JP2008533440A - Analysis of blood and cells using an imaging flow cytometer - Google Patents

Abstract

  Multiple mode / multispectral images of the cell population are collected simultaneously. The photometric and / or morphological features identifiable in the image are used to separate the cell population into multiple subpopulations. If the cell population includes diseased cells and healthy cells, the images can be separated into healthy and diseased subpopulations. If the cell population does not contain disease cells, one or more ratios of different cell types of patients who do not have a disease state can be compared with corresponding ratios of patients with a disease state to detect disease states It becomes possible to do. For example, blood cells can be separated into different types based on the image, and an increase in the number of lymphocytes, which is a phenomenon related to chronic lymphocytic leukemia, can be easily detected.

Description

  The present invention relates to a disease condition detection method for detecting a disease condition from images collected from a cell population, and more particularly, to a disease condition detection method for blood and cell analysis using an imaging flow cytometer. About.

  Cellular hematopathology has traditionally included morphological analysis and cytoenzymology of morphological analysis of blood smears with Maygrunwald / Giemsa staining or Wright / Giemsa staining. Identification and inspection has been performed by a simple slide-based approach. In addition, in order to understand these pathologies, other methods such as flow cytometry analysis of cell populations and polymerase chaining to determine gene expression, gene mutation, chromosomal translocation and replication Molecular methods such as PCR (polymerase chain reaction) and in situ hybridization were added.

  The use of such techniques has led to improvements in diagnostic capabilities, understanding disease mechanisms and progression, and the evaluation of new therapies, but such techniques are related to standardization and robustness, respectively. It has problems and most have not been used in routine clinical tests.

  Traditional hematology clinical laboratory testing involves techniques for rapid and automatic analysis of large numbers of peripheral blood samples with minimal human intervention. Abbott Laboratories (Abbott Park, Illinois), Beckman Coulter Inc. (Fullerton, CA), and companies such as TOA Corporation (Kobe, Japan), these technologies have advanced with respect to moderately increasing the throughput, accuracy of the analysis, and the information content collected during each sample inspection. continuing. However, for samples that suggest cellular blood pathology, i.e., out of tolerance for certain parameters, conventional slide-based techniques are widely used to determine possible causes of abnormalities.

  In hematology, diagnostic criteria are based on morphological identification of cell number, size, shape, and abnormalities in staining patterns. These have been complemented for decades by cell population analysis by staining various cell surface determinants with monoclonal antibodies and collecting data by flow cytometry, but the most important element of diagnostic evaluation Is a visual inspection of peripheral blood smears and a biopsy of bone marrow and lymph nodes using a microscope, which allows subjective classification of the presumed abnormalities.

  Evaluating a patient's tissue and blood smear manually is cumbersome and time consuming, and the variability between laboratories and observers is very large. This process includes staining variability (which adversely affects longitudinal analysis), evaluator bias, and suboptimal sample preparation (blood smears are often "contaminated" cells and irregular lymphocytes) , Affected by many causes of variability and negligence. Manually classifying hundreds of cells by morphological overview is poor and unreliable in assessing changes observed over time or as a result of treatment.

  Chronic lymphocytic leukemia (CLL) is a type of cancer in which the bone marrow produces excess lymphocytes (a type of white blood cell) due to a malignant transformation event (eg, chromosomal translocation). This CLL is the most common type of leukemia in Western society. Normally, stem cells (immature cells) grow into mature blood cells by a process of differentiation that is ordered by the bone marrow. There are three types of mature blood cells: It is (1) red blood cells that carry oxygen to all tissues in the body; (2) white blood cells that fight infection; and (3) platelets that form thrombus to help prevent bleeding. In general, the number and type of these blood cells are strictly defined. In CLL, a type of white blood cell called lymphocyte is chronically overproduced pathologically. There are three types of lymphocytes: It includes (1) B lymphocytes that produce antibodies that help fight infection; (2) T lymphocytes that help B lymphocytes produce antibodies that fight infection; and (3) attack cancer cells and viruses. It is a killer cell. CLL is a disease with an increase in the number of B lymphocyte cells in peripheral blood, generally representing clonal expansion of malignantly transformed CD5 + B lymphocyte cells.

  Currently, this disease state is being treated using established chemotherapy, but many new therapies involving monoclonal antibodies (eg, Rituximab) that target cell surface antigens expressed on CLL cells Has been developed. According to recent data from the National Cancer Data Base, the 5-year survival rate for this disease is about 48%, with only 23% of patients with this disease surviving after 10 years. In recent years, many prognostic factors have been identified that can stratify a patient population into two subpopulations with individual clinical outcomes. Factors that tend to correlate with decreased survival are: It is an increase in ZAP70 expression (tyrosine kinase required for T lymphocyte cell signaling), CD38 expression, non-mutated Ig Vh gene, and chromosomal abnormalities. However, routine assessment of these factors has not reached standard clinical trials due to technical problems in data standardization and interpretation.

  Morphological assessment is still the “gold standard” of hematopathology, and patients with CLL show morphological heterogeneity. Attempts have been made to correlate specific morphological profiles with clinical prognosis, but there is currently no widely recognized association, and morphological subclassification of CLL and its correlation with clinical prognosis remains Searching.

US Pat. No. 6,249,341 US Pat. No. 6,211,955 US Pat. No. 6,473,176 US Pat. No. 6,583,865 US Patent Application No. 09 / 989,031 US Pat. No. 6,608,682 US Pat. No. 6,532,061 US Pat. No. 6,763,149

  Therefore, it automatically analyzes blood, including peripheral blood leukocytes, and cellular components such as bone marrow and lymph nodes (cells are easy to process in suspension) to facilitate investigation of blood-related diseases and abnormalities. It would be desirable to provide a method and apparatus suitable for doing so. To provide a method and apparatus for rapidly collecting images from blood and other body fluids (and cell compartments) and to analyze such images to identify cell abnormalities or cell distribution abnormalities associated with disease It is particularly desirable to provide software tools.

  The present invention has been made in view of such a situation, and an object of the present invention is to provide a disease state detection method for detecting a disease state from an image collected from a cell population.

