnet (Tm:YAG) laser, a ytterbium-doped yttrium aluminum garnet (Yb:YAG) laser, a ytterbium-doped glass (Yb:glass) laser, a holmium yttrium aluminum garnet (Ho:YAG) laser, a chromium-doped zinc selenide (Cr:ZnSe) laser, a cerium-doped lithium strontium aluminum fluoride (Ce:LiSAF) laser, a cerium-doped lithium calcium aluminum fluoride (Ce:LiCAF) laser, an erbium-doped glass (Er:glass) laser, an erbium ytterbium co-doped glass (Er / Yb:glass) laser, a uranium-doped calcium fluoride (U:CaF2) laser, or a samarium-doped calcium fluoride (Sm:CaF2) laser.

[0060]

[0128] The light source can include semiconductor lasers or diode lasers such as gallium nitride (GaN) lasers, indium gallium nitride (InGaN) lasers, aluminum gallium indium phosphide (AlGaInP) lasers, aluminum gallium arsenide (AlGaAs) lasers, indium gallium arsenide phosphide (InGaAsP) lasers, vertical cavity surface emitting lasers (VCSELs), or quantum cascade lasers.

[0061]

[0129] The light source can be configured to emit light. The light can include infrared (IR) light, visible light, or ultraviolet (UV) light. The light can include a plurality of wavelengths. The light can include one or more wavelengths of at least about 200 nm, 210 nm, 220 nm, 230 nm, 240 nm, 250 nm, 260 nm, 270 nm, 280 nm, 290 nm, 300 nm, 310 nm, 320 nm, 330 nm, 340 nm, 350 nm, 360 nm, 370 nm, 380 nm, 390 nm, 400 nm, 410 nm, 420 nm, 430 nm, 440 nm, 450 nm, 460 nm, 470 nm, 480 nm, 490 nm, 500 nm, 510 nm, 520 nm, 530 nm, 540 nm, 550 nm, 560 nm, 570 nm, 580 nm, 590 nm, 600 nm, 610 nm, 620 nm, 630 nm, 640 nm, 650 nm, 660 nm, 670 nm, 680 nm, 690 nm, 700 nm, 710 nm, 720 nm, 730 nm, 740 nm, 750 nm, 760 nm, 770 nm, 780 nm, 790 nm, 800 nm, 810 nm, 820 nm, 830 nm, 840 nm, 850 nm, 860 nm, 870 nm, 880 nm, 890 nm, 900 nm, 910 nm, 920 nm, 930 nm, 940 nm, 950 nm, 960 nm, 970 nm, 980 nm, 990 nm, 1,000 nm, or more.

[0062]

[0130] Light can include one or more wavelengths of at least about 1,000 nm, 990 nm, 980 nm, 970 nm, 960 nm, 950 nm, 940 nm, 930 nm, 920 nm, 910 nm, 900 nm, 890 nm, 880 nm, 870 nm, 860 nm, 850 nm, 840 nm, 830 nm, 820 nm, 810 nm, 800 nm, 790 nm, 780 nm, 770 nm, 760 nm, 750 nm, 740 nm, 730 nm, 720 nm, 710 nm, 700 nm, 690 nm, 680 nm, 670 nm, 660 nm, 650 nm, 640 nm, 630 nm, 620 nm, 610 nm, 600 nm, 590 nm, 580 nm, 570 nm, 560 nm, 550 nm, 540 nm, 530 nm, 520 nm, 510 nm, 500 nm, 490 nm, 480 nm, 470 nm, 460 nm, 450 nm, 440 nm, 430 nm, 420 nm, 410 nm, 400 nm, 390 nm, 380 nm, 370 nm, 360 nm, 350 nm, 340 nm, 330 nm, 320 nm, 310 nm, 300 nm, 290 nm, 280 nm, 270 nm, 260 nm, 250 nm, 240 nm, 230 nm, 220 nm, 210 nm, 200 nm, or less. The light may include one or more wavelengths within a range defined by any two of the aforementioned values.

[0063]

[0131] Light can have a bandwidth of at least about 0.001 nm, 0.002 nm, 0.003 nm, 0.004 nm, 0.005 nm, 0.006 nm, 0.007 nm, 0.008 nm, 0.009 nm, 0.01 nm, 0.02 nm, 0.03 nm, 0.04 nm, 0.05 nm, 0.06 nm, 0.07 nm, 0.08 nm, 0.09 nm, 0.1 nm, 0.2 nm, 0.3 nm, 0.4 nm, 0.5 nm, 0.6 nm, 0.7 nm, 0.8 nm, 0.9 nm, 1 nm, 2 nm, 3 nm, 4 nm, 5 nm, 6 nm, 7 nm, 8 nm, 9 nm, 10 nm, 20 nm, 30 nm, 40 nm, 50 nm, 60 nm, 70 nm, 80 nm, 90 nm, 100 nm, or more. Light can have a bandwidth of at most about 100 nm, 90 nm, 80 nm, 70 nm, 60 nm, 50 nm, 40 nm, 30 nm, 20 nm, 10 nm, 9 nm, 8 nm, 7 nm, 6 nm, 5 nm, 4 nm, 3 nm, 2 nm, 1 nm, 0.9 nm, 0.8 nm, 0.7 nm, 0.6 nm, 0.5 nm, 0.4 nm, 0.3 nm, 0.2 nm, 0.1 nm, 0.09 nm, 0.08 nm, 0.07 nm, 0.06 nm, 0.05 nm, 0.04 nm, 0.03 nm, 0.02 nm, 0.01 nm, 0.009 nm, 0.008 nm, 0.007 nm, 0.006 nm, 0.005 nm, 0.004 nm, 0.003 nm, 0.002 nm, 0.001 nm, or less. Light can have a bandwidth within a range defined by any two of the foregoing values.

[0064]

[0132] The light may have an average optical power of at least about 1 microwatt (μW), 2 μW, 3 μW, 4 μW, 5 μW, 6 μW, 7 μW, 8 μW, 9 μW, 10 μW, 20 μW, 30 μW, 40 μW, 50 μW, 60 μW, 70 μW, 80 μW, 90 μW, 100 μW, 200 μW, 300 μW, 400 μW, 500 μW, 600 μW, 700 μW, 800 μW, 900 μW, 1 milliwatt (mW), 2 mW, 3 mW, 4 mW, 5 mW, 6 mW, 7 mW, 8 mW, 9 mW, 10 mW, 20 mW, 30 mW, 40 mW, 50 mW, 60 mW, 70 mW, 80 mW, 90 mW, 100 mW, 200 mW, 300 mW, 400 mW, 500 mW, 600 mW, 700 mW, 800 mW, 900 mW, 1 watt (W), 2 W, 3 W, 4 W, 5 W, mW, mW, mW, mW, mW, mW, mW, mW, 1 watt (W), 2 W, 3 W, 4 W, 5 W, 6 W, 7 W, 8 W, 9 W, 10 W, or more.

[0065]

[0133] The light may have an average optical power of at most about 10 W, 9 W, 8 W, 7 W, 6 W, 5 W, 4 W, 3 W, 2 W, 1 W, 900 mW, 800 mW, 700 mW, 600 mW, 500 mW, 400 mW, 300 mW, 200 mW, 100 mW, 90 mW, 80 mW, 70 mW, 60 mW, 50 mW, 40 mW, 30 mW, 20 mW, 10 mW, 9 mW, 8 mW, 7 mW, 6 mW, 5 mW, 4 mW, 3 mW, 2 mW, 1 mW, 900 μW, 800 μW, 700 μW, 600 μW, 500 μW, 400 μW, 300 μW, 200 μW, 100 μW, 90 μW, 80 μW, 70 μW, 60 μW, 50 μW, 40 μW, 30 μW, 20 μW, 10 μW, 9 μW, 8 μW, 7 μW, 6 μW, 5 μW, 4 μW, 3 μW, 2 μW, 1 μW, or less. The light may have an optical power within a range defined by any two of the foregoing values.

[0066]

[0134] The light can have a peak optical power of at least about 1 mW, 2 mW, 3 mW, 4 mW, 5 mW, 6 mW, 7 mW, 8 mW, 9 mW, 10 mW, 20 mW, 30 mW, 40 mW, 50 mW, 60 mW, 70 mW, 80 mW, 90 mW, 100 mW, 200 mW, 300 mW, 400 mW, 500 mW, 600 mW, 700 mW, 800 mW, 900 mW, 1 W, 2 W, 3 W, 4 W, 5 W, 6 W, 7 W, 8 W, 9 W, 10 W, 10 W, 20 W, 30 W, 40 W, 50 W, 60 W, 70 W, 80 W, 90 W, 100 W, 200 W, 300 W, 400 W, 500 W, 600 W, 700 W, 800 W, 900 W, 1,000 W, or more.

[0067]

[0135] The light can have a peak optical power of at most about 1,000 W, 900 W, 800 W, 700 W, 600 W, 500 W, 400 W, 300 W, 200 W, 100 W, 90 W, 80 W, 70 W, 60 W, 50 W, 40 W, 30 W, 20 W, 10 W, 9 W, 8 W, 7 W, 6 W, 5 W, 4 W, 3 W, 2 W, 1 W, 900 mW, 800 mW, 700 mW, 600 mW, 500 mW, 400 mW, 300 mW, 200 mW, 100 mW, 90 mW, 80 mW, 70 mW, 60 mW, 50 mW, 40 mW, 30 mW, 20 mW, 10 mW, 9 mW, 8 mW, 7 mW, 6 mW, 5 mW, 4 mW, 3 mW, 2 mW, 1 mW, or less. The light can have a peak optical power within a range defined by any two of the aforementioned values.

[0068]

[0136] Light can have a pulse length of at least about 100 femtoseconds (fs), 200 fs, 300 fs, 400 fs, 500 fs, 600 fs, 700 fs, 800 fs, 900 fs, 1 picosecond (ps), 2 ps, 3 ps, 4 ps, 5 ps, 6 ps, 7 ps, 8 ps, 9 ps, 10 ps, 20 ps, 30 ps, 40 ps, 50 ps, 60 ps, 70 ps, 80 ps, 90 ps, 100 ps, 200 ps, 300 ps, 400 ps, 500 ps, 600 ps, 700 ps, 800 ps, 900 ps, 1 nanosecond (ns), 2 ns, 3 ns, 4 ns, 5 ns, 6 ns, 7 ns, 8 ns, 9 ns, 10 ns, 20 ns, 30 ns, 40 ns, 50 ns, 60 ns, 70 ns, 80 ns, 90 ns, 100 ns, 200 ns, 300 ns, 400 ns, 500 ns, 600 ns, 700 ns, 800 ns, 900 ns, 1 microsecond (μs), 2 μs, 3 μs, 4 μs, 5 μs, 6 μs, 7 μs, 8 μs, 9 μs, 10 μs, 20 μs, 30 μs, 40 μs, 50 μs, 60 μs, 70 μs, 80 μs, 90 μs, 100 μs, 200 μs, 300 μs, 400 μs, 500 μs, 600 μs, 700 μs, 800 μs, 900 μs, 1 millisecond (1 ms), or more.

[0069]

[0137] Light can have a pulse length of up to about 1 ms, 900 μs, 800 μs, 700 μs, 600 μs, 500 μs, 400 μs, 300 μs, 200 μs, 100 μs, 90 μs, 80 μs, 70 μs, 60 μs, 50 μs, 40 μs, 30 μs, 20 μs, 10 μs, 9 μs, 8 μs, 7 μs, 6 μs, 5 μs, 4 μs, 3 μs, 2 μs, 1 μs, 900 ns, 800 ns, 700 ns, 600 ns, 500 ns, 400 ns, 300 ns, 200 ns, 100 ns, 90 ns, 80 ns, 70 ns, 60 ns, 50 ns, 40 ns, 30 ns, 20 ns, 10 ns, 9 ns, 8 ns, 7 ns, 6 ns, 5 ns, 4 ns, 3 ns, 2 ns, 1 ns, 900 ps, 800 ps, 700 ps, 600 ps, 500 ps, 400 ps, 300 ps, 200 ps, 100 ps, 90 ps, 80 ps, 70 ps, 60 ps, 50 ps, 40 ps, 30 ps, 20 ps, 10 ps, 9 ps, 8 ps, 7 ps, 6 ps, 5 ps, 4 ps, 3 ps, 2 ps, 1 ps, 900 fs, 800 fs, 700 fs, 600 fs, 500 fs, 400 fs, 300 fs, 200 fs, 100 fs, or less. Light can have a pulse length within a range defined by any two of the aforementioned values.

[0070]

[0138] Method 100 can further include applying one or more machine learning classifiers to one or more compressed waveforms corresponding to the signal. Method 100 can further include applying one or more machine learning classifiers to one or more of the time waveforms described herein or to compressed waveforms related to one or more of the time waveforms described herein.

[0071]

[0139] One or more machine learning classifiers can be applied to the waveform to identify one or more of the particles described herein. The one or more machine learning classifiers can be configured to identify one or more particles and obtain at least one, two, or three of sensitivity, specificity, and accuracy of at least about 60%, 61%, 62%, 63%, 64%, 65%, 66%, 67%, 68%, 69%, 70%, 71%, 72%, 73%, 74%, 75%, 76%, 77%, 78%, 79%, 80%, 81%, 82%, 83%, 84%, 85%, 86%, 87%, 88%, 89%, 90%, 91%, 92%, 93%, 94%, 95%, 96%, 97%, 98%, 99%, 99.1%, 99.2%, 99.3%, 99.4%, 99.5%, 99.6%, 99.7%, 99.8%, 99.9%, 99.91%, 99.92%, 99.93%, 99.94%, 99.95%, 99.96%, 99.97%, 99.98%, 99.99%, or more.

[0072]

[0140] The one or more machine learning classifiers can be configured to identify one or more particles and obtain at least one, two, or three of sensitivity, specificity, and accuracy of up to about 99.99%, 99.98%, 99.97%, 99.96%, 99.95%, 99.94%, 99.93%, 99.92%, 99.91%, 99.9%, 99.8%, 99.7%, 99.6%, 99.5%, 99.4%, 99.3%, 99.2%, 99.1%, 99%, 98%, 97%, 96%, 95%, 94%, 93%, 92%, 91%, 90%, 89%, 88%, 87%, 86%, 85%, 84%, 83%, 82%, 81%, 80%, 79%, 78%, 77%, 76%, 75%, 74%, 73%, 72%, 71%, 70%, 69%, 68%, 67%, 66%, 65%, 64%, 63%, 62%, 61%, 60%, or less. The machine learning classifier can be configured to identify one or more particles and obtain at least one, two, or three of sensitivity, specificity, and accuracy that are within the range defined by any two of the foregoing values.

[0073]

[0141] One or more machine learning classifiers can include any methods and techniques for using statistical techniques to infer one or more traits from one or more datasets. Such methods and techniques can include supervised, supervised, semi-supervised, or unsupervised machine learning techniques. Machine learning techniques can include regression analysis, regularization, classification, dimensionality reduction, ensemble learning, meta-learning, reinforcement learning, correlation rule learning, cluster analysis, anomaly detection, or deep learning. Machine learning techniques can include k-means, k-means clustering, k-nearest neighbors, learning vector quantization, linear regression, non-linear regression, least squares regression, partial least squares regression, logistic regression, stepwise regression, multivariate adaptive regression splines, ridge regression, principal component regression, least absolute shrinkage, selection operation, least angle regression, canonical correlation analysis, factor analysis, independent component analysis, linear discriminant analysis, multidimensional scaling, non-negative matrix factorization, principal component analysis, principal coordinates analysis, projection pursuit, Sammon mapping, t-distributed stochastic neighbor embedding, AdaBoost, boosting, bootstrap aggregation, ensemble averaging, decision trees, conditional decision trees, boosted decision trees, gradient boosted decision trees, random forests, stacking generalization, Bayesian networks, Bayesian belief networks, naive Bayes, Gaussian naive Bayes, polynomial naive Bayes, hidden Markov models, hierarchical hidden Markov models, support vector machines, encoders, decoders, autoencoders, stacked autoencoders, perceptrons, multi-layer perceptrons, artificial neural networks, feedforward neural networks, convolutional neural networks, recurrent neural networks, long short-term memory, deep belief networks, deep Boltzmann machines, deep convolutional neural networks, deep recurrent neural networks, or generative adversarial networks, but are not limited thereto.

[0074]

[0142] One or more machine learning classifiers can be trained to directly learn the identification of particles from time waveforms or compressed time waveforms. Thus, particles can be analyzed without image reconstruction.

[0075]

[0143] Alternatively or in combination, one or more machine learning classifiers can be trained to learn particle identification from one or more images reconstructed from a time waveform or a compressed time waveform. Thus, method 100 can further include the step of reconstructing one or more images of the particles, and the particles can be analyzed based on the reconstructed images. The one or more images may include one or more brightfield, cross-polarized light, darkfield, phase contrast, differential interference contrast (DIC), reflection interference, Sarfus, fluorescence, epi-fluorescence, confocal, light sheet, multi-photon, super-resolution, near-field scanning optical, near-field optical random mapping (NORM), structured illumination (SIM), spatial modulation illumination (SMI), 4-pi, stimulated emission depletion (STED), ground state depletion (GSD), reversible saturable optical linear fluorescence transition (RESOLFT), binding activation localization (BALM), photoactivation localization (PALM), stochastic optical reconstruction (STORM), direct stochastic optical reconstruction (dSTORM), super-resolution optical fluctuation imaging (SOFI), or offset localization microscopy (OLM) images.

[0076]

[0144] Method 100 can further include the step of reconstructing a plurality of images of the particles, such as at least about 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, or more images of the particles, up to about 100, 90, 80, 70, 60, 50, 40, 30, 20, 10, 9, 8, 7, 6, 5, 4, 3, 2, or 1 image of the particles, or some images of the particles within a range defined by any two of the foregoing values. Each of the plurality of images can include different wavelengths or wavelength ranges.

[0077]

[0145] One or more images may be free of blur artifacts. For example, one or more images may be free of blur artifacts when particles are moving at a speed of at least about 1 millimeter per second (mm / s), 2 mm / s, 3 mm / s, 4 mm / s, 5 mm / s, 6 mm / s, 7 mm / s, 8 mm / s, 9 mm / s, 10 mm / s, 20 mm / s, 30 mm / s, 40 mm / s, 50 mm / s, 60 mm / s, 70 mm / s, 80 mm / s, 90 mm / s, 100 mm / s, 200 mm / s, 300 mm / s, 400 mm / s, 500 mm / s, 600 mm / s, 700 mm / s, 800 mm / s, 900 mm / s, 1 meter per second (m / s), 2 m / s, 3 m / s, 4 m / s, 5 m / s, 6 m / s, 7 m / s, 8 m / s, 9 m / s, 10 m / s, 20 m / s, 30 m / s, 40 m / s, 50 m / s, 60 m / s, 70 m / s, 80 m / s, 90 m / s, 100 m / s, 200 m / s, 300 m / s, 400 m / s, 500 m / s, 600 m / s, 700 m / s, 800 m / s, 900 m / s, 1,000 m / s, or more with respect to a patterned optical structure.

[0078]

[0146] One or more images may be free of blur artifacts when the particles are moving at a speed of up to about 1,000 m / s, 900 m / s, 800 m / s, 700 m / s, 600 m / s, 500 m / s, 400 m / s, 300 m / s, 200 m / s, 100 m / s, 90 m / s, 80 m / s, 70 m / s, 60 m / s, 50 m / s, 40 m / s, 30 m / s, 20 m / s, 10 m / s, 9 m / s, 8 m / s, 7 m / s, 6 m / s, 5 m / s, 4 m / s, 3 m / s, 2 m / s, 1 m / s, 900 mm / s, 800 mm / s, 700 mm / s, 600 mm / s, 500 mm / s, 400 mm / s, 300 mm / s, 200 mm / s, 100 mm / s, 90 mm / s, 80 mm / s, 70 mm / s, 60 mm / s, 50 mm / s, 40 mm / s, 30 mm / s, 20 mm / s, 10 mm / s, 9 mm / s, 8 mm / s, 7 mm / s, 6 mm / s, 5 mm / s, 4 mm / s, 3 mm / s, 2 mm / s, 1 mm / s, or less with respect to the patterned optical structure. One or more images may be free of blurred articles when the particles are moving at a speed within a range defined by any two of the aforementioned values with respect to the patterned optical structure.

[0079]

[0147] Method 100 can further include the step of sorting at least a subset of the particles into one or more groups of sorted particles. For example, the method may further include sorting the particles into one or more groups of sorted particles based on the morphology of the particles. Particles (e.g., cells) can be identified and separated, isolated, or sorted by activating one or more actuators (e.g., piezoelectric units) to direct or redirect one or more particles, or a fluid stream containing one or more particles, for example, from one channel to another. Such one or more actuators may provide, for example, a fluid or pressure pulse (e.g., via an ultrasonic signal).

[0080]

[0148] Particles (e.g., cells) can be identified and separated, isolated, or sorted at a rate of at least about 1 particle / second, 2 particles / second, 3 particles / second, 4 particles / second, 5 particles / second, 6 particles / second, 7 particles / second, 8 particles / second, 9 particles / second, 10 particles / second, 20 particles / second, 30 particles / second, 40 particles / second, 50 particles / second, 60 particles / second, 70 particles / second, 80 particles / second, 90 particles / second, 100 particles / second, 200 particles / second, 300 particles / second, 400 particles / second, 500 particles / second, 600 particles / second, 700 particles / second, 800 particles / second, 900 particles / second, 1,000 particles / second, 2,000 particles / second, 3,000 particles / second, 4,000 particles / second, 5,000 particles / second, 6,000 particles / second, 7,000 particles / second, 8,000 particles / second, 9,000 particles / second, 10,000 particles / second, 20,000 particles / second, 30,000 particles / second, 40,000 particles / second, 50,000 particles / second, 60,000 particles / second, 70,000 particles / second, 80,000 particles / second, 90,000 particles / second, 100,000 particles / second, 200,000 particles / second, 300,000 particles / second, 400,000 particles / second, 500,000 particles / second, 600,000 particles / second, 700,000 particles / second, 800,000 particles / second, 900,000 particles / second, 1,000,000 particles / second, or more.

[0081]

[0149] Particles (e.g., cells) can be identified and separated, isolated, or sorted at a speed of up to about 1,000,000 particles per second, 900,000 particles per second, 800,000 particles per second, 700,000 particles per second, 600,000 particles per second, 500,000 particles per second, 400,000 particles per second, 300,000 particles per second, 200,000 particles per second, 100,000 particles per second, 90,000 particles per second, 80,000 particles per second, 70,000 particles per second, 60,000 particles per second, 50,000 particles per second, 40,000 particles per second, 30,000 particles per second, 20,000 particles per second, 10,000 particles per second, 9,000 particles per second, 8,000 particles per second, 7,000 particles per second, 6,000 particles per second, 5,000 particles per second, 4,000 particles per second, 3,000 particles per second, 2,000 particles per second, 1,000 particles per second, 900 particles per second, 800 particles per second, 700 particles per second, 600 particles per second, 500 particles per second, 400 particles per second, 300 particles per second, 200 particles per second, 100 particles per second, 90 particles per second, 80 particles per second, 70 particles per second, 60 particles per second, 50 particles per second, 40 particles per second, 30 particles per second, 20 particles per second, 10 particles per second, 9 particles per second, 8 particles per second, 7 particles per second, 6 particles per second, 5 particles per second, 4 particles per second, 3 particles per second, 2 particles per second, 1 particle per second, or less. Particles (e.g., cells) can be identified and separated, isolated, or sorted and can be achieved at a speed within a range defined by any two of the aforementioned values.

[0082]

[0150] Method 100 can further include the step of collecting one or more of the selected groups of particles to produce a concentrated particle mixture. One or more of the selected groups of particles can have a purity of at least about 60%, 61%, 62%, 63%, 64%, 65%, 66%, 67%, 68%, 69%, 70%, 71%, 72%, 73%, 74%, 75%, 76%, 77%, 78%, 79%, 80%, 81%, 82%, 83%, 84%, 85%, 86%, 87%, 88%, 89%, 90%, 91%, 92%, 93%, 94%, 95%, 96%, 97%, 98%, 99%, 99.1%, 99.2%, 99.3%, 99.4%, 99.5%, 99.6%, 99.7%, 99.8%, 99.9%, 99.91%, 99.92%, 99.93%, 99.94%, 99.95%, 99.96%, 99.97%, 99.98%, 99.99%, or more.

[0083]

[0151] One or more of the selected groups of particles can have a purity of up to about 99.99%, 99.98%, 99.97%, 99.96%, 99.95%, 99.94%, 99.93%, 99.92%, 99.91%, 99.9%, 99.8%, 99.7%, 99.6%, 99.5%, 99.4%, 99.3%, 99.2%, 99.2%, 99.1%, 99%, 98%, 97%, 96%, 95%, 94%, 93%, 92%, 91%, 90%, 89%, 88%, 87%, 86%, 85%, 84%, 83%, 82%, 81%, 80%, 79%, 78%, 77%, 76%, 75%, 74%, 73%, 72%, 71%, 70%, 69%, 68%, 67%, 66%, 65%, 64%, 63%, 62%, 61%, 60%, or less. One or more of the selected groups of particles can have a purity within a range defined by any two of the foregoing values.

[0084]

[0152] Method 100 can further include subjecting one or more particles from one or more groups of selected particles to one or more assays. The one or more assays can include one or more members selected from the group consisting of lysis, nucleic acid extraction, nucleic acid amplification, nucleic acid sequencing, and protein sequencing.

[0085]

[0153] Method 100 can further include subjecting the particles to hydrodynamic flow focusing. The particles can be subjected to hydrodynamic flow focusing before operation 110, during operation 110, after operation 110, before operation 120, during operation 120, after operation 120, before operation 130, during operation 130, and / or after operation 130.

[0086]

[0154] Method 100 can further include collecting a partial transmission speckle pattern of the particles when the particles move relative to a patterned optical structure. The speckle pattern can be generated by diffraction of light that has passed through the patterned optical structure. For example, coherent light can interact with the patterned optical structure to generate structured illumination, which can be directed to the focus of an objective lens. After focusing, the coherent light spreads and generates a speckle pattern by diffraction. At a significant length away from the focus, the speckle pattern can be large enough such that a first portion of the speckle pattern can be blocked while a second portion of the speckle pattern is transmitted. Such blocking can be implemented, for example, using an iris. This small portion of the original speckle pattern can be detected using any of the detectors described herein. When cell particles pass through the structured illumination, changes in the phase and intensity of the transmitted light can result in changes in the speckle pattern, which can cause changes in the intensity of the light detected by the detector. From the modulation of the detected signal, the phase and / or intensity information of the particles can be obtained.

[0087]

[0155] The systems and methods described herein may be used in various therapeutic applications.

[0088]

[0156] In some cases, method 100 can be used to identify one or more target cells from a plurality of cells. A method for identifying one or more target cells from a plurality of cells can include any one or more operations of method 100, such as any one or more of operations 110, 120, and 130 of method 100 described herein. For example, a method for identifying one or more target cells from a plurality of cells can include: (a) obtaining spatial information from the movement of a plurality of cells relative to a patterned optical structure (as described herein with respect to operation 110 of method 100, for example); and (b) using the spatial information to identify one or more target cells from the plurality of cells.

[0089]

[0157] The step of using spatial information to identify one or more target cells from a plurality of cells can include using the spatial information to analyze one or more target cells (as described herein with respect to operation 130 of method 100, for example). For example, the morphology of a particle can be at least partially computationally reconstructed by using a combination of one or more temporal waveforms including one or more light intensity distributions imparted by a patterned optical structure, as described herein (e.g., with respect to FIG. 1A).

[0090]

[0158] The step of using spatial information to identify one or more target cells from a plurality of cells can include applying one or more machine learning classifiers to identify target cells from the plurality of cells. The one or more machine learning classifiers can comprise any machine learning classifier described herein. The one or more machine learning classifiers can utilize spatial information (or a signal associated with the spatial information) to identify target cells. For example, the one or more machine learning classifiers can be applied to spatial information (or a signal related to the spatial information) as described herein with respect to method 100.

[0091]

[0159] One or more machine learning classifiers can be trained to learn the identification of a plurality of cells (or target cells) directly from spatial information (or from signals associated with the spatial information). Thus, a plurality of cells (or target cells) can be analyzed without image reconstruction.

[0092]

[0160] Alternatively, or in combination, one or more machine learning classifiers can be trained to learn the identification of a plurality of cells (or target cells) from one or more images reconstructed from spatial information (or from signals associated with the spatial information). Thus, the step of using the signal to identify one or more target cells can further include the step of reconstructing one or more images of a plurality of cells (or target cells). A plurality of cells (or target cells) can be analyzed based on the reconstructed images. The one or more images can comprise any of the images described herein (e.g., with respect to method 100).

[0093]

[0161] The spatial information can correspond to features, characteristics, or information regarding a plurality of cells. The spatial information can correspond one-to-one with features, characteristics, or information regarding a plurality of cells. The features, characteristics, or information regarding a plurality of cells can include one or more members selected from the group consisting of: metabolic state, proliferation state, differentiation state, maturation state, expression of marker proteins, expression of marker genes, cell morphology, organelle morphology, organelle location, organelle size or extent, cytoplasm morphology, cytoplasm location, cytoplasm size or extent, nucleus morphology, nucleus location, nucleus size or extent, mitochondria morphology, mitochondria arrangement, mitochondria size or extent, lysosome morphology, lysosome arrangement, lysosome size or extent, intracellular molecule distribution, intracellular peptide, polypeptide or protein distribution, intracellular nucleic acid distribution, intracellular carbohydrate or polysaccharide distribution, and intracellular lipid distribution.

[0094]

[0162] The step of using a signal to isolate one or more target cells may further include, as described herein (e.g., with respect to any of method 100 or FIGS. 5A, 5B, 5C, 12A, 12B, 12C, 12D, and 14), sorting a plurality of cells into one or more groups of sorted cells based on the morphology of the plurality of cells. The one or more groups of sorted cells may include the target cells.

[0095]

[0163] The step of using a signal to isolate one or more target cells may further include, as described herein (e.g., with respect to any of method 100 or FIGS. 5A, 5B, and 5C), collecting one or more groups of sorted cells to generate a concentrated cell mixture.

[0096]

[0164] Spatial information can be compression-converted into signals that sequentially reach a detector, as described herein with respect to method 100.

[0097]

[0165] One or more target cells can be isolated from a plurality of cells based on spatial information (or a signal related to spatial information). The step of isolating one or more target cells can include (i) analyzing a plurality of cells using spatial information (or a signal related to spatial information), (ii) sorting the plurality of cells into one or more groups of sorted cells based on the results of analyzing the plurality of cells, and (iii) collecting one or more target cells from the one or more groups of sorted cells.

[0098]

[0166] The step of using spatial information (or signals associated with spatial information) to isolate one or more target cells from a plurality of cells may comprise the step of using spatial information (or signals associated with spatial information) to analyze one or more target cells (as described herein, for example, with respect to operation 130 of method 100). For example, the morphology of the particles can be at least partially computationally reconstructed by using a combination of one or more temporal waveforms including one or more light intensity distributions imparted by a patterned optical structure (as described herein, for example, with respect to FIG. 1A).

[0099]

[0167] The step of using spatial information (or signals associated with spatial information) to isolate one or more target cells from a plurality of cells may include applying one or more machine learning classifiers to one or more temporal waveforms or compressed temporal waveforms corresponding to the spatial information (or signals associated with the spatial information), as described herein (e.g., with respect to method 100).

[0100]

[0168] The step of using spatial information (or signals associated with spatial information) to isolate one or more target cells may further include sorting the plurality of cells into one or more groups of selected cells based on the results of analyzing the plurality of cells. For example, the one or more groups of selected cells may be based on the morphology of the plurality of cells, as described herein (e.g., with respect to any of method 100 or FIGS. 5A, 5B, 5C, 12A, 12B, 12C, 12D, and 14). The one or more groups of selected cells may include the target cells.