  Embodiments of the concepts disclosed herein measure at least one characteristic of a population using identifiable photometric and / or morphological markers in the collection of multispectral images and collection of images from a population of cells. With respect to analysis of collected images to do, markers are associated with disease states. The term marker refers to an optical or spatial characteristic of a cell (or group of cells) that is determined using one or more images of the cell (or group of cells). In exemplary applications, the cells are obtained from body fluids and cell compartments, in particularly preferred implementations from blood, most preferably the cell compartments are bone marrow and lymph nodes, and more particularly preferred implementations are both photometric and morphological markers. Is used for analysis. Particularly preferably, although not limiting the implementation, a plurality of images of individual objects are respectively collected simultaneously.

  An exemplary step that can be used to analyze biological cells according to one embodiment of the concepts disclosed herein is to collect image data from a cell population and one or more cell sub-lines from the image data. Including identifying the population. In one embodiment, a subpopulation corresponding to cells exhibiting abnormalities associated with a disease state is identified. Such subpopulations can be identified based on experimental evidence indicating that one or more photometric and / or morphological features are generally associated with cellular abnormalities associated with a disease state. For example, photometric and / or morphological data from collected images is analyzed. Such data can be related to one or more characteristics of the cell. The term feature is intended to refer to a particular structure, region, feature, characteristic, or portion of a cell that can be easily distinguished from one or more images of the cell. Photometric and / or morphological data from the collected images is analyzed to allow measurement of at least one feature of the selected feature. Features that are experimentally associated with cellular abnormalities found in a particular disease state can be detected in the data to determine whether a particular disease state is found in the initially imaged cell population.

  In yet other embodiments, a disease state can be detected even if the cell itself does not exhibit an abnormality that can be identified by photometry and / or morphological parameters. In such implementations, the sample includes a plurality of different subpopulations each identified by normal characteristic morphology and photometric markers. In the absence of a disease state, the ratio of subpopulations to each other varies from patient to patient within a measurable range. When a malignant disease state is present, the disease state can change the ratio between subpopulations and can indicate that the disease state is present when the ratio change exceeds the normal range. For example, CLL (the disease state described in the background art above) changes the proportion of lymphocytes in the blood. Although the lymphocytes themselves do not show any abnormalities, a disease state is suggested by an increase in the number of lymphocytes that can be considered as a result of a normal response to infection or a malignant transformation event beyond the normal range.

  Consider the blood cell population of healthy patients. The ratio of lymphocytes to other types of blood cells can be determined by analyzing the image data of the entire population of blood cells to classify the images according to blood cell type. When the same process is applied to CLL patient blood cell populations, the ratio of lymphocytes to other types of blood cells is significantly different from the ratios identified in patients not suffering from CLL. Thus, a disease state can be detected by analyzing a cell population to identify subpopulations present in the population and measuring changes in the proportion of subpopulations that indicate the presence of the disease state.

  In yet other embodiments, the disease state can be detected by the presence of an uncharacterized cell type. In such embodiments, the sample includes a plurality of different subpopulations, each identified by a characteristic morphology and photometric marker. In the absence of a disease state, only the expected sub-population within the sample is evident, although it varies from patient to patient within a measurable range. If a disease state is present, a completely irregular cell type can be revealed in the sample. For example, breast cancer metastasis is evidenced by the presence of some level of apparent epithelial cells in the blood. Thus, a disease state can be detected by analyzing a cell population to identify subpopulations present in the population and measuring the prevalence of an irregular subpopulation that indicates the presence of the disease state. Disease states can be further clarified by analyzing the morphology of irregular cell populations and photometric markers to determine whether they contain characteristic subpopulations. For example, cancer can be characterized as aggressive when most cells divide at a fast rate, as evidenced by markers that define a high ratio of nuclear size to cell size.

  In yet another embodiment, detecting a disease state not only by analyzing cell subpopulations and their relative abundance, but also by analyzing biomolecules floating in the cell sample (not associated with cells) In such embodiments, a reagent can be added to the cell sample, and the reagent includes a reaction substrate that each indicates the abundance of a particular biomolecule. Each reaction substrate (eg, microsphere) includes a unique optical feature that identifies the type of biomolecule to which it preferentially binds and indicates the abundance of the biomolecule in the sample. By analyzing the sample image in which the reaction substrate and the cells are mixed, the reaction substrate and the cells can be distinguished, and analysis of molecules and cells in the sample can be performed in a complex manner.

  Population and subpopulation image data can be manipulated using several different techniques. An exemplary approach is called gating and is the manipulation of data relating to photometry or morphological imaging. Another exemplary approach is backgating, which further includes defining a subset of the gated data. Although not strictly necessary, signal processing is preferably performed on the collected image data, particularly image data collected using simultaneous multiple channel imaging, to reduce crosstalk and enhance spatial resolution.

  In particularly preferred embodiments, image data from a cell population indicative of a disease state is collected. One or more photometric or morphological markers associated with the disease state are identified. As described above, such markers can suggest measurable differences in some parameters between healthy cells and diseased cells. Such photometric or morphological markers used to distinguish healthy cells from diseased cells are generally associated with specific features. The identified marker can indicate data present in image data collected from diseased cells, but is likely not present in image data collected from healthy cells. Identified markers can also indicate data present in image data collected from cells exhibiting a disease state and are also present in image data collected from healthy cells, but are quantifiable and identifiable The degree is likely to be different. A marker can represent a measurable change in a subpopulation associated with a disease state, rather than a subpopulation associated with the absence of a disease state. Using the example described above, an increase in the number of blood lymphocytes relative to other blood cell types suggests a CLL disease state.

  After one or more identification markers are experimentally established, the cell population is imaged and analyzed to determine whether the identification marker is present in the sample population and to determine whether a disease state is present. be able to. Particularly preferably, although not limiting practice, the disease state is chronic lymphocytic leukemia and the marker is related to an increase in the size or shape of the lymphocyte subpopulation.

  To facilitate analysis, at least one aspect of the concepts disclosed herein includes labeling either diseased cells or healthy cells and imaging a mixed population of healthy and diseased cells together. The identification marker is measured from a mixed population of cells. Labeling allows a labeled subpopulation of cells to be extracted from the imaged data collected from the mixed population sample. Therefore, it is easy to separate aggregate image data into images corresponding to diseased cells and images corresponding to healthy cells by labeling, and photometric and / or morphological markers corresponding to diseased cells can be more easily identified. Is possible.