[0101]

[0169] The step of using spatial information (or signals related to spatial information) to isolate one or more target cells can further include the step of collecting one or more target cells from one or more groups of sorted cells. For example, one or more groups of sorted cells can be collected to generate an enriched cell mixture as described herein (e.g., with respect to Method 100 or any of FIGS. 5A, 5B, and 5C).

[0102]

[0170] The target cells can include one or more therapeutic cells. The target cells are selected from the group consisting of stem cells, mesenchymal stem cells, induced pluripotent stem cells, embryonic stem cells, cells differentiated from induced pluripotent stem cells, cells differentiated from embryonic stem cells, genetically engineered cells, blood cells, red blood cells, white blood cells, T cells, B cells, natural killer cells, chimeric antigen receptor T cells, chimeric antigen receptor natural killer cells, cancer cells, and blast cells. It can be one or more members selected from the group consisting of

[0103]

[0171] In another aspect, the present disclosure provides an image-free method for classifying or sorting particles based at least in part on the morphology of the particles without using unnatural labels with at least 80% or greater accuracy.

[0104]

[0172] FIG. 1C shows a flowchart of an example of an image-free optical method 150 for classifying or sorting particles. In a first operation 160, method 150 can include the step of classifying or sorting particles based at least in part on the morphology of the particles without using unnatural labels. Operation 160 can comprise any one or more of the operations of method 100 described herein, such as any one or more of operations 110, 120, and 130 described herein.

[0105]

[0173] Many variations, modifications, and adaptations based on method 100 or 150 provided herein are possible. For example, the order of operations of method 100 or 150 may be changed as needed, some of the operations may be removed, some of the operations may be replicated, and additional operations may be added. Some of the operations may be performed sequentially. Some of the operations may be performed in parallel. Some of the operations may be performed once. Some of the operations may be performed two or more times. Some of the operations may include sub-operations. Some of the operations may be automated and some of the operations may be manual.

[0106]

[0174] In another aspect, the present disclosure provides a system for particle analysis. The system can include a fluid flow path configured to direct particles. The system can further include a detector in sensing communication with at least a portion of the fluid flow path and one or more computer processors operably coupled to the detector. The one or more computer processors can be individually or collectively programmed to perform method 100 or 150 described herein. For example, the one or more computer processors can be individually or collectively programmed to (i) obtain spatial information from the movement of particles relative to an optically patterned structure that is random or pseudo-randomly patterned, (ii) compressively transform the spatial information into signals that sequentially reach the detector, and (iii) analyze the particles using the signals. In some embodiments, the detector is a multi-pixel detector. Alternatively, the detector may be a single-pixel detector.

[0107] Computer system

[0175] The present disclosure provides a computer system programmed to implement the disclosed methods and systems. FIG. 13 shows a computer system 1301 including a central processing unit (CPU, also referred to herein as "processor" and "computer processor") 1305, which can be a single-core or multi-core processor, or multiple processors for parallel processing. The computer system 1301 also includes a memory or memory location 1310 (e.g., random access memory, read-only memory, flash memory), an electronic storage device 1315 (e.g., hard disk), a communication interface 1320 (e.g., network adapter) for communicating with one or more other systems, and peripheral devices 1325 such as a cache, other memories, data storage, and / or an electronic display adapter. The memory 1310, storage device 1315, interface 1320, and peripheral devices 1325 communicate with the CPU 1305 via a communication bus (solid lines), such as a motherboard. The storage device 1315 can be a data storage device (or data repository) for storing data. The computer system 1301 can be operably coupled to a computer network ("network") 1330 with the aid of the communication interface 1320. The network 1330 can be the Internet, the Internet and / or an extranet, or an intranet and / or extranet that communicates with the Internet. The network 1330 can be, in some cases, a telecommunications and / or data network. The network 1330 can include one or more computer servers that enable distributed computing, such as cloud computing. The network 1330 can, in some cases, implement a peer-to-peer network that can enable devices coupled to the computer system 1301 to operate as clients or servers, with the aid of the computer system 1301.

[0108]

[0176] CPU1305 can execute a sequence of machine-readable instructions that can be embodied in a program or software. The instructions can be stored at a memory location such as memory 1310. The instructions can be directed to CPU1305, and CPU1305 can then program or otherwise configure CPU1305 to implement the methods of the present disclosure. Examples of operations performed by CPU1305 can include fetch, decode, execute, and write-back.

[0109]

[0177] CPU1305 can be part of a circuit such as an integrated circuit. One or more other components of system 1301 can be included in the circuit. In some cases, the circuit is an application specific integrated circuit (ASIC).

[0110]

[0178] Storage device 1315 can store files such as drivers, libraries, and saved programs. Storage device 1315 can store user data, such as user preferences and user programs. Computer system 1301 can include one or more additional data storage devices external to computer system 1301, such as being located on a remote server that communicates with computer system 1301 via an intranet or the Internet in some cases.

[0111]

[0179] The computer system 1301 can communicate with one or more remote computer systems via the network 1330. For example, the computer system 1301 can communicate with a user's remote computer system. Examples of remote computer systems include personal computers (e.g., portable PCs), slates or tablet PCs (e.g., Apple® iPad®, Samsung® Galaxy Tab), telephones, smartphones (e.g., Apple® iPhone®, Android-enabled devices, Blackberry®), or personal digital assistants. A user can access the computer system 1301 via the network 1330.

[0112]

[0180] The methods described herein (e.g., one or more methods for particle analysis, image-free optical methods, or methods for identifying one or more target cells from a plurality of cells as described herein) can be implemented by machine (e.g., computer processor) executable code stored in an electronic memory location of the computer system 1301, such as the memory 1310 or the electronic storage device 1315. The machine executable code or machine readable code can be provided in the form of software. In use, the code can be executed by the processor 1305. In some cases, the code can be retrieved from the storage device 1315 and stored on the memory 1310 for easy access by the processor 1305. In some situations, the electronic storage device 1315 can be excluded and the machine executable instructions can be stored in the memory 1310.

[0113]

[0181] The code can be pre-compiled and configured for use with a machine having a processor adapted to execute the code, or can be compiled during runtime. The code can be supplied in a programming language that can be selected such that the code can be executed in a pre-compiled form or in a form that remains compiled.

[0114]

[0182] Aspects of the systems and methods provided herein, such as computer system 1301, may be embodied in programming. Various aspects of the technology may typically be considered a "product" or "article of manufacture" in the form of machine (or processor) executable code and / or associated data carried on or embodied in a type of machine-readable medium. The machine executable code can be stored in an electronic memory device such as a memory (e.g., read only memory, random access memory, flash memory) or a hard disk. A "storage" type of medium can include any or all of various semiconductor memories, tape drives, disk drives, etc., that can provide non-transitory storage for a computer, a tangible memory for a processor, or for software programming at any time. All or part of the software may sometimes be communicated via the Internet or various other electrical communication networks. Such communication can, for example, enable the loading of software from one computer or processor to another, such as from a management server or host computer to an application server computer platform. Accordingly, another type of medium that can carry software elements is one that is used over a physical interface between local devices through wired and optical fixed networks as well as via various air links and includes light, electrical, and electromagnetic waves. Physical elements that carry such waves, such as wired or wireless links, optical links, etc., can also be considered a medium that carries software. As used herein, unless limited to non-transitory, tangible "storage" media, terms such as computer or machine "readable media" refer to any medium involved in providing instructions to a processor for execution.

[0115]

[0183] Thus, machine-readable media such as computer-executable code can take many forms, including but not limited to tangible storage media, carrier wave media, or physical transmission media. Non-volatile storage media includes, for example, optical or magnetic disks, such as any storage device in any computer that can be used to implement a database shown in the drawings. Volatile storage media includes dynamic memory, such as the main memory of such a computer platform. Tangible transmission media includes coaxial cables, copper wire, and fiber optics, including wires that comprise a bus within a computer system. Carrier wave transmission media can take the form of electrical or electromagnetic signals, or acoustic or light waves, such as those generated during radio frequency (RF) data communications and infrared (IR) data communications. Thus, common forms of computer-readable media include, for example, floppy (registered trademark) disks, flexible disks, hard disks, magnetic tape, any other magnetic media, CD-ROM, DVD, or DVD-ROM, any other optical media, punch card paper tape, any other physical storage media with patterns of holes, RAM, ROM, PROM, and EPROM, FLASH (registered trademark)-EPROM, any other memory chip or cartridge, carrier waves that carry data or instructions, cables or links that carry such carrier waves, or any other media that a computer can read programming code and / or data from. Many of these forms of computer-readable media can be involved in carrying one or more sequences of one or more instructions to a processor for execution.

[0116]

[0184] Computer system 1301 can include, or be communicable with, for example, an electronic display 1335 that includes a user interface (UI) 1340 for providing information to a user. Examples of UIs include, but are not limited to, graphical user interfaces (GUIs) and web-based user interfaces.

[0117]

[0185] The methods and systems of the present disclosure can be implemented by one or more algorithms (e.g., one or more methods for particle analysis, image - free optical methods, or one or more algorithms corresponding to methods for identifying one or more target cells from a plurality of cells described herein). The algorithms can be implemented by software when executed by a central processing unit 1305.

[0118]

[0186] The methods and systems can be combined with, or modified by, other methods and systems such as, for example, WO2016136801 and WO2017073737, each of which is hereby incorporated by reference in its entirety.

[0119] Examples Example 1: Materials and Methods

[0187] All photomultiplier tubes used in this study were purchased from Hamamatsu Photonics Inc. In FIGS. 2, 3B, and 3C, the PMT signals were recorded using a 10 6 ohm resistor with an oscilloscope (8 - bit, 1 GHz, Tektronix). A PMT with a 10 MHz bandwidth incorporating an amplifier (H10723 - 210MOD2) was used for the detection of blue and green signals, and another PMT (7422 - 40) connected to a current amplifier (HCA - 40M - 100K - C, FEMTO) was used for the detection of red signals.

[0120]

[0188] In FIGS. 3D, 4, and 5, the PMT signals were recorded using an electronic filter with an FPGA development board (TR4, Terasic) having a digitizer (M4i.4451 or M2i.4932 - Expa, Spectrum, Germany) or an analog - to - digital converter (ADC, partially modified, Texas Instruments, AD11445EVM). The digitizer and / or FPGA continuously collected fixed - length signal segments simultaneously from each color channel using fixed trigger conditions applied to the green signal.

[0121]

[0189] In FIGS. 17A - E, 18A - F, and 19A - 19D, the GC signal was detected using a PMT with a bandwidth of 10 MHz having a built - in amplifier (H10723 - 210mOD2), while the fluorescence and side - scatter (SSC) signals were detected using a PMT with a bandwidth of 1 MHz (H10723 - 210mOD3) or 10 MHz (H10723 - 210mOD2). The forward - scatter (FSC) signal was acquired using a photodetector (PDA100A or PDA100A2, Thorlabs). The PMT signals were recorded with an electronic filter using a digitizer (M4i.4451 or M2i.4932 - Expa, Spectrum, Germany) or an FPGA development board (TR4, Terasic) having an analog - to - digital converter (partially modified, Texas Instrument, AD11445EVM). The digitizer and / or FPGA continuously collected fixed - length signal segments from each color channel simultaneously using fixed - trigger conditions applied to the green signal.

[0122]

[0190] The image of the object was reconstructed from the time signal measured using GC as defined in Equation (1) by solving the following optimization problem using the TwIST procedure (13).

[0123]

Equation

[0124] Here, g is the vector of the measured time signal, H(x, y) is the system matrix of the GC, i is the vector reformed from the input object image l, R is a regularization term having sparsity, and τ is a constant for regularization.

[0125]

Equation

[0126] is

[0127]

Equation

[0128] Represents a norm. In CS, the image object can be sparse in the regularization space and is incoherent with the randomized convolution measurement process shown in Equation (1) (32).

[0129]

[0191] When reconstructing the image in panel (i) of FIG. 3C, the wavelet transform was selected using the Fejer-Korovkin scaling filter as the regularization term (33). When reconstructing other images, total variation was selected as the regularization term (34). The two-dimensional total variation R TV is written as follows.

[0130]

Number

[0131] Total variation improves smoothness while preserving edges within the image of the object. TwIST encodes using either the wavelet transform or total variation.

[0132]

[0192] The reconstructed image I in panel (iv) of FIG. 3C was evaluated using the baseline image I' in panel (v) of FIG. 3C taken by an array pixel color camera, based on the peak signal-to-noise ratio (PSNR). The PSNR is calculated as follows.

[0133]

Number

[0134] Here, MAX(I') is the maximum pixel intensity in I', and MSE(I, I') is the mean squared error between I and I'. A larger PSNR means a better reconstruction, and its maximum value is infinity.

[0135]

[0193] All reagents were purchased from either Wako, Sigma-Aldrich, or Invitrogen, unless otherwise specified. For the phosphate-buffered saline (PBS) solution, D-PBS(-) (Wako) was used.

[0136]

[0194] Stem Fit AK02N (Ajinomoto) was used as an induced pluripotent stem cell (iPSC) culture medium together with all supplements, 10 μM of Y-27632 (Wako), and 0.25 μg / cm 2 of iMatrix-511 silk (Nippi). Stem Fit AK02N without Supplement C, containing 10 μM of SB431542 (Wako) and 10 μM of DMH (Wako), was used as a neural progenitor cell (NPC) differentiation medium. Stem Fit AK02N without Supplement C, containing 10 ng / mL of activin A (Wako) and 3 μM of CHIR99021 (Wako), was used as an endoderm differentiation medium. Stemsure DMEM (Wako) containing 20% Stemsure Serum Replacement (Wako), 1% L-alanyl L-glutamine solution (Nacalai Tesque), 1% monothioglycerol solution (Wako), 1% MEM non-essential amino acid solution (Nacalai Tesque), and 1% DMSO (Sigma-Aldrich) was used as a hepatoblast differentiation medium.

[0137]

[0195] mCF-7 and MIA PaCa-2 cells were purchased from the RIKEN BRC CELL BANK. Peripheral blood mononuclear cells (PBMCs) were purchased from Astarte Biologics, Inc. Jurkat cells, Y-79 retinoblastoma cells, and Raji cells were purchased from JCRB. T cells were purchased from ProMab. iPSCs were prepared as previously reported and can be purchased from the Coriell Institute (GM25256).

[0138]

[0196] mIA PaCa-2 cells were cultured in D-MEM (high glucose) containing sodium pyruvate (Wako) supplemented with L-glutamine, phenol red, and 10% FBS. Jurkat cells and Raji cells were cultured in RPMI-1640 containing L-glutamine and phenol red (Wako) supplemented with 10% and 20% FBS, respectively. Y-79 cells were cultured in RPMI 1640 medium (Gibco) containing 20% FBS.

[0139]

[0197] NPCs were differentiated from iPSCs by the following procedure. iPSCs were seeded at 2,500 cells / cm 2 in iPS culture medium (day 0). On day 1, the medium was replaced with NPC differentiation medium. On days 3 and 5, the medium was replaced with the same medium. On day 6, more than 90% of the cells were PAX6-positive NPCs.

[0140]

[0198] Hepatoblasts were differentiated from iPSCs by the following procedure. iPSCs were seeded at 2,500 cells / cm 2 in iPS culture medium (day 0). On day 1, the medium was replaced with endoderm differentiation medium. On day 3, the medium was replaced with the same medium. On day 5, the medium was replaced with hepatoblast differentiation medium, and the medium was replaced with the same medium on days 7 and 9. On day 10, the cells showed AFP positivity under immunostaining tests.

[0141]

[0199] mIA PaCa-2 cells were detached from the culture flask using trypsin. Jurkat cells were removed from the culture flask by pipetting. After washing with PBS solution, each type of cell was fixed with 4% paraformaldehyde in PBS solution (Wako) for 15 minutes. Then, the cells were washed with PBS solution. FITC mouse anti-human CD45 (BD Biosciences) was added to Jurkat cells and cultured for 30 minutes. The stained Jurkat cells were washed with PBS solution. Next, each type of cell was mixed and flowed into the sorting system.

[0142]

[0200] Either iPSCs, T cells, or Raji cells were used in each experiment. Before separation, the culture medium for iPSCs was exchanged with 0.5 mM EDTA in PBS solution and cultured for 3 minutes. Then, the medium was returned to the iPSC culture medium, and the cells were detached using a cell scraper. T cells and Raji cells were removed from the culture flask by pipetting. After washing with binding buffer, the cells were stained with PI (Institute of Medical Biology) and Annexin V-FITC (Institute of Medical Biology) for 15 minutes and then flowed into the system without washing. The cells were first gated using an FSC-SSC scatter plot to remove debris, and then gated using an FSC height vs width scatter plot to remove doublets. The gated cells were further gated and labeled as live cells, dead cells, or apoptotic cells according to the fluorescence intensity of PI and Annexin V-FITC. 1,000 cells randomly selected (without repetition) using the same number of undifferentiated cells and differentiated cells were used as training data and test data, respectively.

[0143]

[0201] iPSCs, NPCs, and hepatoblasts were each separated from the culture flask using a cell scraper. After washing with buffer (1% FBS, X% EDTA in D-PBS), they were stained with Calcein AM (Dojindo) and rBC2LCN-635 (Wako), and then washed with buffer. For the classification of undifferentiated cells (iPSCs) and differentiated cells (NPCs or hepatoblasts), each cell was mixed at an equal concentration. The mixed suspension was flowed into the iSGC system. The cells were first gated using an FSC-SSC scatter plot to remove dead cells and debris, then gated using an FSC height vs width scatter plot to remove doublets, and again gated using an FSC-Calcein AM scatter plot to remove the remaining dead cells. The gated cells were further gated and labeled as undifferentiated cells or differentiated cells according to the fluorescence intensity of rBC2LCN-635 and Calcein AM. 5,000 and 1,000 cells randomly selected (without repetition) using the same number of undifferentiated cells and differentiated cells were used as training data and test data, respectively.

[0144]

[0202] The cells were flowed through either a quartz flow cell (Hamamatsu) or a PDMS microfluidic device using a customized pressure pump (IsoFlow, Beckman Coulter) for the sheath fluid and a syringe pump (KD Scientific) for the sample fluid. The quartz flow cell had channel cross-sectional dimensions of 250 μm × 500 μm at the measurement position, and when using this, the sheath flow was driven at a pressure of approximately 25 kPa unless otherwise specified. The PDMS device had channel cross-sectional dimensions of 50 × 50 μm at the measurement position, and when using this, the sheath flow was driven at a pressure of approximately 170 kPa unless otherwise specified. The sample fluid was driven at a flow rate of 20 μL / min unless otherwise specified.

[0145]

[0203] Before cell staining, the cell count was adjusted to approximately 8 × 10 6 cells per test using a cell counting chip. After staining, the cell concentration was adjusted to approximately 1 × 10 6A cell suspension of cells / mL was prepared. When MCF-7 cells were stained with multiple colors (as shown in FIGS. 3B to 5), their membranes were first labeled with PE-CF594-EpCAM (BD, PE-CF594 mouse anti-human CD326 antibody) by adding 100 μL of 0.5% BSA in PBS and then 5 μL of the PE-CF594-EpCAM antibody solution, followed by culturing at room temperature for 15 minutes. After fixing the cells with formaldehyde, 2 mL of a 3-60 μM DAPI stock solution in PBS was added to the cell pellet, and their nuclei were labeled with DAPI (Invitrogen, DAPI nucleic acid stain) by culturing them at room temperature for 15 minutes. After washing the cells thoroughly with PBS, the cytoplasm was finally labeled with FG (Thermo Fisher Scientific, LIVE / DEAD® Fixable Green Dead Cell Stain Kit, for 488 nm wavelength excitation) by adding 200 μL of 1% Tween 20 in PBS and then 5 μL of the FG fluorescent reactive dye solution to the cell pellet, and culturing for over 1 hour at room temperature. For cell imaging, the cell suspension was sandwiched between a pair of cover slides to fix the position of the cells, and the sample was placed on an electronic stage (Sigma Koki, Japan) to move the cells in the desired direction at a constant speed of 0.2 mm / s.

[0146]

[0204] In the image-free GC analysis experiments shown in FIGS. 4C to 4E, the cytoplasm of all MCF-7, MIA PaCa-2 cells and PBMC was labeled green using FG. The reaction time and dye concentration were adjusted so that the total fluorescence intensity was equal among different cell types. This confirmed that accurate classification was based on morphological information and not on total intensity. The membranes of MCF-7 cells were labeled for blue fluorescence using BV421-EpCAM (BV421-conjugated mouse anti-human CD326 antibody, BD) by adding 420 μL of 0.5% BSA in PBS and 80 μL of the BV421-EpCAM antibody solution to the cell pellet after fixation with formaldehyde and subsequent culturing at room temperature for 2 hours.

[0147]

[0205] In the label-free GC sorting experiment for classifying MCF-7 cells and MIA PaCa-2 cells (as shown in Fig. 5B), the membrane of MCF-7 cells was labeled for blue fluorescence using BV421-EpCAM in the same manner as above. In the experiment for classifying MIA PaCa-2 cells and PBMC (as shown in Fig. 5C), the cytoplasm of MIA PaCa-2 cells was labeled for blue fluorescence using primary mouse monoclonal anti-pan-cytokeratin AE1 / AE3 antibody (Abcam) and secondary AF405-conjugated donkey anti-mouse IgG antibody (Abcam). This staining was performed by first adding 450 μL of 0.5% Tween 20 and 0.5% BSA in PBS, and 50 μL of the primary antibody solution to the fixed cell pellet, followed by incubation overnight at 4 °C, then adding 490 μL of 0.5% Tween 20 and 0.5% BSA in PBS, and 10 μL of the secondary antibody solution, followed by incubation for 1 hour at room temperature. The cytoplasm of MCF-7, MIA PaCa-2 cells and PBMC was labeled green with FG after the above immunostaining in both experiments. The reaction time and dye concentration were adjusted so that the total fluorescence intensity was equal among different cell types. The cells were washed rigorously after each staining process.

[0148]

[0206] In FIG. 4, to train the model, a time waveform training dataset was collected by flowing each cell type (MCF-7 and MIA PaCa-2) separately through the optical structure using a syringe pump (Harvard) and a customized high-pressure pump. Building a model that is sufficiently generalized to avoid overfitting is the key to making reliable predictions for untrained signals. For this purpose, various time waveforms were collected by varying the flow rate of the inner fluid while maintaining the flow rate of the outer fluid: the inner flow rate was varied at 20 μL / min and 30 μL / min, and the outer flow rate was driven at a pressure of 180 kPa. Before testing the model, mixed solutions of MCF-7 and MIA PaCa-2 cells were prepared at different concentration ratios. When experimentally testing the model, the inner flow rate was maintained at 20 μL / min. Throughout the training and testing of the SVM model, the centerline of the cell flow stream was maintained at the same position. Finally, the same conditions were applied to the demonstration of cancer cell detection from composite PBMCs.

[0149]

[0207] All signals were detected using a PMT, collected and converted using an ADC board, and processed using Python. When preparing the dataset, the waveforms of single flowing objects were selectively collected by thresholding based on the average level of the entire recorded signal for training. The SVM with a radial basis function (RBF) used in this study estimates the label of the input time signal / waveform g by calculating the following scoring function f.

[0150]

Equation

[0151]

[0208] where, α j (0 ≤ α j ≤ C) and y j is the j-th trained time signal g jThe trained coefficients and class labels for, b is the bias term, γ is the kernel scale parameter, N is the number of trained time signals, and C is the regularization coefficient. C and γ were optimized by grid search. For high-throughput processing in the FPGA, N and the length of g were adjusted on an offline computer according to the biological sample and classification purpose.

[0152]

[0209] For cell classification, the support vector machine (SVM) technique was used with either a linear or a radial basis function kernel. The hyperparameters in the SVM were adjusted by grid search. Unless otherwise specified, training and validation were performed using an equal mixture of cell types for each class label, and accuracy, ROC curve, and ROC-AUC were verified with 10 random samplings and represented as the mean and standard deviation of 10 trials.

[0153] Example 2: Estimation of the sensitivity of ghost cytometry

[0210] For the entire system, the minimum number of detectable fluorophores can be estimated as follows under the assumption that the model fluorophore is a fluorescein molecule, the PMT can detect a single photon, the fluorophore is far from optical saturation, and the optical losses due to dichroic mirrors and bandpass filters can be ignored.

[0154]

[0211] The minimum number of detectable fluorophore molecules is given by the following.

[0155]

Equation

[0156]

[0212] The efficiency of photon collection in the experimental setup is given by the following.

[0157]

Equation

[0158]

[0213] Numerical aperture a = 0.75. The number of fluorescent photons emitted when a group of fluorophores passes through a unit spot defined by a random optical pattern is N s which is given by N f The number of fluorophores is given by N

[0159]

Equation

[0160]

[0214] The incident photon flux is given by the following equation

[0161]

Equation

[0162]

[0215] The incident photon flux on the bright spot of the structured illumination on the sample surface is given by f. The absorption cross-section is given by the following equation. s = 1.5×10 -20 [m 2 (35). The quantum efficiency is given by the following equation. q = 0.95(36). The exposure time is given by the following equation

[0163]

Equation

[0164]

[0216] The turnover rate (1 / lifetime) is given by the following equation. γ t = 1 / 4 [1 / nsec] (37). Therefore, in an ideal case, about three fluorophores (N f ~ 3) are required

[0165] Example 3: Motion-based compressed fluorescence imaging

[0217] Figures 2A - 2G show the demonstration of motion - based compressed fluorescence imaging. The optical setups for the SI mode and the SD mode are shown in FIGS. 2A and 2B, respectively. Aggregates of fluorescent beads were moved on a glass coverslip across the optical structure using an electronically translatable stage. As shown in FIG. 2(c), in the SI mode, the beads pass through the structured illumination in response to the motion, and a time waveform of the fluorescence intensity is generated. As shown in FIG. 2D, from this acquired time signal, a 2D fluorescence image is computationally reconstructed. As shown in FIG. 2E, in the SD mode, the beads are illuminated by uniform light. Then, the combined fluorescence image of the beads passes through a structured pinhole array, resulting in the generation of a time waveform. As shown in FIG. 2F, from this time signal, a 2D fluorescence image is computationally reconstructed. FIG. 2G shows the fluorescence image of the same aggregated beads obtained using an array pixel camera.

[0166]

[0218] As an experimental proof of concept for GC in imaging mode, fluorescent beads placed on a glass coverslip were imaged and moved using an electronic translation stage over an optically structured pattern that was either random or pseudo-randomly patterned. The beads were kept in focus as they moved in a direction parallel to the column direction of H(x,y). Incorporating the patterned optical structure either before or after the sample in the optical path is mathematically equivalent. This means that GC-based imaging can be experimentally realized by either randomly pseudo-randomly structured illumination (as shown in Fig. 2A) or random structured detection (as shown in Fig. 2B). These configurations respectively correspond to computerized ghost imaging and single pixel compressed imaging. The encoding operator H(x,y) in Equation (1) in SI mode can be experimentally measured as the excitation intensity distribution in the sample plane. On the other hand, the operator H(x,y) in SD mode can be measured as the product with respect to the excitation intensity distribution in the sample plane and the transmittance distribution of the photomask at the coupling plane between the sample and the detector. The exact operator H(x,y) can be experimentally calibrated by placing a thin sheet of fluorescent polymer on the sample plane and measuring its intensity distribution using a spatially resolved multi-pixel detector. An incoherent blue light-emitting diode (LED) was used as the excitation source. During the movement of the object, the PMT collected photons emitted from the object as a temporally modulated fluorescence intensity (as shown in Figs. 2C and 2E respectively for SI mode and SD mode). Figs. 2D and 2F show the computationally restored fluorescence images of multiple beads for each waveform. For comparison, Fig. 2G shows an image obtained using an array detector-based scientific complementary metal oxide semiconductor (CMOS) camera. The morphological features of the beads are clear, confirming GC imaging in both SI mode and SD mode.

[0167] Example 4: Multicolor high-throughput fluorescence cell imaging

[0219] Figures 3A - 3D show the demonstration of multi - color and high - throughput fluorescence cell imaging by GC. Figure 3A shows the optical setup for multi - color motion - based compressed fluorescence imaging (SD mode). This setup utilized a coherent 473 nm wavelength blue laser and an incoherent 375 nm wavelength UV LED as excitation light sources coupled by a dichroic mirror (Figure 7). The experiment used cultured MCF - 7 cells having membranes, cytoplasm, and nuclei fluorescently stained red, green, and blue, respectively. As the labeled cells moved on an electronically translated stage, their combined images passed through an optical encoder to generate a time waveform. The signal was then split, using a dichroic mirror, into three color channels of (i) red, (ii) green, and (iii) blue and finally recorded by different PMTs, and a representative trace is shown in Figure 3B. As shown in Figure 3C, from the time signals for each PMT, the fluorescence images of the labeled cells were computationally restored in red (panel (i) of Figure 3C), green (panel (ii) of Figure 3C), and blue (panel (iii) of Figure 3C), respectively. Panel (iv) of Figure 3C shows a pseudo - color multi - color image combined from panels (i), (ii), and (iii) of Figure 3C. Panel (iv) of Figure 3C shows a multi - color fluorescence image acquired using an array pixel color camera. Figure 3D shows multi - color sub - millisecond fluorescence imaging of cells flowing at a throughput rate exceeding 10,000 cells / second. In the experiment, a 488 nm wavelength blue laser and a 405 nm wavelength violet laser passed through diffractive optical elements to generate random structured illumination for the cell flow. Panels (i) and (ii) of Figure 3D show the green and blue fluorescence signals from the cytoplasm and nucleus, respectively. From the time signals for each PMT, the fluorescence images of the labeled cells were computationally restored in green (panel (iii) of Figure 3D) and blue (panel (iv) of Figure 3D), respectively. Panel (v) of Figure 3D shows the reconstructed multi - color fluorescence image.

[0168]

[0220] The simple design of the GC optical system means that by adding multiple light sources, dielectric mirrors, and optical filters, as shown in Fig. 3A, multi-color fluorescence imaging using a single photomask becomes possible. To verify GC for cell imaging, MCF-7 cells (a human cell line derived from breast cancer) were stained in three colors - the membrane, nucleus, and cytoplasm were stained with red (EpCAM PE-CF594), blue (DAPI), and green (FG: fixable green) dyes respectively. The stained cells were placed on a cover glass slip. A coherent blue continuous wave (CW) laser and an incoherent ultraviolet (UV) LED light source were used to excite the fluorophores. The SD mode was used, and the operator H(x, y) was experimentally estimated for each excitation light source. In the experiment, the stage with the cover glass was moved, and the time signals from each color channel were measured using three PMTs respectively. The computationally reconstructed fluorescence images for each color clearly show the fine features of the cell membrane, cytoplasm, and nucleus. For comparison, panel (iv) of Fig. 3C shows the overlaid color image, and panel (v) of Fig. 3C shows the fluorescence image obtained using a conventional color camera (Wraycam SR130, WRAYMER Inc., Japan). The average peak signal-to-noise ratio of the red, green, and blue channels calculated according to Equation (8) is 26.2 dB between the reconstructed image of GC (shown in panel (iv) of Fig. 3C) and the camera image (shown in panel (v) of Fig. 3C). These results demonstrate the good performance of multi-color GC imaging in depicting the morphological features of cells.

[0169]

[0221] GC can also achieve high-speed multi-color continuous fluorescence imaging of flowing cells. Using a flow cell assembly (Hamamatsu Photonics, Japan) to focus the flow of fluorescent cells in three-dimensional space, the cells can be focused in a plane perpendicular to the flow aligned parallel to the length of the encoder H(x, y). Using a diffractive optical element that generates structured illumination within the flow cell, a continuous spread optical point scan of 100 pixels was performed perpendicular to the flow corresponding to the image size in the y direction in computer reconstruction. Panels (i) and (ii) of FIG. 3D show the time waveforms from each color channel of a single MCF-7 cell whose cytoplasm was labeled with FG and whose nucleus was labeled with DAPI. The fluorescence images were computationally reconstructed for each waveform in panels (iii) and (iv) of FIG. 3D, respectively. Panel (v) of FIG. 3D shows the computationally reconstructed multi-color fluorescence image, which has clearly resolved morphological features of the cells. Cells were flowed at a throughput higher than 10,000 cells / second, which is the speed at which array pixel cameras such as charge-coupled device (CCD) and CMOS generate completely motion-blurred images. The total input excitation intensity of the structured illumination after the objective lens was ~58 mW and ~14 mW for the 488 nm and 405 nm lasers, respectively, while that assigned to individual random spots was <43 μW and <10 μW on average. By designing and employing an appropriate DOE, the light pattern was created with minimal loss, suggesting high sensitivity of GC imaging, calculated as the minimum number of detectable fluorophores (as detailed in the method) close to the single molecule level.