  Yet another embodiment of the approach disclosed herein relates to monitoring treatment of patients who are indicative of a disease state. Baseline data is collected by imaging the patient's cell population prior to treatment. Preferably, the cell population is obtained from a body fluid such as blood. Additional data is obtained by imaging additional cell populations collected from the patient during and after the various stages of the treatment process during treatment. Such data quantitatively suggests an improved state of patients in the disease state, as indicated by either the amount of cells representing the disease state versus normal cells, or a change in the proportion of subpopulations present in the population. To do. It is important that such quantification is not feasible with standard microscopy and / or conventional flow cytometry.

  In a preferred implementation of the techniques disclosed herein, images collected from a population of biological cells include a collection of multimodal images. That is, the collected images include at least two of the following types of images: It includes one or more images corresponding to light emitted from the cell, one or more images corresponding to light transmitted by the cell, and one or more images corresponding to light scattered by the cell It is. Such multi-mode imaging can include any or a combination of the following types of images: It comprises (1) one or more fluorescent images and at least one bright field image; (2) one or more fluorescent images and at least one dark field image; (3) one or more fluorescent images, bright A field image and a dark field image; and (4) a bright field image. Simultaneous collection of multiple different fluorescent images (separated by spectrum) can be beneficial, as is simultaneous collection of multiple different bright field images (using light transmitted with two different spectral filters). Preferably, multi-mode images are acquired simultaneously.

  Although the present disclosure has been presented in order to introduce some concepts in a simplified form, it is described in further detail in the following best mode for carrying out the invention. However, the disclosure of the present invention does not identify key or important features of the claimed subject matter, but is used as an aid in determining the scope of the claimed subject matter. not.

  Various aspects and attendant advantages of one or more exemplary embodiments, as well as modifications thereof, will be better understood by reference to the following detailed description when read in conjunction with the accompanying drawings and as follows: Will be readily apparent.

<Figures and disclosed embodiments are not limiting>
Exemplary embodiments are illustrated in referenced figures of the drawings. The embodiments and figures disclosed herein are to be considered illustrative rather than limiting.

<Overview>
The disclosure of the present invention combines the speed of flow cytometry, sample handling, cell sorting capabilities with multiple forms of microscopy images, sensitivity, and resolution as well as complete visible / near infrared spectral analysis, It includes a method of using a flow imaging system that can collect and analyze data related to a condition and in particular detect and monitor chronic lymphocytic leukemia.

  One embodiment of the concept disclosed herein relates to a system and method for imaging and analyzing biological cells entrained in a fluid flow. In at least one embodiment, multiple images of a plurality of living cells, ie, including at least two of the following types of images, are collected simultaneously. It is a bright field image, a dark field image, and a fluorescence image. Images are collected for a population of living cells. After the images are collected, the images can be processed to identify image sub-populations, which share light intensity and / or profile characteristics that are experimentally measured for association with disease states.

  The term “population of cell” with respect to the following disclosure and the claims attached thereto refers to a group of cells comprising a plurality of objects. Thus, the cell population must contain multiple cells.

A preferred image system used to collect the image data necessary to carry out the techniques disclosed herein incorporates the following principle characteristics:
1. High speed measurement;
2. Ability to process large or continuous samples;
3. High spectral resolution and bandwidth;
4). Good spatial resolution;
5. High sensitivity; and Low measurement variability.

  In particular, the recently developed imaging flow cytometer technology embodied in the ImageStream ™ instrument (Amnis Corporation, Seattle WA) has made significant progress in achieving each of the above-described principles of characteristics. The ImageStream ™ instrument is a commercial embodiment of a flow imaging system that will be described in detail below with respect to FIG. These significant advances in the technical field of flow cytometry are described in patents assigned to the assignee of the present invention (see, for example, US Pat. While the ImageStream ™ platform represents a particularly preferred imaging device for collecting image data that is processed according to the concepts disclosed herein, the concepts disclosed herein are specific to a particular device. It is not limited to use only.

  As described above, in addition to collecting image data from a biological cell population, one embodiment of the concepts disclosed herein includes at least one characteristic associated with a disease state in the imaged cell population. Processing the collected image data to measure. A preferred image analysis software package is IDEAS ™ (Amnis Corporation, Seattle WA). The IDEAS ™ package evaluates about 200 features in all cells, including multiple morphologies and fluorescence intensity measurements that can be used to define and characterize cell populations. The IDEAS ™ package allows the user to define biologically relevant subpopulations of cells and analyze the subpopulations using standard cytometric analysis such as gating and backgating. Is possible. However, other image analysis methods or software packages can be implemented to apply the concepts disclosed herein, and the preferred image analysis software package disclosed limits the concepts disclosed herein. Rather than illustrative.

<Overview of preferred imaging system>
FIG. 1 is a schematic diagram illustrating an exemplary flow imaging system that can be used to simultaneously collect multiple images from an object in flow. That is, in a schematic diagram of a preferred flow imaging system 510 (functionally describing the ImageStream ™ platform) that uses TDI when collecting images of an object 502 (such as a living cell) entrained in a fluid flow 504. is there.

  The flow imaging system 510 includes a velocity detection subsystem that is used to synchronize the flow of fluid through the TDI imaging detector 508 and the flow imaging system 510. Significantly, the imaging system 510 can collect multiple images of the object simultaneously. Also, a particularly preferred implementation of imaging system 510 is configured for multi-spectral imaging and can operate on the following six spectral channels. It is DAPI fluorescence (400-460 nm), dark field (460-500 nm), FITC fluorescence (500-560 nm), PE fluorescence (560-595 nm), bright field (595-650 nm), and deep red (650-700 nm) It is. The TDI detector 508 is capable of 10-bit digital resolution per pixel. The preferred imaging system 510 has a numerical aperture of typically 0.75 and a pixel size of about 0.5 microns. However, those skilled in the art will appreciate that this flow imaging system 510 is not limited to six spectral channels, nor is it limited to the numerical aperture or pixel size and resolution described above.