[0170]

[0222] Using the motion of the object across the light pattern H(x, y) for optical encoding and sparse sampling, motion-free high frame rate imaging was achieved with a high signal-to-noise ratio. The frame rate r and pixel scan speed p are defined as follows.

[0171]

Number

[0172]

Number

[0173] Here, I is the final image, and p is the reciprocal of the time required for the fluorophore to pass through each excitation spot in H(x, y). First, compression encoding is defined by the length of H(x, y), reduces the number of sampling points required for one frame acquisition, and as a result, effectively reduces the required bandwidth to achieve a high frame rate. Since shot noise increases as the bandwidth increases, this reduction is particularly important in ultrafast imaging using a small number of photons. Due to this feature, the excitation power of the fluorescence signal can be effectively reduced to overcome noise. Second, at a sufficient signal-to-noise ratio, GC can fully utilize the high bandwidth of a single pixel detector as H(x, y), which can be temporally static, unlike other techniques that modulate the excitation intensity temporally before the object passes through the pixels of the excitation light. Therefore, GC provides a non-blurred image as long as the pixel scanning speed does not exceed at least twice the bandwidth speed of the PMT. For example, for H(x, y) with a spot size of 500 nm and a PMT with a high bandwidth of 100 MHz, GC provides a non-blurred image for flow rates up to 100 m / s.

[0174] Example 5: Machine Learning Classification of Fluorescently Labeled Cells

[0223] Figures 4A-4E show high-throughput and high-precision fluorescence image-free "imaging" cytometry by GC through direct machine learning of compression signals. Figure 4A shows the procedure for training a classifier model in GC. Panel (i) of Figure 4A shows that different but morphologically similar cell types (MCF-7 and MIA PaCa-2 cells) were fluorescently labeled. For both cell types, the cytoplasm was stained green by FG, and the membrane of only MCF-7 cells was stained blue by BV421-EpCAM. Scale bar = 20 μm. Panel (ii) of Figure 4A shows flowing different cell types separately through the encoded optical structure used in Figure 3D at a throughput rate of >10,000 cells / second. Panel (iii) of Figure 4A shows the compression waveforms of each cell type collectively extracted from the time-modulated signals of fluorescence intensity. Panel (iv) of Figure 4A shows a library of waveforms labeled with each cell type used as a training dataset for constructing a cell classifier. A support vector machine model was used in this study. Figure 4B shows the procedure for testing the classifier model. Panel (i) of Figure 4B shows that different types of cells were experimentally mixed at various concentration ratios before analysis. Panel (ii) of Figure 4B shows the flow of the cell mixture through the encoder at the same throughput rate. Panel (iii) of Figure 4B shows applying the trained model directly to the waveforms to classify cell types. As shown in Panel (i) of Figure 4C, the blue plots are the concentration ratios of MCF-7 cells in each sample estimated by directly applying an SVM-based classification trained on the waveform of FG intensity (shown in detail in Panel (iv) of Figure 4C) compared to that obtained by measuring the total intensity of BV421 (shown in detail in Panel (ii) of Figure 4C). The red plots are the concentration ratios of MCF-7 cells estimated by applying the same procedure of SVM-based classification to the total FG intensity: applying the trained SVM-based classification (shown in detail in Panel (iii) of Figure 4C) to the total FG intensity compared to the measured total intensity of BV421. Seventy samples were measured for various concentration ratios, and each sample included 700 tests of randomly mixed cells.The results of image-free GC (shown as blue plots in panel (i) of Fig. 4C) reveal a small root mean square error (RMSE) of 0.046 from y = x and an area under the receiver operating characteristic (ROC, shown in Fig. 4D) curve (AUC) of 0.971 over ~5000 cells, and the morphology of these cells appears to be similar to that of the human eye. Each point on the ROC curve corresponds to a threshold applied to the scores from the trained SVM model, and the red and blue in the histogram are labeled according to the intensity of BV421. In contrast, the red plot in panel (i) of Fig. 4C shows a poor classification result with a large RMSD of 0.289 and a poor ROC-AUC of 0.596. As shown in Fig. 4E, when classifying model cancer (MCF-7) cells against a complex mixture of peripheral blood mononuclear cells (PBMCs), ultra-fast image-free GC recorded a high value of AUC ~0.998, confirming its robust and accurate performance in practical use.

[0175]

[0224] In addition to the use of GC as a powerful imaging device, the direct analysis of the compressively generated signals from GC enables high-throughput and accurate classification of cell morphology at significantly lower computer costs compared to other imaging methods, leading to the realization of ultra-fast fluorescence “imaging” activated cell sorting and selection (FiCS). This is achieved because compressive sensing in GC can significantly reduce the size of the imaging data while retaining sufficient information to reconstruct the object image. Human perception cannot directly classify waveforms, but machine learning methods can analyze waveforms without image generation. Supervised machine learning applied directly to waveforms measured at a rate of ~10,000 cells / second classified fluorescently labeled cells with high performance exceeding that of existing flow cytometers and human image perception. Furthermore, this image-free GC can be integrated with a microfluidic sorting system to enable real-time FiCS.

[0176]

[0225] Imageless GC can include two steps: 1) training a model for cell classification and 2) testing the model. The model was constructed based on support vector machine (SVM) technology (23) by computationally mixing the fluorescence signal waveforms from different cell types. This training data of the waveforms was collected by experimentally passing each cell type separately through an optical encoder. Then, the experimentally mixed cells were flowed through, and the model was tested by classifying the cell types. Before the experiment, two different types of cells were cultured, fixed, and fluorescently labeled: MCF-7 cells and MIA PaCa-2 cells, as shown in panel (i) of Fig. 4A. For both cell types, the cytoplasm was labeled green with FG for classification by imageless GC, as shown in panel (i) of Fig. 4A. On the other hand, only in MCF-7 cells, the membrane was labeled blue with BV421-EpCAM. MIA PaCa-2 cells had only low autofluorescence in the blue channel, providing a contrast that was easily distinguishable in this channel, as shown in Figs. 10A - 10C. This was used to verify the classification of GC that relied on similar cytoplasmic labeling in both cell types. To excite these fluorophores, both violet and blue CW lasers were used, while a digitizer (M4i.4451, Spectrum, Germany) recorded the signals obtained from the PMT. A microfluidic system was developed to spatially control the position of the cell flow relative to a random optical structure corresponding to the position of the cells in the reconstructed image. Using this optical-fluidic platform, the waveforms of the green fluorescence intensity from the cytoplasm were collected for each cell type. From this training dataset, an SVM-based classifier without any feature extraction was constructed. To test this trained classifier, a series of solutions containing combinations of different cell types mixed at various concentration ratios were introduced. Then, each classifier identified ~700 waveforms of the mixed cells as a single dataset and estimated the concentration ratio for each. Since the combination of MCF-7 and MIA PaCa-2 was used, the classification results could be quantitatively scored by measuring the total fluorescence intensity of BV421 in the membrane of MCF-7 cells.

[0177]

[0226] The plot of the concentration ratio measured by applying the concentration ratio vs. model of MCF-7 and MIA PaCa-2 cells measured using the cyan fluorescence intensity to the green fluorescence waveform gives a line on the diagonal with a small root mean square error (RMSE) of 0.046 from y = x. Using the measured values of BV421 to evaluate the GC-based classification of ~49,000 mixed cells, an area under the receiver operating characteristic (ROC) curve (AUC) of 0.971 was obtained, confirming that cell classification by GC is accurate. Each point on the ROC curve corresponds to a threshold applied to the score obtained from the trained SVM model, and the red and blue in the histogram are labeled according to the intensity of BV421 in Fig. 4D. To confirm that the high performance of GC is due to the spatial information encoded in the waveform, the same procedure for SVM-based classification was applied to the total green fluorescence intensity obtained by integrating each GC waveform over time. The result seen as the red plot in Fig. 4C gives a poor ROC-AUC of 0.596 and a large RMSE of 0.289 from y = x, thus indicating little contribution of the total fluorescence intensity to the high performance of GC. In addition, by calculating a simple linear fit (as shown in Fig. 11), it was confirmed that SVM-based classification consistently maintains its accuracy over a wide range of concentration ratios. Therefore, label-free GC is an accurate cell classifier even when target cells are similar in size, total fluorescence intensity, and apparent morphological features and are present in a mixture at different concentrations. Indeed, classifying such similar cell types in the absence of molecular-specific labels has been an important challenge for previous cytometers and even for human perception.

[0178]

[0227] In addition to classifying two cell types that share a similar morphology, GC can accurately classify specific cell types from complex cell mixtures at high throughput. Such techniques are important, for example, when detecting rare circulating tumor cells (CTCs) in a patient's peripheral blood (24 - 27). An image-free GC workflow was applied to classify model cancer cells (MCF-7) from a heterogeneous population of blood cells including peripheral blood mononuclear cells (PBMCs from Astarte Biologics, Inc), lymphocytes, and monocytes. Again, the cytoplasm of all cells was labeled green with FG for classification by image-free GC, while the membranes of only MCF-7 cells were labeled blue with BV421-EpCAM to confirm the GC classification results. The classifier SVM model was first trained by experimentally collecting green fluorescence waveforms for the labeled MCF-7 cells and PBMC cells and computationally mixing them. The model was then tested by flowing the experimentally mixed cells and classifying their cell types one by one. All signals were measured at a throughput exceeding 10,000 cells / second. 1,000 MCF-7 cells and 1,000 PBMCs were used to train the model. 1,000 cells from a random mixture of MCF-7 cells and PBMCs were used to test the model. After performing cross-validation 10 times, an AUC of 0.998 was recorded for the SVM-based classifier of GC waveforms. Figure 4E shows one of the ROC curves, demonstrating the ability for ultra-fast and accurate detection of specific cell types from complex cell mixtures.

[0179] Example 6: Ultra-Fast Precision Cell Sorting

[0228] Figures 5A-5C demonstrate the proof-of-concept of a machine learning-based fluorescence “imaging” activated cell sorter (FiCS). As shown in the left panel of Fig. 5A, the microfluidic device can include three functional parts. The cell flow stream is first focused by 3D hydrodynamic flow focusing, then subjected to random structured light illumination, and finally reaches the sorting region. During the sorting operation, a lead zirconate titanate (PZT) actuator is driven by an input voltage and bent to laterally displace the fluid towards the junction to sort the target cells to the collection outlet. As shown in the right panel of Fig. 5A, for real-time classification and selective isolation of cells, the analog signal measured by the PMT was digitized and then analyzed by an FPGA implementing a trained SVM-based classifier. If the classification result was positive, the FPGA sent a time-delay pulse, as a result, activating the PZT device. The experiment was carried out at a throughput speed of ~3,000 cells / second. The cytoplasm of all MIA PaCa-2, MCF-7, and PBMC cells was labeled green using FG. The membrane of MCF-7 cells (shown in Fig. 5B) and the cytoplasm of MIA PaCa-2 cells (shown in Fig. 5C) were labeled blue using BV421-conjugated EpCAM antibody and anti-pan-cytokeratin primary / AF405-conjugated secondary antibody, respectively.

[0180]

[0229] Figure 5B shows the accurate isolation of MIA PaCa-2 cells from morphologically similar MCF-7 cells. GC directly classified the green fluorescence waveforms without image reconstruction. Panel (i) of Fig. 5B shows the histogram of the maximum blue fluorescence intensity measured for the cell mixture, showing a purity of 0.626 for MIA PaCa-2, while panel (ii) of Fig. 5B shows the histogram of the maximum blue fluorescence intensity for the same mixture after applying FiCS, showing a purity of 0.951 for MIA PaCa-2. A dashed line corresponding to a threshold of 0.05 was used to distinguish the populations of the two cell types.

[0181]

[0230] Figure 5C shows the accurate isolation of model cancer (MIA PaCa-2) cells against a complex mixture of PBMCs. Panel (i) of Figure 5C shows a histogram of the maximum blue fluorescence intensity measured for the cell mixture, indicating a purity of 0.117 for MIA PaCa-2, while panel (ii) of Figure 5C shows a histogram of the maximum blue fluorescence intensity for the same mixture after applying FiCS, indicating a purity of 0.951 for MIA PaCa-2. The dashed line corresponds to the 40 thresholds used to distinguish the populations of the two cell types.

[0182]

[0231] Data size reduction by compressive sensing and avoidance of image reconstruction in GC shorten the computation time required to classify individual waveforms. By combining this efficient signal processing with a microfluidic system, ultra-fast and accurate cell sorting based on real-time analysis of "imaging" data was achieved. The ability to isolate specific cell populations from another population of morphologically similar cells and from complex cell mixtures was demonstrated with high throughput and high precision. The left panel of Fig. 5A shows a microfluidic device made of polydimethylsiloxane. As shown in Fig. 12A, three functional sites were designed. First, as shown in Fig. 12B, the cell flow stream was focused into a dense stream by a 3D hydrodynamic flow focusing (28,29) structure. The cells then received random structured light illumination from the GC and finally reached the sorting junction. A lead zirconate titanate (PZT) actuator was connected to this junction through a channel oriented in a direction perpendicular to the main flow stream as shown in Fig. 12C. For the sorting operation, the actuator was driven by an input voltage to bend the fluid and displace it laterally towards the junction to sort the target cells to the collection outlet. As shown in Fig. 5A and Fig. 12D, for real-time classification and selective isolation of cells, their fluorescence signals were recorded as analog voltages by a PMT, digitized by an analog-to-digital converter, and then analyzed by a field programmable gate array (FPGA, Stratix IV, Altera). An SVM-based classifier was pre-implemented. When the FPGA classified a cell of interest as positive, the FPGA sent a time-delay pulse, which as a result drove the PZT actuator within the chip. The computation time in the FPGA for classifying each compressed waveform was short enough (<10 μs) to enable reproducible sorting. Through experiments, the length of the GC waveform was maintained at approximately 300 μs corresponding to a throughput of approximately 3,000 cells / second. After measuring the green fluorescence waveforms of positive and negative cells with those labels created by the maximum blue fluorescence intensity, the classifier model was constructed offline on a computer and implemented in the FPGA.In the experiment, all the cytoplasm of MIA PaCa-2, MCF-7, and PBMC cells was labeled green using FG. Furthermore, the membrane of MCF-7 cells (shown in Fig. 5B) and the cytoplasm of MIA PaCa-2 cells (shown in Fig. 5C) were labeled blue using BV421-conjugated EpCAM antibody and anti-pan-cytokeratin primary / AF405-conjugated secondary antibody, respectively.

[0183]

[0232] Integrated FiCS enabled the accurate isolation of MIA PaCa-2 cells from MCF-7 cells, which are similar in size, total fluorescence intensity, and apparent morphology. Two hundred waveforms of MIA PaCa-2 cells and two hundred of MCF-7 cells were used to train the SVM model. The two cell types were mixed, and their maximum blue fluorescence intensity was measured using a homemade flow cytometer (analyzer), yielding two distinct peaks corresponding to the two cell types that appeared in the histogram (as shown in panel (i) of Fig. 5B). After applying machine learning-driven FiCS to the same cell mixture by classifying the green fluorescence waveforms, the maximum blue fluorescence intensity of the selected mixture was measured in the same way. As a result, the peak with a stronger intensity corresponding to MCF-7 cells disappeared, and the purity of MIA PaCa-2 cells increased from 0.625 (shown in panel (i) of Fig. 5B) to 0.951 (shown in panel (ii) of Fig. 5B). This confirmed that FiCS can recognize and physically isolate seemingly similar cell types with high accuracy and throughput based on their morphology using only cytoplasmic staining (FG) that does not specifically label the target molecule.