  The moving object 502 is illuminated using a light source 506. The light source 506 can be a laser, light emitting diode, filament lamp, gas discharge arc lamp or other suitable light source, and the flow imaging system 510 can provide broadband or one or more desired light wavelengths to the velocity of the object. And optical adjustment elements, such as lenses, apertures, and filters, used to illuminate the object with the intensity required to detect one or more other features. Light from the object is divided into two optical paths by a beam splitter 503. Light traveling along one optical path is directed toward the velocity detector subsystem, and light traveling along the other optical path is directed toward the TDI imaging detector 508. A plurality of lenses 507 are used to direct the light in the desired direction and focus the light. Although not shown, for example, only a light with a narrowband wavelength corresponding to the wavelength emitted by fluorescent or phosphorescent molecules in / on the object, or only light having a wavelength provided by the light source 506, and / or the velocity detection subsystem and / or Or, it can have one or a set of filters to reach the TDI imaging detector 508. As a result, light from unwanted light sources is substantially eliminated.

  The velocity detector subsystem includes an optical diffraction grating 505a that amplifies modulated light from an object, a light sensitive detector 505b (photomultiplier tube or solid state photodetector), a signal conditioning unit 505c, a velocity calculation unit 505d, and a system. A timing control unit 505e that reliably synchronizes the TDI imaging detector 508 to the fluid flow 504 through the. The optical diffraction grating preferably includes a plurality of alternating transparent and opaque bars that modulate light received from the object to form modulated light having a modulation frequency corresponding to the velocity of the object that sent the light. Preferably, the optical magnification and the marking pitch of the optical diffraction grating are selected so that the size of the object to be irradiated and the width of the bar are substantially the same. Thus, light collected from cells or other objects is alternately blocked and passed through the optical grating grating lines as the object traverses the interrogation region, i.e., the field of view. The modulated light is directed toward the light sensitive detector and generates a signal that can be analyzed by the processor to measure the velocity of the object. A timing signal is sent to the TDI imaging detector 508 using the velocity measurement subsystem.

  Preferably, the signal conditioning unit 505c includes a programmable computing device, although an ASIC chip or a digital oscilloscope can be used for this purpose. The frequency of the photodetector is measured and the velocity of the object is calculated as a function of that frequency. A speed dependent signal is periodically sent to the TDI detector timing control 505e to adjust the clock speed of the TDI imaging detector 508. Those skilled in the art will appreciate the speed of the image of the object passing through the TDI detector within a small tolerance range selected such that the vertical image smearing in the output signal of the TDI detector is within an acceptable range. The TDI detector clock speed is adjusted to match. The rate update rate must occur at a frequency sufficient to keep the clock frequency within the tolerance band as the flow (object) velocity changes.

  A beam splitter 503 is used to split the light portion from the object 502 to the light sensitive detector 505b and the light portion from the object 502a to the TDI imaging detector 508. In the optical path toward the TDI imaging detector 508, there are a plurality of stacked two-color filters 509 that separate light from the object 502a into a plurality of wavelengths. One of the lenses 507 is used to form an image of the object 502a on the TDI imaging detector 508.

  The theory of operation of a TDI detector as used in the flow imaging system 510 is as follows. As the object moves through the flow tube 511 (FIG. 1) and passes through the volume imaged by the TDI detector, the light from the object forms an image of the object, and these images move across the surface of the TDI detector. To do. The TDI detector preferably moves the charge in a row-by-row manner in each clock period so that a given charge train remains fixed or synchronized with a certain train in the image. A specially designed charge coupled device (CCD) array. As the charge column reaches the bottom layer of the array, it moves from the array to the memory. As the image and the corresponding generated signal propagate through the CCD array, the intensity of each row of signals generated by the TDI detector corresponding to the image of the object is integrated over time. This approach greatly improves the signal-to-noise ratio of the TDI detector compared to a non-integrating type detector. This is a significant advantage of detectors aimed at responding to images of low-level fluorescent radiation of objects. Proper operation of the TDI detector requires that the charge signal be clocked across the CCD array when synchronized with the speed of the image of the object moving through the CCD array. An accurate clock signal that facilitates this synchronization can be provided by measuring the velocity of the object, and in the concept disclosed herein, the velocity of the object, and therefore through the CCD array of the TDI detector. Use an accurate estimate of the speed of the image as it moves. This type of flow imaging system is disclosed in a patent assigned to the assignee of the present invention (see, for example, US Pat. No. 6,096,049), the complete disclosure, specification and drawings of which are hereby incorporated by reference. Specially incorporated.

  FIG. 2 is a diagram illustrating an image recorded (generated) by the flow imaging system illustrated in FIG. 1. It is a figure showing the image produced | generated by the flow imaging system of FIG. Column 520 is labeled “BF” and includes an image formed by a spherical object 502 entrained in fluid flow 504 absorbing light from light source 506. “BF” label refers to “bright field”, a term derived from a method of creating contrast in an image, whereby light passes through the region and light is absorbed by objects in the region. As a result, a dark region is formed in the image. Therefore, in this image, the background is light and the object is dark. Thus, column 520 is a “bright field channel”. The inclusion of bright field images is exemplary rather than limiting the scope of the concepts disclosed herein. Preferably, the concepts disclosed herein use a combination of bright field images and fluorescent images, or a combination of dark field images and fluorescent images.

  The remaining three columns 522, 524, 526 shown in FIG. 2 are labeled “λ1”, “λ2”, and “λ3”, respectively. These columns contain images generated using light emitted by objects entrained in the fluid flow. Preferably, such light is emitted through a fluorescence process (as opposed to an image generated using transmitted light). As will be apparent to those skilled in the art, fluorescence is the emission of light (or other electromagnetic radiation) by an object excited by absorption of incident light. In general, fluorescence lasts only while excitation radiation continues. Many substances (especially fluorescent dyes) can be identified based on the spectrum of light produced when they emit fluorescence. Thus, columns 522, 524, and 526 are referred to as “fluorescence channels”.

  Other exemplary flow imaging systems are disclosed in patents assigned to the assignee of the present invention (see, for example, US Pat. It is specifically incorporated herein by reference as background material. The imaging systems detailed above and in these two patents and incorporated by reference herein have significant advantages over conventional imaging systems used to acquire images of biological cell populations. These advantages are obtained by using several imaging systems of the optical dispersion system in conjunction with a TDI detector that produces an output signal in response to images of cells and other objects in the direction of the TDI detector. It is done. It is important to be able to collect multiple images of a single object simultaneously. The image of each object can be resolved into a spectrum using a common TDI detector for analysis to distinguish object features by absorption, scattering, reflection or radiation. Other flow imaging systems include multiple detectors, each dedicated to a single spectral channel.

  Use these flow imaging systems to measure the morphology, photometry and spectral properties of cells and other objects by measuring optical signals including light scattering, reflection, absorption, fluorescence, phosphorescence, luminescence, etc. Can do. Morphological parameters include area, perimeter, texture, or spatial frequency, centroid location, shape (ie, circular, oval, barbell, etc.), volume, and the ratio of a selected pair (or subset) of these parameters. Contains. Similar parameters can be used to measure the nucleus, cytoplasm, or other subcompartments of cells according to the concepts disclosed herein. Photometric measurements in a preferred imaging system can measure the optical density of the nucleus, the optical density of the cytoplasm, the optical density of the background, and the ratio of selected pairs of these values. An object that is imaged by the concepts disclosed herein can be excited to fluorescence or phosphorescence to emit light, or can be luminescent light that generates light without excitation. . In each case, the light from the object is imaged on a TDI detector using the concepts disclosed herein to determine the presence and amplitude of the emitted light, the cell or other that produces the optical signal. The number of discrete positions in the object, the relative position of the signal source, and the color (wavelength or band) of the light emitted at each position of the object are measured.

<Analysis of disease state of body fluid using imaging system>
As noted above, embodiments of the concepts disclosed herein may include a plurality of biological cell populations for identifying at least one photometric or morphological feature that has been experimentally determined to be associated with a disease state. Includes both spectral image collection and analysis of collected images. Thus, one disclosed embodiment of the present invention is derived from a multi-modal image of an object (eg, a cell) in the flow to identify the characteristics of the cells in the cell population and facilitate detection of the presence of the disease state. Related to the use of both photometric and morphological features. Comprehensive to provide morphological and photometric data to allow quantization of features presented by both normal and diseased cells, making it easier to detect diseased or abnormal cells that are indicative of a disease state A method for analyzing cells in suspension or flow that can be combined with such multispectral imaging is described in detail below. To date, such methods have not been possible with standard microscopy and / or flow cytometry.

As described above, a preferred flow imaging system (eg, ImageStream ™ platform) is used to simultaneously acquire multispectral images of cells in the flow, supporting four channels of brightfield, darkfield, and fluorescence. Image data can be collected. The ImageStream ™ platform is a commercial embodiment based on the imaging system described above. In general, cells are hydrodynamically focused into the core stream and illuminated at right angles for dark field and fluorescent images. Cells are simultaneously transmitted and illuminated by spectrally limited light sources (eg, filtered white light or light emitting diodes) for bright field imaging. Light is collected from the cells with an imaging objective and projected onto a CCD array. The optical system has a numerical aperture of 0.75, an object space CCD pixel size of 0.5 microns ² and allows high resolution imaging at an event rate of approximately 100 cells per second. In this example, each pixel is digitized with 10-bit intensity resolution, providing a minimum dynamic range of 30 per pixel. In practice, the distribution of the signal across multiple pixels generally provides an effective dynamic range of over 40 per image. Furthermore, the sensitivity of the CCD can be controlled independently for each multispectral image, resulting in a total of about 60 dynamic ranges across all images associated with one object. Although the ImageStream (TM) platform represents a particularly preferred flow imaging system for acquiring image data in accordance with the concepts disclosed herein, the ImageStream (TM) platform does not limit the disclosed concepts and is illustrative Represents a simple imaging system. An imaging instrument that can collect images of a biological cell population sufficient to allow image analysis, as detailed below, can be implemented according to the concepts described herein.

  Referring back to the preferred imaging system, the ImageStream ™ platform, the light passes through a spectrally resolved optical system that directs different spectral bands to different lateral positions of the detector (that Such spectral decomposition is described above in conjunction with the description of various preferred embodiments of the imaging system). Using this technique, the image is optically resolved into a set of multiple sub-images (preferably six sub-images including bright field, dark field and four different fluorescent images), each sub-image being It corresponds to different spectral (ie color) components and is spatially separated from the remaining sub-images. This process facilitates identification and quantization of intracellular signals by physically separating detector signals that may arise from overlapping portions of the cells. Spectral decomposition also allows multi-mode imaging, ie simultaneous detection of bright field, dark field and multiple colors of fluorescence. The process of spectral decomposition is performed during the image creation process rather than by conventional digital image processing of composite images.

  The CCD can be operated using TDI to maintain sensitivity and image quality even when the relative motion between the detector and the object being imaged is fast. As with any CCD, image photons are converted to photocharges in an array of pixels. However, in TDI operation, photocharge moves continuously from pixel to pixel down the detector and parallel to the flow axis. When the moving speed of the photo charge is synchronized with the speed of the cell image, the effect is similar to physically panning the camera. Despite the signal integration time being on the order of magnitude longer than conventional flow cytometry, image streaking is avoided. For example, the instrument operates at a continuous data rate of about 30 megapixels per second, can integrate a 10 millisecond signal from each object, and detect even a few fluorescent probes in acquired cell images at a relatively high speed. It is possible. By paying attention to the pump and fluid system design to achieve high laminar non-pulsatile flow, cell rotation or lateral movement on the time scale of the imaging process is eliminated (see, for example, US Pat. ).

  The real-time algorithm analyzes all pixels read from the CCD to detect the presence of an object image and calculate the number of basic features and photometric features that can be used as a basis for data storage . Data files containing 10,000 to 20,000 cells are typically about 100 MB in size and can therefore be stored and analyzed using a standard personal computer. The TDI readout process runs continuously without “dead time”, which means that all cells can be imaged, and two or more cells can be imaged simultaneously in contact or non-contact This indicates that there is no obstacle to data acquisition.