[0184]

[0233] FiCS can also accurately enrich MIA PaCa-2 cells for a complex mixture of PBMCs. Two hundred waveforms of MIA PaCa-2 cells and two hundred of PBMCs were used to train the SVM model. The two cell types were mixed, and their maximum blue fluorescence intensities were measured using a homemade flow cytometer (analyzer). The peak at a stronger intensity corresponding to the population of MIA PaCa-2 cells was relatively small (shown in panel (i) of Figure 5C). After applying FiCS to the same cell mixture by classifying the green fluorescence waveforms, the maximum blue fluorescence intensity of the sorted mixture was measured in the same way. As a result, the purity of MIA PaCa-2 cells increased from 0.117 (shown in panel (i) of Figure 5C) to 0.951 (shown in panel (ii) of Figure 5C). This confirmed that FiCS can significantly enrich model cancer cells against the background of a complex cell mixture with high accuracy and throughput without any specific biomarker.

[0185]

[0234] Recent studies have widely used imaging flow analyzers for the detection and / or characterization of important cells in various fields, including oncology, immunology, and drug screening (30, 31). The ability of GC to significantly increase the analysis throughput and selectively isolate cell populations in real time according to high-content information can lead to the integration of morphological-based analysis with comprehensive downstream omics analysis at the single-cell level. Beyond conventional image generation and processing that rely on limited human knowledge and capabilities, machine learning methods directly applied to compressive modalities can have broad applicability for the real-time application of large and high-dimensional data.

[0186] Example 7: Acquisition of Optical Encoder Pattern

[0235] FIG. 6 shows the acquisition of the optical encoder pattern H(x, y) used in the fluorescence imaging of beads in the SI mode and in the SD mode. The lower left part of FIG. 6 shows the optical setup used to estimate the exact intensity distribution of the encoder H(x, y) (as shown in the upper right part of FIG. 6). A photomask fabricated with a chromium layer having the same pseudo-random hole array was placed on each of the combined image planes before (plane 1 shown in FIG. 6) and after (plane 2 shown in FIG. 6) the samples for the SI mode and the SD mode. The fabricated holes were squares with a side length of 8 μm and were randomly spread over a rectangular area of 800×9,600 μm 2 and had a filling rate of ~1%. The sparsity (filling rate) of the optical structure was designed to make the GC measurement process robust against experimental noise (38, 39). These 8-μm holes correspond to spots of approximately 1 μm on the sample surface after diffraction and geometric contraction and define the spatial resolution of GC-based imaging. To minimize the error due to chromatic aberration caused by each optical component, a thin film of fluorescent polymer was placed on the sample surface and its patterned emission was recorded.

[0187]

[0236] In the SI mode, the polymer film was excited by light passing through the photomask. The fluorescence image of the film was transferred to plane 2 and further transferred to the image sensor through a 4f microscope system equipped with two chromatic aberration correction lenses each having a focal length of 200 mm. In the SD mode, the polymer film was excited by uniformly spread light. The fluorescence image of the film was transferred to plane 2 where the photomask was placed. Photons passing through the pseudo-randomly patterned holes formed its image through a 4f microscope system equipped with two chromatic aberration correction lenses each having a focal length of 200 mm. The scale bar is 50 μm. In the experiment, a blue LED (UHP-T-LED-460UV LED, Prizmatix, Israel) was used as the excitation light source, and a scientific CMOS camera (Flash4.0V2, Hamamatsu Photonics) was used for image acquisition. The objective lens was 20× and obtained from Olympus (UPlanSApo). The imaging beads were purchased from Spherotech Inc. (FP-10052-2 fluorescent yellow particles, diameter: 10.0 - 14.0 μm). The excitation light from the blue LED passed through a 474 / 27 nm bandpass filter (Semrock), was reflected by a dichroic mirror (ZT488rdc, Chroma), and irradiated the sample. The emitted light passed through the mirror and was collected after passing through a 525 / 50 nm bandpass filter (Semrock).

[0188]

[0237] Figure 7 shows the optical setup of the encoder H(x, y) used for three-color imaging of fluorescently labeled cells and the measured intensity distribution. A photomask of a chromium layer with a pseudo-random hole array was placed on the combined image plane behind the sample for the SD mode. The fabricated holes were squares with a side length of 8 μm and randomly spread over a rectangular area of 800×10,800 μm 2 with a filling rate of ~1%.

[0189]

[0238] Light from a cyan laser (GEM473-500, Laser Quantum) passed through a clearing filter (Example 1, Semrock), and light from a UV LED (UHP-T-LED-385UV LED, Prizmatix) passed through a 377 / 50 nm bandpass filter (Example 2, Semrock). These excitation light sources were combined with a 420 long-pass dichroic mirror (DM1, Chroma) and reflected by a triple-edge dichroic mirror (DM2, Di01-R405 / 488 / 594, Semrock), and the sample was uniformly irradiated through a 20x objective lens. The light emitted from the sample passed through DM2, a triple-bandpass emission filter (Em.1, FF01-432 / 523 / 702, Semrock), and an optically encoded photomask. In GC-based imaging, the emitted light after the photomask was split into three color channels by the combined use of dichroic mirrors (DM3 is FF-573-Di01 manufactured by Semrock, and DM4 is ZT488rdc manufactured by Chroma), and recorded using three PMTs through each bandpass filter (Em.2, Em.3, and Em.4 are a 593 long-pass filter, a 535 / 50 nm bandpass filter, and a 435 / 40 nm bandpass filter manufactured by Semrock). For the calibration of H(x, y), a thin film of a fluorescent polymer was placed on the sample surface, and the patterned emission was recorded separately under illumination with different excitations using a 536 / 40 nm bandpass filter (Em.5, Chroma) and an sCMOS camera. The scale bar is 50 μm.

[0190]

[0239] Figure 8 shows the optical setup of the optical encoder used for ultra-high speed multi-color fluorescence imaging and image-free imaging cytometry by GC, and the calibrated intensity distribution. The lower left part of Figure 8 shows the optical setup used for image-free GC, where structured illumination using a diffractive optical element (DOE) is illuminated for each wavelength laser at the combined image plane in front of the sample. The structured illumination is designed to have a 100×1350 pixel pattern with a pseudo-random pattern having ~1% bright spots, and each square pixel has a side length of ~10.5 μm for 405 nm (DOE-1) using a lens with a focal length of 150 mm and ~12 μm for 488 nm (DOE-2). A spatial filter was placed at the combined image plane in front of the sample to block the 0th and multiple diffraction patterns.

[0191]

[0240] Light from a 488 nm blue laser (Cobolt06-MLD488nm, Cobolt) and light from a 405 nm violet laser (Stradus405-250, Vortran) were combined with a long-pass dichroic mirror (DM1, ZT405rdc-UF1 from Chroma) and reflected by a quad-band dichroic mirror (DM2, Di03-R405 / 488 / 561 / 635-t3, Semrock), and the sample was uniformly irradiated through a 20× objective lens. The light emitted from the sample passed through DM2 and was split into two color channels by a dichroic mirror (DM3, FF506-Di03 from Semrock), and recorded using two PMTs through each band-pass filter (Em.1 and Em.2 are 535 / 50 nm band-pass filter and 440 / 40 nm band-pass filter from Semrock). Calibration was performed in the same way using Em.5 and sCMOS as described herein with respect to Figure 7. When performing image-free imaging cytometry (Figure 4), DOE-1 was replaced with a cylindrical lens. The scale bar is 50 μm.

[0192] Example 8: Hydrodynamic flow focusing

[0241] Figures 9A - 9F show the effect of controlling hydrodynamic focusing intensity on the performance of classification by image - free GC. The geometric shape of the flow cell (Hamamatsu Photonics K.K., Japan) shown in Figure 9A brings about hydrodynamic three - dimensional (3D) flow focusing of the inner fluid by the outer fluid, enabling the flowing cells to be kept in a focused state and enabling control of their positions in the image as shown. Further, as shown in Figure 9B, the device enables control of the focusing intensity by varying the flow rate ratio between the inner fluid and the outer fluid. That is, when one type of cell is flowed in a loosely focused state, a more diverse and generalized time waveform is generated compared to the strongly focused state. After comprehensively investigating various combinations of the following focusing intensities, a combination of fluid conditions that enables the construction of a highly generalized, robust, and sensitive classifier model was used: combining loose focusing conditions and tight focusing conditions to train the model and using tight focusing conditions to test the model.

[0193]

[0242] To determine the effect of flow focusing on cell classification using SVM - based image - free GC, as shown in Figures 9C - 9F, various combinations of focusing intensities were comprehensively tested when training and testing the model. To evaluate the classification results of GC, waveforms were randomly selected from the set of waveforms labeled with each type (MCF - 7 or MIA PaCa - 2 cells) regarding training the model. As shown in Figures 9C - 9F, the model improves by increasing the number of waveforms used to train the classifier. To test the model, 1,000 waveforms were selected for each cell type from other waveforms and repeated 10 times to obtain error bars. Here, a tightly focused stream refers to a scenario where the inner flow rate in volume and the outer velocity in flow velocity are 20 μL / min and > 10 m / s, respectively. A loosely focused stream refers to a scenario where the inner velocity is 30 μL / min.

[0194]

[0243] The curve in FIG. 9C shows that when a tightly focused stream is used for both training and testing under highly controlled experimental conditions (minimum experimental error), the accuracy rapidly reaches a high plateau with a small number of training points. The curve in FIG. 9D shows that when a loosely focused stream is used for both training and testing, the accuracy increases slowly and requires a larger number of training points. These results indicate that tightly focused conditions help increase the sensitivity of the classifier. However, a model trained under strict conditions may overfit and lose accuracy if the test conditions are not exactly the same as the training conditions, which is often the case in practice. Indeed, the curve in FIG. 9E shows that when a tightly focused stream is used for training and a loosely focused stream is used for testing, the accuracy increases much more slowly and reaches a lower plateau.

[0195]

[0244] To simultaneously pursue high sensitivity and robustness for cell classification, the training dataset was prepared to include waveforms randomly selected from a dataset collected using both tightly focused and loosely focused streams. When this trained model was used to test a dataset measured using a tightly focused stream, as shown in FIG. 9F, the accuracy increased rapidly and ultimately reached a high plateau. Thus, the model was generalized and functioned well. These results indicate that training the model using a loosely focused stream reduces sensitivity while increasing the robustness of the model with respect to samples, which is equivalent to the generalization ability in a machine learning model. To achieve highly sensitive and robust cytometry, the inventors combined loose focusing conditions and tight focusing conditions for training the model and used tight focusing conditions for testing the model.

[0196] Example 9: Verification of Classification by Image-Free Ghost Cytometry

[0245] Figures 10A-10D show the classification of image-free GC and the BV421-EpCAM-based validation of the image-free GC workflow. To evaluate the performance of image-free GC when classifying temporally modulated green fluorescence intensity, the total blue fluorescence (BV421-EpCAM) intensity of the same cells was quantified. Figure 10A shows a histogram of the total blue fluorescence intensity of MCF-7 cells only, and Figure 10B shows a histogram of the total blue fluorescence intensity of MIA PaCa-2 cells only. Figure 10C shows an intensity histogram for a dataset of an example cell mixture, which corresponds to a single dot in Figure 4D. Throughout the experiments in Figures 4A-4E, the total fluorescence intensity of BV421-EpCAM was used, with 2,500 used as the upper gating threshold for distinguishing MIA PaCa-2 cells and 4,000 used as the lower gating threshold for distinguishing MCF7 cells. The cells excluded by this gating were only 0.7% of the total population and were considered negligible. In contrast to the distinct distribution of the total intensity of BV421, the histogram of the total fluorescence intensity of FG for the same sample cells shows an indistinguishable distribution, as shown in Figure 10D.

[0197]

[0246] Figure 10D shows a histogram of the total fluorescence intensity of FG where distinguishable peaks are not evident. In contrast, Figure 4D shows a histogram of the scores calculated using Equation (9), which has two distinguishable peaks. Each point on the ROC curve in Figure 4D corresponds to a threshold applied to the scores obtained from this trained SVM model. The SVM-based cell classification method demonstrated high performance as measured by an area under the ROC curve (AUC) of 0.971. The sample and fluid conditions employed in the demonstrations of Figures 4 and 11 are summarized in Table 1.

[0198]

Table 1

[0199] Example 10: Confirmation of Consistent Accuracy of SVM-Based Classification

[0247] Figure 11 shows the confirmation of consistent accuracy of SVM-based classification over a wide range of concentration ratios. Simple linear fitting was calculated to demonstrate that the model consistently maintained its accuracy over a wide range of concentration ratios. In Figure 11, if the true positive rate and false positive rate for distinguishing MCF-7 by GC are denoted as α and β, respectively, the results for GC with respect to the results for the measurement of the intensity of BV421 can be estimated as follows. n = αm + β(1 - m) Here, m is the concentration ratio of MCF-7 estimated by the measurement of BV421, and n is that obtained by GC classification. The quantities α and β were calculated by comparing the GC classification results for DAPI measurements on pure solutions of MCF-7 and MIA PaCa-2. The calculated α and β are 0.949 and 0.068 for the fitting shown in Figure 4D. This demonstrates a consistently good agreement between the plot and the fit for different concentration ratios. As shown in panel (i) of Figure 4A, it is difficult to visually identify clearly different morphological features such as size and circularity between the green fluorescent cytoplasmic images of MIA PaCa-2 and MCF-7 cells.

[0200] Example 11: Machine Learning-Based Fluorescent "Imaging" Activated On-Chip Cell Sorting

[0248] Figures 12A - 12D show a machine - learning - based fluorescence “imaging” activated on - chip cell sorting system (FiCS). Figures 12A - 12C show a functional microfluidic device that enables FiCS in combination with a GC system. The cell sorting chip shown in Figure 12A was fabricated using standard soft - lithography techniques. The features of the channels were patterned by photolithography using an SU - 8 photoresist mold and transferred to polydimethylsiloxane (PDMS). The PDMS channels were fixed onto a glass substrate by oxygen plasma - activated bonding. Both the sample and sheath injections were driven by a syringe pump (Harvard Apparatus Pump11 Elite Syringe Pump) at a constant flow rate of approximately 10 μL / min and 400 μL / min, respectively. The chip design (28) shown as a microscope image in Figure 12B enabled the sample flow to be focused into a narrow cell stream by 3D hydrodynamic flow focusing in front of the optical measurement site. The chip design (29) shown as a microscope image in Figure 12C enabled selective cell sorting based on classification. At this sorting site, a piezoelectric PZT actuator was connected to the junction through a channel running perpendicular to the main flow stream. Target cells entering the sorting site were deflected towards the collection outlet by displacement fluid driven by the bending action of the PZT actuator. The diameter of the piezoelectric actuator was 20 mm and the natural resonance frequency was approximately 6.5 kHz. The scale bars in Figures 12B and 12C are 50 μm.

[0201]

[0249] Figure 12D shows the process flow of the real-time classification and electrical control system used in this study. The GC waveform detected as an analog voltage by the PMT was first digitized via an analog-to-digital converter (ADC, a partially modified Texas Instrument AD11445EVM) with a sampling rate of 20 MHz and a resolution of 14 bits. In the SVM-based classification by a field-programmable gate array (FPGA, Stratix IV, Altera) on an FPGA development board (Terasic TR4), each waveform was decimated to 1,000 data points. When the FPGA determines the presence of target cells based on the classification, it automatically outputs a time-delay signal to drive a function generator (Tektronix AGF1022). The rectangular pulse immediately sent from the function generator was amplified via a high-voltage amplifier (Trek model 2210) and sent to the electrodes of the PZT actuator to induce a sorting effect. The FPGA was also connected to a computer to collect training data, upload SVM parameters, set trigger conditions, and store test data.

[0202]

[0250] To evaluate the performance of the cell sorting system, a homemade analyzer including an optical setup with a flow cell (Hamamatsu Photonics) (as shown in FIG. 8) was used to analyze the samples before and after sorting. Here, the maximum blue fluorescence intensity (as a label) and green fluorescence intensity from the sample were simply detected by the trigger conditions applied to the green fluorescence signal.

[0203]

[0251] In the hydrodynamic flow focusing device of the example, the position of the cell stream was controlled in three-dimensional space by changing the ratio of the velocities between the inner flow and the outer flow.