  Using such an imaging system to measure the morphology, photometry and spectral characteristics of cells and other objects by measuring optical signals including light scattering, reflection, absorption, fluorescence, phosphorescence, luminescence, etc. Can do. As used herein, morphological parameters (ie, morphometry) can be either basic (eg, nuclear shape) or complex (eg, cytoplasmic as the difference between cell size and nucleus size). Size). For example, morphological parameters include core area, perimeter, texture or spatial frequency component, centroid position, shape (ie, circular, elliptical, barbell shape, etc.), volume, and ratio of selected pairs of these morphological parameters. be able to. Cell morphology parameters can also include cytoplasm size, texture or spatial frequency components, volume, and the like. As used herein, photometric measurements in the imaging system described above measure the optical density of the nucleus, the optical density of the cytoplasm, the optical density of the background, and the ratio of selected pairs of these values. can do. The object to be imaged can be excited to fluorescence or phosphorescence to emit light, or luminescent to generate light without excitation, in each case the light from the object is the TDI detector of the imaging system The presence and amplitude of the emitted light, the number of discrete positions in the cell or other object generating the light signal, the relative position of the signal source, and the color (wavelength or band) of the light emitted at each position of the object ) Can be measured.

  The present disclosure provides a method that uses both photometric features and morphology obtained from a multi-mode image of an object in flow. Such a method can be used as a cell analyzer for measuring whether a marker corresponding to a disease state is present in the cell population to be imaged. As noted above, the marker can indicate a cell abnormality associated with the disease state, or the marker can indicate a change in the proportion of subpopulations present in the cell population being imaged. Suggests a disease state. Preferably, the cell population is imaged while being entrained in the fluid flowing through the imaging system. As used herein, gating refers to a subset of data associated with photometry or morphological imaging. For example, a gate can be a numerical or geometric boundary of a subset of data that can be used to further define the characteristics of the particle to be analyzed. Here, a gate is a plot that includes, for example, “in focus” cells, or sperm cells with tails, or sperm cells without tails, or cells other than sperm cells, or sperm cell populations or cell debris. It is defined as a boundary. Further, backgating can be a subset of subset data. For example, a forward scatter vs. side scatter plot combined with histograms from other markers can be used to backgate a subset of cells within the initial subset of cells.

  When using the imaging system described herein, if the object is luminescent (ie, the object generates light), no other light source is required to create an image of the object (cell) It will be clear. However, many imaging system applications described herein may need to use one or more light sources to provide incident light to the object being imaged. One skilled in the art will appreciate that the position of the light source substantially affects the interaction of the type of information obtained from the image using the incident light of the object and the detector.

  In addition to illuminating the object with the incident light and imaging it, the light source can also be used to excite the emission of light from the object. For example, a cell in contact with a probe bound to a fluorescent dye (FITC, PE, APC, Cy3, Cy5, or Cy5.5) fluoresces when excited by light, and the corresponding characteristic emission spectrum is excited. Generated from the fluorescent dye probe and can be imaged on a TDI detector. Alternatively, a light source is used to excite a fluorophore probe on the object, and the TDI detector detects the fluorescent spot created by the probe as a result of spectral dispersion of light from the object provided by the prism. Enable imaging at different locations above. The placement of these fluorescent spots on the surface of the TDI detector depends on the emission spectrum and its position on the object.

  Each light source can generate light that can be either coherent, non-coherent, broadband, or narrowband light, depending on the desired imaging system application. Thus, tungsten filament light sources can be used for applications where a narrow band light source is not required. Narrowband laser light is preferred for applications such as exciting fluorescent emission from a probe, which may be a spectrally resolved, undistorted image of an object generated from light scattered by the object. To make it possible. This scattered light image is resolved separately from the fluorescent spot generated on the TDI detector as long as the emission spectrum of any spot is at a wavelength different from the wavelength of the laser light. The light source can be either a continuous wave (CW) or a pulse type such as a pulsed laser. When using a pulse type illumination source, it is possible to integrate signals from multiple pulses by extending the integration period associated with TDI detection. Furthermore, it is not necessary to pulse the light in synchronism with the TDI detector.

  In particular, for use in collecting image data of cell populations found in body fluids such as blood, a 360 nm UV laser is used as the light source and the optical system of the imaging system is diffracted into a 400 to 460 nm (DAPI radiation) spectral band. It may be desirable to optimize for limited image performance.

  It should be understood that in an embodiment consistent with the disclosure herein, there is relative motion between the object being imaged and the imaging system. In most cases, it is more convenient to move the object than to move the imaging system. In some cases, it is contemplated to keep the object stationary and move the imaging system relative to the object. As a further alternative, both the imaging system and the object can be moved and their motion can be in different directions and / or different velocities.

<Example High-Level Method Steps>
FIG. 3 shows a flowchart of overall method steps performed in one implementation of the concepts disclosed herein. That is, flowchart 400 schematically illustrates exemplary steps that can be used to analyze a cell population based on an image of the cell population to identify a disease state.

  In particularly preferred embodiments, the cell population is obtained from a bodily fluid such as blood. First, at block 402, image data is collected from a first population of biological cells known to have a disease state using an imaging system, such as the exemplary imaging system described above. At block 404, at least one photometric or morphological marker associated with the disease state is identified. In the experimental tests described below, two individually different types of markers were developed. One type of marker relates to photometry and / or identification of morphological differences between healthy and diseased cells. One technique for identifying such markers is to label cancer cells with a fluorescent label and compare images of the fluorescently labeled cancer cells with images of healthy cells to compare multiple photometric and morphological features associated with cancer cells. It is to identify the marker. As described in detail below, such markers include differences in the average nucleus size between healthy and cancer cells, and differences in fluorescence images between healthy and cancer cells. These differences are quantified based on the processing of the image data of the cell population so that images that are likely to be images of cancer cells are identified and images that are likely to be healthy cells are identified. be able to.

  Another type of marker relates to the identification of some differences between subpopulations found in cell populations without disease state and subpopulations found in cell populations in disease state. For example, CLL is a disease state in which the number of lymphocytes in the blood is increased compared to the number of other types of blood cells. Thus, a disease state can be suggested by a change in the ratio of the number of lymphocytes and other types of blood cells.

  Once photometric and / or morphological markers associated with a disease state are identified, image data is collected from a second population of cells where the disease state is present or not known to be absent. At block 408, image data is collected from the second cell population, and the image data is then analyzed for the presence of the previously identified marker to determine whether a disease state is present in the second cell population.