[0204] Example 12: Label-Free Sorting Based on a Partially Transmitted Speckle Pattern

[0252] The systems and methods described herein can be used to perform label-free sorting. Such label-free sorting can be achieved by obtaining phase information of cells flowing at high speed in a microfluidic channel by compressive sensing. This can be done by measuring the modulation waveform of the transmission speckle pattern using a detector (such as a single-pixel PMT or any other detector described herein) when the cells pass through the structured illumination described herein. Compared with phase retrieval imaging techniques, this method can obtain only a part of the entire speckle pattern required to regenerate the image. Nevertheless, it may still be sufficient to classify cells with high accuracy, and this reduction in the optical dimension before acquisition may enable fast and real-time classification and sorting.

[0205]

[0253] FIG. 14 shows a label-free cell sorter using GC. From the waveforms acquired by a photodetector (PD), a field-programmable gate array (FPGA) can implement a machine learning classifier that classifies each cell passing through the PD. Next, the FPGA can send a pulse to a PZT to extrude the cells identified as target cells into adjacent channels.

[0206]

[0254] FIG. 15 shows motion-driven compressive ghost cytometry for a label-free cell sorter. When cells pass through structured illumination, the transmission speckle pattern can be modulated by the structured illumination. By acquiring a part of this speckle pattern with a photodetector (PD), a modulation waveform can be obtained.

[0207]

[0255] MIA PaCa-2 cells and Jurkat cells (two cell lines with similar sizes and apparent morphologies) were classified and sorted using the systems and methods of the present disclosure. The cells were flowed at a rate of 3 m / s, and the waveforms of each cell were acquired at 110 microseconds (μs), which corresponds to a throughput of over 9,000 cells / second. To train and validate the mechanical classifier, MIA PaCa-2 cells were stained with green fluorescence. By training the mechanical classifier using 150 cells of each cell type, the two types of cells in the mixture could be classified with 93% accuracy. When the actually sorted cells were analyzed by fluorescence, they had a purity of 89%. Conventional forward scatter (FSC) and side scatter (SSC) plots indicate that it is difficult to completely classify the two cells by size.

[0208]

[0256] Figure 16A shows the sorting of Jurkat and MIA PaCa-2 cells using a label-free cell sorter. The upper left part of Figure 16A shows the phase contrast image of Jurkat cells. The upper right part of Figure 16A shows the phase contrast image of MIA PaCa-2 cells. The lower left part of Figure 16A shows the FSC and SSC plots of Jurkat cells. The lower right part of Figure 16A shows the FSC and SSC plots of MIA PaCa-2 cells. As shown in Figure 16A, the FSC and SSC plots of Jurkat cells and MIA PaCa-2 cells overlap, making it difficult to distinguish the two types of cells using size determination by FSC and SSC plots alone.

[0209]

[0257] Figure 16B shows the histograms of Jurkat and MIA PaCa-2 cell populations before and after cell sorting. Only MIA PaCa-2 cells were stained green. The upper part of Figure 16B shows the sample before sorting. The lower part of Figure 16B shows the sample after sorting. After sorting, the ratio of MIA to PaCa-2 increased from 51% to 89%.

[0210] Example 13: Classification of Dead and Early Apoptotic Induced Pluripotent Stem Cells (iPSCs)

[0258] Figure 17A shows an example of a scatter plot of fluorescence intensities of propidium iodide (PI) and annexin V for classifying induced pluripotent stem cells (iPSCs) as viable, early apoptotic, or dead. As shown in Figure 17A, the populations within the blue, red, and green regions are labeled as viable cells, dead cells, and early apoptotic cells, respectively.

[0211]

[0259] Figure 17B shows an example of a scatter plot of forward scatter (FSC) and side scatter (SSC) for iPSCs. The blue, red, and green dots correspond to viable cells, dead cells, and early apoptotic cells, respectively. Using conventional methods such as FSC and SSC, viable cells can be mostly distinguished from dead / apoptotic cells, but it is difficult to distinguish between dead cells and apoptotic cells from each other.

[0212]

[0260] Figure 17C shows an example of an ROC curve and an SVM score histogram for the classification of viable and dead cells using the system and method of the present disclosure. The dotted line is the ROC curve for classification using only conventional FSC and SSC for reference. As shown in Figure 17C, conventional FSC and SSC enable the discrimination of viable and dead cells with an AUC of 0.876, while the classification using the system and method of the present disclosure achieves an AUC of 0.995, showing very excellent performance.

[0213]

[0261] Figure 17D shows an example of an ROC curve and an SVM score histogram for the classification of viable and early apoptotic cells using the system and method of the present disclosure. The dotted line is the ROC curve for classification using only conventional FSC and SSC for reference. Similar to the scenario shown in Figure C, the classification by the system and method of the present disclosure shows excellent performance (AUC = 0.947) compared to that obtained using conventional FSC and SSC (AUC = 0.875).

[0214]

[0262] Figure 17E shows an example of an ROC curve and an SVM score histogram for the classification of dead cells and early apoptotic cells using the systems and methods of the present disclosure. The dotted line is the ROC curve for classification using only conventional FSC and SSC for reference. Using conventional FSC and SSC alone, it is difficult to distinguish between the two populations (AUC = 0.657), but with the systems and methods of the present disclosure, they can be distinguished with an AUC of 0.931.

[0215]

[0263] The systems and methods described herein may demonstrate significant applicability to cell manufacturing processes for use in processes such as regenerative medicine and cell therapy. In such processes, the quality of the cells should be monitored and controlled according to various requirements. The systems and methods described herein may be applied to a pipeline that uses induced pluripotent stem cells (iPSCs). The pipeline may begin with the thawing of cryopreserved iPSCs, which may then pass through multiple differentiation steps to yield a final cell product. Throughout this pipeline, it may be necessary to monitor the viability, expression state, and purity of the cell population, and it may be necessary to sort the cells. Previous tests of iPSCs have required molecular staining that is often toxic to the cells or otherwise has an adverse effect on the cells (e.g., due to an immune response). Additionally, chemical contamination of the cell product may have potential side effects for the patient when they are introduced into the human body. In contrast, the systems and methods described herein may enable high-speed evaluation of cells based on their viability, expression state, and purity without staining actual production line cells, thus solving problems potentially caused by molecular staining.

[0216]

[0264] The systems and methods described herein can distinguish cells in the early apoptotic state from live cells with high accuracy, not only dead cells. This can be essential when monitoring cell populations for quality control, because the remaining dead cells can potentially cause stress to surrounding cells and alter the differentiation of iPSCs or other stem cells. Further, contamination by dead cells and early apoptotic cells can lead to inaccurate estimation of the number of functional cells in transplantation. Viability and apoptosis analysis of cultured iPSCs separated from culture flasks were performed. As shown in FIG. 17A, training labels were created using the fluorescence intensities of propidium iodide (PI) and annexin V, which are indicators of dead cells and early apoptotic cells, respectively. Dead cells and early apoptotic cells can be distinguished from live cells with some accuracy from conventional FSC and SSC plots, as shown in FIG. 17B. However, such classification may not be completely accurate, and conventional methods may require additional dead cell exclusion staining to eliminate dead cells.

[0217]

[0265] Using the systems and methods of the present disclosure, a training data set of waveforms was prepared using a cell population gated and labeled as live cells, dead cells, and early apoptotic cells. As shown in FIG. 17B, dead cells were distinguished from live cells having a high ROC-AUC of 0.996±0.001 (as shown by the solid line in FIG. 17C), and early apoptotic cells were distinguished from live cells having a ROC-AUC of 0.947±0.005 (as shown by the solid line in FIG. 17D). In contrast, using only conventional FSC and SSC data with the same labels, conventional flow cytometry achieved limited performance for dead-live and apoptosis-live discrimination, with ROC-AUCs of 0.890±0.009 (as shown by the dashed line in FIG. 17C) and 0.875±0.013 (as shown by the dashed line in FIG. 17D), respectively. In addition, the systems and methods of the present disclosure were able to distinguish early apoptotic cells from dead cells with a ROC-AUC of 0.931±0.006 (as shown by the solid line in FIG. 17E), which was difficult with only conventional FSC and SSC achieving a ROC-AUC of 0.657±0.012 (as shown by the dashed line in FIG. 17E).

[0218] Example 14: Classification of Differentiated and Undifferentiated iPSCs

[0266] FIG. 18A shows an example of a scatter plot of calcein AM and rBC2LCN-635 intensities for a mixture of neural progenitor cells (NPCs) and iPSCs used to train a machine learning classifier. Populations within the blue and red regions were labeled as NPCs and iPSCs, respectively.

[0219]

[0267] FIG. 18B shows an example of an ROC curve (outer panel of FIG. 18B) and an SVM score histogram (inner panel of FIG. 18B) for the classification of NPCs and iPSCs using the systems and methods of the present disclosure. The AUC of the ROC curve was 0.933. The blue and red SVM score histograms correspond to cells labeled as NPCs and iPSCs, respectively, derived from FIG. 18A.

[0220]

[0268] FIG. 18C shows an example of a scatter plot of calcein AM and rBC2LCN-635 intensities for a mixture of hepatoblasts and iPSCs used to train a machine learning classifier. Populations within the blue and red regions were labeled as hepatoblasts and iPSCs, respectively.

[0221]

[0269] FIG. 18D shows examples of an ROC curve (outer panel of FIG. 18D) and an SVM score histogram (inner panel of FIG. 18D) for the classification of hepatoblasts and iPSCs using the systems and methods of the present disclosure. The AUC of the ROC curve was 0.947. The blue and red SVM score histograms correspond to cells labeled as hepatoblasts and iPSCs, respectively, derived from FIG. 18C.

[0222]

[0270] The systems and methods of the present disclosure can classify undifferentiated cells from differentiated cells, which can represent one of the most important processes in a cell manufacturing pipeline. Such classification can be important because the remaining undifferentiated cells can still have the ability to proliferate, which can ultimately turn the cells into cancer cells. Undifferentiated iPSCs were compared to differentiated cells in the form of neural progenitor cells (NPCs) and hepatoblasts, both derived from the same iPSCs. In the training process, cells were labeled with an undifferentiated marker that stained only the undifferentiated iPSCs. A label-free cytometry-like mode was also implemented. The training label was created by staining the cells with rBCLCN-635 (Wako), a marker indicating undifferentiated cells, and measuring the rBCLCN-635 intensity (shown in FIGS. 18A and 18C). Using the waveforms for each cell type labeled with the undifferentiated marker as a training dataset, the classifier distinguished iPSCs and NPCs with an ROC-AUC of 0.933±0.008 (as shown in FIG. 18B) and iPSCs and hepatoblasts with an ROC-AUC of 0.944±0.004 (as shown in FIG. 18D), demonstrating its high classification ability. In contrast, using only conventional FSC and SSC information, the ROC-AUC for the classification of each pair of cells was limited to 0.862±0.013 and 0.697±0.020, respectively.

[0223] Example 15: Classification of Cells Based on Function

[0271] FIG. 19A shows an example of a histogram of the fab FITC intensity of T cells. Cells to the right of the red threshold line were labeled as CAR T cells. The threshold was determined by a negative control sample.

[0224]

[0272] FIG. 19B shows an example of an ROC curve and an SVM score histogram for classifying different states in the cell cycle using the labels from FIG. 19A.

[0225]

[0273] FIG. 19C shows an example of a histogram of the 2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose (2-NBDG) intensity of raji cells. Cells to the left of the green threshold line were labeled as low glucose cells, and cells to the right of the red threshold were labeled as high glucose cells.

[0226]

[0274] FIG. 19D shows an example of an ROC curve and an SVM score histogram for classifying cells with high and low levels of glucose using the labels from FIG. 19C.

[0227]

[0275] The systems and methods described herein can be used to improve the quality of a cell population by selecting cells based on their function. This can be important, for example, in cancer treatments where immune cells can be modified to attack cancer cells.

[0228]

[0276] The classification of chimeric antigen receptor (CAR) T cells from a mixture of non-transduced T cells and CAR T cells was performed using the systems and methods of the present disclosure, using machine learning trained on a fab marker expressed only on CAR T cells, as shown in FIG. 19A. The ROC-AUC was 0.887 ± 0.007 (as shown in FIG. 19B), indicating that the systems and methods of the present disclosure can purify CAR T cells with high precision in a label-free manner such that these cells can be used directly to treat patients.

[0229]

[0277] Cell classification based on glycolytic levels was also performed. Differences in glycolytic levels result in different cell metabolisms and different cell activities. For example, in CAR T cells, cells with low glucose levels have been reported to have higher efficacy against cancer. To monitor glycolytic levels, 2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose (2-NBDG), a fluorescent glucose analog, is commonly used, but it is highly toxic to cells. Therefore, conventional methods for sorting live cells based on glycolytic levels have been difficult. Raji cells were treated with 2-NBDG. Cells with high and low levels of 2-NBDG within the entire population were extracted offline (as shown in Figure 19C). Using these cells with high and low 2-NBDG levels, a classifier was trained and then verified. As a result, the classifier was able to classify the two populations with an accuracy of 0.88 (as shown in Figure 19D), indicating that the systems and methods of the present disclosure can classify cells based on their glycolytic levels.

[0230] Example 16: Detection of Specific Cell Types in a Mixed Fluid

[0278] The systems and methods of the present disclosure may be used to detect or sort specific cell types (such as diseased cells) in a mixed fluid (such as blood, urine, tears, saliva, pleural effusion, cerebrospinal fluid, or any other fluid described herein). The systems and methods may enable the detection or sorting of such specific cell types without the need for labeling with fluorophores or immunostaining with antibodies, as may be required in various conventional laboratory tests.

[0231]

[0279] Figures 20A - I show label-free detection of various types of white blood cells in a mixed fluid.

[0232]

[0280] Figure 20A shows a training dataset containing a population of neutrophils. Neutrophils were characterized by a relatively high level of expression of the CD66 marker and a relatively low level of expression of the CD193 marker. Neutrophils were passed through the patterned optical structures described herein and the optical signals corresponding to the neutrophils were recorded.

[0233]

[0281] Figure 20B shows examples of label-free optical signals corresponding to neutrophils and non-neutrophils. The optical signals were acquired without prior labeling of neutrophils and non-neutrophils. Using the optical signals, a machine learning classifier was trained to distinguish neutrophils from non-neutrophils.

[0234]

[0282] Figure 20C shows an example of an SVM score histogram for the classification of neutrophils and non-neutrophils. The classification was acquired without prior labeling of neutrophils and non-neutrophils. The SVM classification achieved an accuracy of 91.4% and an ROC-AUC of 0.941.

[0235]

[0283] Figure 20D shows a training dataset containing a population of eosinophils. Eosinophils were characterized by a relatively high level of expression of the CD66 marker and a relatively high level of expression of the CD193 marker. Eosinophils were passed through the patterned optical structures described herein and the optical signals corresponding to the eosinophils were recorded.

[0236]

[0284] Figure 20E shows examples of label-free optical signals corresponding to eosinophils and non-eosinophils. The optical signals were acquired without prior labeling of eosinophils and non-eosinophils. Using the optical signals, a machine learning classifier was trained to distinguish eosinophils from non-eosinophils.

[0237]

[0285] Figure 20F shows an example of an SVM score histogram for the classification of eosinophils and non-eosinophils. The classification was acquired without prior labeling of eosinophils and non-eosinophils. The SVM classification achieved an accuracy of 94.7% and an ROC-AUC of 0.992.

[0238]

[0286] FIG. 20G shows a training data set containing a population of basophils. Basophils were characterized by a relatively low level of expression of the CD66 marker and a relatively low level of expression of the CD193 marker. The basophils were passed through the patterned optical structures described herein and the optical signals corresponding to the basophils were recorded.

[0239]

[0287] FIG. 20H shows examples of label-free optical signals corresponding to basophils and non-basophils. The optical signals were acquired without prior labeling of basophils and non-basophils. The optical signals were used to train a machine learning classifier to distinguish basophils from non-basophils.

[0240]

[0288] FIG. 20I shows an example of an SVM score histogram for the classification of basophils and non-basophils. The classification was acquired without prior labeling of basophils and non-basophils. The SVM classification achieved an accuracy of 94.0% and an ROC-AUC of 0.970.