  When collecting image data from a cell population using the imaging system described above, it is important that the image data can be collected very quickly. In general, analysis (ie, analysis of collected image data to initially identify a marker or determine the presence of a previously identified marker in a cell population) is offline, ie after collection of image data. To be implemented. Current imaging processing software can analyze relatively large cell populations (ie, tens of thousands of cells) within tens of minutes using a readily available personal computer. However, as more powerful computing systems are developed and readily available, it will be possible to analyze image data in real time. Accordingly, off-line processing of image data is exemplary rather than limiting, and it is contemplated that real-time processing of image data is an alternative.

  If the marker relates to some photometric and / or morphological difference between healthy and diseased cells, especially if the first population contains a mixture of both diseased and healthy cells, the imaging device is used to Prior to collecting image data of a population (a population known to be associated with a disease state), it may be desirable to label either diseased cells or healthy cells. This method facilitates the separation of the collected image data into an image corresponding to diseased cells and an image corresponding to healthy cells, and can be used to distinguish between the two photometric and / or morphological markers To facilitate identification. However, the first population can contain only disease cells, and when comparing image data of a cell population known to contain only healthy cells and image data of the first population, the disease cells are treated as healthy cells. Photometric and / or morphological markers can be easily identified that can be used to distinguish from.

  If the marker relates to some photometric and / or morphological differences between a subpopulation found in a cell population without a disease state and a subpopulation found in a cell population associated with the disease state, the first cell population (disease Image data corresponding to a state known to be related to a condition) is analyzed to determine a subpopulation found in a healthy cell population and a cell population having a disease state (cellular population having the disease condition). Image data corresponding to the subpopulation found in the healthy cell population must be provided before some photometric and / or morphological differences from the subpopulation found in the disease cell population can be identified .

  Although not strictly necessary, in the working embodiments of the technology described herein, other processing has been performed to reduce multi-channel imaging crosstalk and spatial resolution. The implemented crosstalk reduction process is described in a patent assigned to the assignee of the present invention (see, for example, Patent Document 8), and is hereby incorporated by reference in its specification, disclosure and drawings as background material. Specifically incorporated herein by reference. One skilled in the art can alternatively implement other types of crosstalk reduction methods.

<Example Photometric / Identification of Morphological Disease State Markers>
In the context of the present disclosure, the multi-spectral imaging flow cytometer described above is UV-excited to quantify DNA content and nuclear morphology for the purpose of detecting and monitoring disease states such as chronic lymphocytic leukemia. Use performance and algorithms. In addition to using flow imaging equipment that includes 360 nm UV lasers and optical systems optimized for imaging performance that are diffraction limited to the 400-460 nm (DAPI radiation) spectral band, imaging processing systems are used to process image data. The Image processing software running on a personal computer represents an exemplary imaging processing system. The imaging processing software incorporates algorithms that allow cell photometry and / or determination of morphological characteristics based on cell images. Exemplary algorithms include masking algorithms, algorithms that define nuclear morphology, algorithms for cell cycle quantification histograms, algorithms for analyzing DNA content, algorithms for analyzing non-uniform chromaticity, N Includes algorithms for analyzing the / C ratio, algorithms for analyzing granularity, algorithms for analyzing CD45 expression, and algorithms for analyzing other parameters. In addition, the imaging processing software incorporates an algorithm called a classifier, which is an analytical tool based on software configured to evaluate a sample population of cells to determine whether a disease state marker is present. In determining the presence of cancer cells, the classifier analyzes the image of the sample population for images having photometric and / or morphological characteristics corresponding to the photometric and / or morphological characteristics associated with the previously identified cancer cells. .

  For samples of cell populations that are analyzed to detect CLL, separate the images into different cell subpopulations (based on different types of blood cells) and whether the subpopulation ratios indicate the presence of CLL The image of the sample population is analyzed by the classifier so as to determine (for example, the number of lymphocytes is larger than usual). Preferably, the classification device configured to detect CLL separates blood cells into the following subpopulations. It is lymphocytes, monocytes, basophils, neutrophils, and eosinophils. A classifier configured to detect CLL is based on experimental data from healthy and CLL patients. The CLL profile of the classifier can be improved by collecting and comparing classifier data for various patients who have received the same diagnosis. Preferably, for optimization of the classifier, a large (10,000-20,000 cells) data set is collected from each patient to assess the presence and diagnostic significance of a subpopulation of CLL cells. . Such an optimized classification device can be used to monitor a patient's treatment response and assess residual disease after treatment.

  It is important that high-speed data acquisition is necessary for the detection of epithelial cell carcinoma. Such cells have been reported to be one in 100,000 to 1,000,000 peripheral blood lymphocytes (peripheral blood leukocytes). The ImageStream ™ cytometer and IDEAS ™ analysis software package described above are ideally suited for this application. Images from peripheral blood lymphocytes can generally be obtained in the preparation of blood smears that are free of artifacts. A large number of cells (tens of thousands to hundreds of thousands) can be accumulated per sample, and analysis of subpopulations can provide high confidence. Immunofluorescent staining with recognized markers (CD5, CDI9, etc.) makes correlation with morphology easier. A quantitative cell sorter eliminates the subjectivity of human assessment and provides a confidence level that was previously unobtainable for comparisons between patients. Longitudinal studies are also very beneficial by quantitative analysis, especially for storing and retrieving large numbers of cell image files digitally, especially when compared to conventional methods of searching digital photographs of microscope slides and / or relatively few cells. be able to.

<Identification of morphological features using fluorescence method>
The technique used to detect cancer cells in body fluids based on image data of cell populations from body fluids has been the development of initial absorbance and fluorescence staining protocols for simultaneous morphological analysis of bright field and fluorescence images.

  First, in the test, the simultaneous use of dye staining and fluorescent dye was considered. The imaging system described above has the ability to generate bright field images as well as multiple color fluorescent images of each cell, so the possibility of using both conventional dye staining and fluorescent dyes for analysis simultaneously was considered. However, because dye staining does not normally penetrate the cell membrane of viable cells, and the optical system described above can collect side scatter images of the laser, it is related to cell granularity previously obtained with dyes such as eosin. Much information could be obtained using laser side scatter images without the need for cell staining. Numerous cell-penetrating fluorescent dyes provide nuclear morphology without the need for fixation and dye staining. Based on these considerations, it became clear that fluorescence alternatives to distinguishing morphological features provide better methods than conventional staining methods.