[0241]

[0289] FIGS. 21A - 22B show label-free detection of diseased cells in a mixed fluid.

[0242]

[0290] FIG. 21A shows an example of an SVM score histogram for the classification of oral cells and HeLa cells. The oral cells and HeLa cells were passed through the patterned optical structures described herein and the optical signals corresponding to the oral cells and HeLa cells were recorded. The optical signals were acquired without prior labeling of the oral cells or HeLa cells. The optical signals were used to train a machine learning classifier to distinguish oral cells from HeLa cells. The SVM classification achieved an ROC-AUC of 0.995.

[0243]

[0291] Figure 21B shows an example of an SVM score histogram for the classification of BM1 cells and K562 cells. The BM1 cells and K562 cells were flowed through the patterned optical structures described herein, and the optical signals corresponding to the BM1 cells and K562 cells were recorded. The optical signals were acquired without pre-labeling of the BM1 cells or K562 cells. Using the optical signals, a machine learning classifier was trained to distinguish BM1 cells from K562 cells. An ROC-AUC of 0.969 was obtained by SVM classification.

[0244] Example 16: Differentiation of Different Types of Particles Simultaneously Present in the Detection Region

[0292] The systems and methods of the present disclosure can be applied to distinguish different types of particles even when different types of particles are simultaneously present within the region where the detector of the present disclosure has sensitivity. When two or more particles are simultaneously located in such a region, an overlap can occur in the signals detected by the detector. Since the systems and methods described herein can be expressed in terms of linear equations, such events can be represented as the sum of the signals resulting from each particle independent of all other particles within the region.

[0245]

[0293] The systems and methods described herein can be used to learn different waveforms corresponding to different types of particles, as described herein. Such waveforms can be arbitrarily summed using any time offset to account for different particles moving through the region during different time intervals. A wide variety of such sums can be generated and compared to the measured waveform resulting from the simultaneous presence of different types of particles. In this way, the particles can be distinguished.

[0246]

[0294] Alternatively, or in combination, the machine learning procedures described herein may be trained using the overlapping waveforms resulting from the simultaneous presence of different types of particles.

[0247]

[0295] FIG. 22A shows examples of signals due to the first type of particle (upper part of FIG. 22A), signals due to the second type of particle (central part of FIG. 22A), and signals due to the simultaneous presence of the first and second types of particles (lower part of FIG. 22A).

[0248]

[0296] FIG. 22B shows an example of a confusion matrix for distinguishing different types of particles that are simultaneously present in the detection region of the present disclosure. The confusion matrix shows six states generated for a mixture of particles including MIA PaCa-2 cells and MCF-7 cells. As shown in FIG. 22B, the systems and methods of the present disclosure accurately distinguished MIA PaCa-2 from MCF-7 cells even in the case of overlapping signals corresponding to the simultaneous presence of two types of cells in the region detectable by the detector.

[0249]

[0297] Preferred embodiments of the present invention are shown and described herein, but it will be apparent to those skilled in the art that such embodiments are provided by way of example only. The present invention is not intended to be limited by the specific examples provided herein. Although the present invention has been described with reference to the foregoing specification, the description and illustration of the embodiments herein are not intended to be construed in a limiting sense. Those skilled in the art will envision numerous variations, modifications, and substitutions without departing from the present invention. Further, it should be understood that all aspects of the present invention are not limited to the specific depictions, configurations, or relative ratios described herein, which depend on various conditions and variables. It should be understood that various alternatives to the embodiments of the present invention described herein may be used in practicing the present invention. Accordingly, it is intended that the present invention include any such alternative, modification, variation, or equivalent thereof. The following claims define the scope of the present invention, and it is intended that the methods and structures within these claims and their equivalents be encompassed thereby.

Claims

1. 1. A method for particle processing or analysis comprising: (a) acquiring spatial information during motion of particles relative to a patterned optical structure; (b) compressing the spatial information into signals that arrive sequentially at a detector; (c) using the signals to identify at least a subset of the particles with an accuracy of at least 70%; and (d) sorting the at least a subset of the particles identified in (c) into one or more groups at a rate of at least 10 particles / second.

2. 2. The method of claim 1 , wherein (a) comprises the steps of: (i) directing light from a light source through the patterned optical structure; (ii) directing light from the patterned optical structure to the particles; and (iii) directing light from the particles to the detector.

3. 3. The method of claim 1 or 2, wherein (a) comprises the steps of: (i) directing light from a light source to the particles; (ii) directing light from the particles through the patterned optical structure; and (iii) directing light from the patterned optical structure to the detector.

4. The method of claim 2 or 3, wherein the light comprises ultraviolet light or visible light.

5. 4. The method of claim 1, wherein (b) comprises applying a two-stage iterative shrinkage / thresholding (TwIST) procedure.

6. 6. The method of claim 1 , wherein (c) comprises a step of computationally reconstructing the morphology of the particles at least in part through the combined use of one or more temporal waveforms comprising one or more intensity distributions imparted by the patterned optical structure.

7. The method of claim 1 , wherein the particles comprise one or more biological particles.

8. The method of claim 1 , wherein the particles comprise one or more cells.

9. The method of claim 1 , wherein the particles comprise one or more rare cells.

10. 10. The method of claim 1, wherein the particles comprise one or more cancer cells.

11. The method of claim 1 , wherein the particles comprise one or more circulating tumor cells.

12. 12. The method of claim 1, wherein (c) comprises applying one or more machine learning classifiers to a compressed waveform corresponding to the signal to identify the at least a subset of particles.

13. 13. The method of claim 12, wherein the one or more machine learning classifiers are selected from the group consisting of support vector machines, random forests, artificial neural networks, convolutional neural networks, deep learning, ultra-deep learning, gradient boosting, adapoost, decision trees, linear regression, and logistic regression.

14. 14. The method of claim 1, wherein the particles are analyzed without image reconstruction.

15. The method of claim 1 , wherein the detector comprises a single pixel detector.

16. The method of claim 15 , wherein the single pixel detector comprises a photomultiplier tube.

17. 17. The method of claim 1, further comprising the step of reconstructing one or more images of the particles.

18. The method of claim 17 , wherein the one or more images include a fluorescent image.

19. 19. The method of claim 17 or 18, further comprising the step of reconstructing a plurality of images of the particle, each of the plurality of images comprising a different wavelength or range of wavelengths.

20. 20. The method of any one of claims 17 to 19, wherein the one or more images are free of blurring artifacts.

21. 21. The method of claim 1, wherein the particles move relative to the patterned optical structure at a speed of at least 1 m / s.

22. 22. The method of any one of claims 1 to 21, wherein the velocity is at least 100 particles / second.

23. 23. The method of any one of claims 1 to 22, wherein the velocity is at least 1,000 particles / second.

24. 24. The method of claim 1, wherein (d) comprises collecting one or more of the groups to produce a concentrated particulate mixture.

25. 25. The method of claim 24, wherein the one or more groups have a purity of at least 70%.

26. 26. The method of any one of claims 1 to 25, further comprising subjecting one or more particles of said one or more groups to one or more assays.

27. 27. The method of claim 26, wherein the one or more assays are selected from the group consisting of lysis, nucleic acid extraction, nucleic acid amplification, nucleic acid sequencing, and protein sequencing.

28. 28. The method of claim 1, further comprising, prior to (a), subjecting the particles to hydrodynamic flow focusing.

29. 29. The method of claim 1, further comprising collecting a partial transmission speckle pattern of the particle as the particle moves relative to the patterned optical structure.

30. 30. The method of claim 1, wherein the patterned optical structure comprises a regular patterned optical structure.

31. 31. The method of claim 1 , wherein the patterned optical structure comprises a random patterned optical structure.

32. The method of claim 31 , wherein the disordered optical structures comprise non-periodic patterned optical structures.

33. The method of claim 31 , wherein the disordered optical structures comprise randomly or pseudo-randomly patterned optical structures.

34. 34. The method of claim 1, wherein the optical structure comprises a static optical structure.

35. An image-free optical method for classifying or sorting particles based at least in part on the morphology of said particles without the use of non-native labels with an accuracy of at least 70%.

36. 36. The method of claim 35, wherein the particles comprise one or more biological particles.

37. 37. The method of claim 36, wherein the biological particle comprises one or more cells.

38. 1. A system for particle processing or analysis, comprising: a fluid flow path configured to direct particles; a detector in sensing communication with at least a portion of the fluid flow path; and one or more computer processors operably coupled to the detector, the one or more computer processors individually or collectively programmed to: (a) acquire spatial information during motion of particles relative to a patterned optical structure; (b) compress and convert the spatial information into signals that sequentially arrive at the detector; (c) use the signals to identify at least a subset of the particles with an accuracy of at least 70%; and (d) sort the at least the subset of the particles identified in (c) into one or more groups at a rate of at least 10 particles / second.

39. 40. The system of claim 38, wherein the detector is a single pixel detector.

40. 1. A non-transitory computer readable medium comprising machine-executable code that, when executed by one or more computer processors, implements a method for particle analysis, the method comprising: (a) acquiring spatial information during motion of particles relative to a patterned optical structure; (b) compressing the spatial information into signals that arrive sequentially at a single pixel detector; (c) using the signals to identify at least a subset of the particles with an accuracy of at least 70%; and (d) sorting the at least the subset of the particles identified in (c) into one or more groups at a rate of at least 10 particles / second.

41. 1. A method for cell processing or analysis comprising: (a) acquiring spatial information during movement of cells relative to a patterned optical structure; (b) condensing the spatial information into signals that sequentially arrive at a detector; (c) using the signals to identify at least a subset of the cells as cancerous; and (d) sorting at least the subset of the cells identified in (c) into one or more groups of cancerous cells and one or more groups of non-cancerous cells.

42. 1. A method for cell processing or analysis comprising: (a) acquiring spatial information during movement of cells relative to a patterned optical structure; (b) condensing the spatial information into signals that sequentially arrive at a detector; (c) using the signals to identify at least a subset of the cells as therapeutic; and (d) sorting the at least the subset of cells identified in (c) into one or more groups of therapeutic cells and one or more groups of non-therapeutic cells.

43. 1. A method for identifying one or more target cells from a plurality of cells, comprising: (a) acquiring spatial information during movement of the plurality of cells relative to a patterned optical structure; and (b) inputting the spatial information into a trained machine learning algorithm to identify the one or more target cells from the plurality of cells.

44. 1. A method for cell processing comprising: (a) acquiring spatial information of a plurality of cells; and (b) using at least said spatial information to separate or isolate a subset of said plurality of cells from said plurality of cells at a rate of at least 1,000 cells / second.

45. 1. A method for processing one or more target cells from a plurality of cells, comprising: (a) acquiring spatial information during movement of the plurality of cells relative to a patterned optical structure; (b) identifying the one or more target cells from the plurality of cells using the spatial information; and (c) separating or isolating the one or more target cells from the plurality of cells at a rate of at least 10 cells / second based at least in part on the one or more target cells identified in (b).

46. 46. ​​The method of claim 45, wherein (a) comprises the steps of: (i) directing light from a light source through the patterned optical structure; (ii) directing light from the patterned optical structure to the plurality of cells; and (iii) directing light from the plurality of cells to a detector.

47. 47. The method of claim 45 or 46, wherein (a) comprises the steps of: (i) directing light from a light source to the plurality of cells; (ii) directing light from the plurality of cells through the patterned optical structure; and (iii) directing light from the patterned optical structure to a detector.

48. 48. The method of any one of claims 45 to 47, wherein the patterned optical structure comprises a random patterned optical structure.

49. 49. The method of any one of claims 45 to 48, wherein (c) comprises a step of computationally reconstructing the morphology of the cell at least in part through the combined use of one or more temporal waveforms comprising one or more intensity distributions imparted by the patterned optical structure.

50. 50. The method of any one of claims 45 to 49, wherein the target cells comprise one or more cancer cells or circulating tumor cells.

51. 51. The method of any one of claims 45 to 50, wherein the target cells comprise one or more therapeutic cells.

52. 52. The method of claim 51 , wherein the target cells comprise one or more members selected from the group consisting of stem cells, mesenchymal stem cells, induced pluripotent stem cells, embryonic stem cells, cells differentiated from induced pluripotent stem cells, cells differentiated from embryonic stem cells, genetically engineered cells, blood cells, red blood cells, white blood cells, T cells, B cells, natural killer cells, chimeric antigen receptor T cells, chimeric antigen receptor natural killer cells, cancer cells, and blast cells.

53. 53. The method of any one of claims 45 to 52, wherein (b) comprises applying one or more machine learning classifiers to compressed waveforms corresponding to the spatial information to identify the one or more target cells.

54. 54. The method of claim 53, wherein the one or more machine learning classifiers achieve one or more of sensitivity, specificity, and accuracy of at least 70%.

55. 55. The method of claim 53 or 54, wherein the one or more machine learning classifiers are selected from the group consisting of support vector machines, random forests, artificial neural networks, convolutional neural networks, deep learning, ultra-deep learning, gradient boosting, adapoost, decision trees, linear regression, and logistic regression.

56. 56. The method of any one of claims 45 to 55, wherein the plurality of cells is processed without image reconstruction.

57. 57. The method of any one of claims 45 to 56, wherein the detector comprises a single pixel detector.

58. 58. The method of claim 57, wherein the single pixel detector comprises a photomultiplier tube.

59. 59. The method of any one of claims 45 to 58, further comprising the step of reconstructing one or more images of the plurality of cells.

60. 54. The method of claim 53, further comprising reconstructing a plurality of images of the plurality of cells, each of the plurality of images comprising a different wavelength or range of wavelengths.

61. 55. The method of claim 53 or 54, wherein the one or more images are free of blurring artifacts.

62. 62. The method of any one of claims 45 to 61, wherein the plurality of cells move relative to the patterned optical structure at a speed of at least 1 m / s.

63. 63. The method of any one of claims 45 to 62, wherein (c) comprises: (i) sorting the plurality of cells into one or more groups of sorted cells based on results of analyzing the plurality of cells; and (ii) collecting the one or more target cells from the one or more groups of sorted cells.

64. 64. The method of any one of claims 45-63, wherein (c) comprises sorting the plurality of cells into one or more groups of sorted cells based on morphology of the plurality of cells.

65. 65. The method of claim 64, wherein the sorting step is accomplished at a rate of at least 10 cells / second.

66. 66. The method of claim 64 or 65, further comprising collecting one or more of the group of sorted cells to produce an enriched cell mixture.

67. 67. The method of any one of claims 64 to 66, wherein the one or more groups of sorted cells have a purity of at least 70%.

68. 68. The method of any one of claims 64 to 67, further comprising subjecting one or more cells of said one or more groups of sorted cells to one or more assays.

69. 69. The method of claim 68, wherein the one or more assays are selected from the group consisting of lysis, nucleic acid extraction, nucleic acid amplification, nucleic acid sequencing, and protein sequencing.

70. 70. The method of any one of claims 45 to 69, further comprising the step of subjecting the cells to hydrodynamic flow focusing prior to (a).

71. 71. The method of any one of claims 45 to 70, further comprising collecting a partial transmission speckle pattern of the plurality of cells as the plurality of cells move relative to the patterned optical structure.

72. 72. The method of any one of claims 45 to 71, wherein the spatial information corresponds to features, characteristics, or information regarding the plurality of cells.

73. 73. The method of claim 72, wherein the spatial information corresponds one-to-one with the features, characteristics, or information regarding the plurality of cells.

74. 74. The method of claim 72 or 73, wherein the features, characteristics or information about the plurality of cells comprises one or more members selected from the group consisting of metabolic state, proliferation state, differentiation state, maturation state, expression of a marker protein, expression of a marker gene, cell morphology, organelle morphology, organelle location, organelle size or extent, cytoplasmic morphology, cytoplasmic location, cytoplasmic size or extent, nuclear morphology, nucleus location, nucleus size or extent, mitochondrial morphology, mitochondrial location, mitochondrial size or extent, lysosomal morphology, lysozyme location, lysozyme size or extent, intracellular molecule distribution, intracellular peptide, polypeptide or protein distribution, intracellular nucleic acid distribution, intracellular sugar or polysaccharide distribution, and intracellular lipid distribution.

Images