  The main alternative to fluorescent dye staining useful in conjunction with the optical system described above is a fluorescent DNA binding dye. Such various dyes can be excited at 488 nm and include several SYTO dyes (Molecular Probes), DRAQ5 (BioStatus), 7-AAD, Propium Iodide (PI), and the like. These dyes are alternatives to nuclear dyes such as Toluidine Blue, Methyl Green, Crystal Violet, Nuclear Fast Red, Caralum, Celestine Blue, and Hematoxylin. Fluorescent DNA binding dyes are generally used in the optical systems described above to define the shape and boundaries of the nucleus, its region, its texture (non-uniform chromaticity similarity), and to provide DNA content information Included in the analysis protocol developed for.

  The software image analysis program IDEAS ™ described above makes it possible to evaluate feature combinations from different images of the same cell in order to expand the use of fluorescent nuclear images. For example, as a means for generating a mask that includes only the cytoplasmic region, the nuclear image mask can be removed from the bright field image mask (which covers the entire cell). Once defined, the cytoplasmic mask can be used to calculate the cytoplasmic region, N / C ratio, relative fluorescence intensity of cytoplasmic and nuclear probes, etc. with an intuitive “Feature Manager”. An example of a Feature Manager session for defining the N / C ratio is shown in FIG. FIG. 4 is a diagram illustrating an exemplary graphical user interface used to perform the method steps shown in FIG. The basic features associated with cell images are selected from a list and combined algebraically using a simple expression builder.

<Measurement of photometric parameters and / or morphological parameters>
In an exemplary embodiment of the concepts disclosed herein, ImageStream ™ data analysis and cell sorting is performed after acquisition using the IDEAS ™ software package. A captured image of the analysis of human peripheral blood on the annotated IDEAS ™ software screen is shown in FIG. IDEAS ™ software allows visualization and photometric / morphological analysis of data files containing images from tens of thousands of cells, thereby combining quantitative image analysis and statistical power of flow cytometry.

  FIG. 5 is a diagram illustrating an exemplary graphical user interface used to perform the method steps of FIG. 3 applied to the analysis of human peripheral blood. The exemplary screenshot shown in FIG. 5 includes images and quantitative data from 20,000 human peripheral blood mononuclear cells. Whole blood was treated with erythrocyte lysing agent and cells were labeled with anti-CD45-PerCP mAb (red) and DNA binding dye (green). The FL1 and FL4 spectral bands were used to image each cell in fluorescence and dark and bright fields. In this figure, images of a plurality of cells in the dark field channel 51a, the green fluorescence channel 51b, the bright field channel 51c, and the red fluorescence channel 51d can be easily identified. Such a thumbnail image gallery (upper left of the interface) allows a “list mode” examination of the cell population. Cell images show four different cell types (eosinophil 53a, NK cell 53b, monocyte 53c and neutrophil 53d) pseudo-stained and overlaid for visualization in the image gallery or shown below the interface Can be expanded.

  The software also allows one-dimensional and two-dimensional plots of features calculated from the images. A dot 55 representing a cell in the two-dimensional plot can be “clicked” to see the associated image in the gallery. The reverse is also possible. The cell image can be selected to highlight the corresponding dot in all plots where that cell is displayed. In addition, a “virtual cell sort” function can be used to define subpopulations that can be gated on the plot and then examined in the gallery. Features calculated from the image or defined by the user (i.e. selected from a list automatically combined using algebraic and simple expression builders) can be plotted. Dot plot 59a (shown on the left side of the center of FIG. 5) is an analysis of CD45 expression (x-axis) versus dark-field granularity measurement (y-axis) similar to the side scatter intensity measured by conventional flow cytometry. The cluster obtained from is shown. Plot 59a represents lymphocytes (green in the full color image), monocytes (red in the full color image), neutrophils (turquoise in the full color image), and eosinophils (orange in the full color image). The dot plot 59b (displayed on the right side of the center of FIG. 5) substitutes for a nuclear texture parameter which is the “nuclear frequency” of CD45 expression on the x-axis, and is estimated NEC cell population ( In a full-color image, purple). Background display of the purple population in the left dot plot indicates that this population has the same mean value of CD45 expression as the lymphocyte population (green in the full color image). This frequency parameter is one of a set of morphological and photometric features that have been developed and incorporated into the IDEAS ™ software package. Table 1 below shows an exemplary list of photometric and morphological definitions that can be identified for the entire image (or subpopulation if appropriate). It should be understood that FIG. 5 has been modified to simplify its duplication. In a full-color image, in order to make cells and data easy to see, the background of each frame including cells is black, and the background of each dot plot is black.

  The features that quantify the morphology are shown in italics in Table 1. Each feature is automatically calculated for six types of images of each cell (dark field, bright field, and four fluorescent images acquired simultaneously) when the image data set is loaded into the software.

  Over 35 features are calculated for each image, and an analysis using all 6 images results in a total of over 200 features per cell without user-defined features. Each cell is also assigned a unique serial number and time stamp, allowing dynamic examination of the cell population.

<Selecting photometric / morphological markers for cancer cells>
It was first proposed to use bladder epithelial cells to examine morphological differences between normal and epithelial cancer cells. However, the initial samples of bladder lavage that were analyzed showed that the number of cells per sample was very variable and was generally too few for practical use on the ImageStream ™ instrument. Therefore, mammary epithelial cells were used instead of bladder cells. Breast cells were selected because normal primary cells of this kind are commercially available (Clonetics / Invitrogen) and grow as adherent cells in short-term tissue culture in a dedicated growth medium. In addition, breast epithelial cancer cells taken from breast cancer patient metastases are available from the American Type Tissue Culture Collection (ATCC). To better manage variability from tumor to tumor, three different breast epithelial cancer cell lines were tested. They are HCC-I 500, HCC-1 569, and HCC-1428. These cell lines were established from metastasis of three separate patients and were purchased as frozen stocks from ATCC. The cell line was fixed on plastic and grown, propagated by a normal tissue culture method, and used experimentally.

  Normal and cancerous breast epithelial cells were harvested separately by short-term culture with trypsin / EDTA at 37 ° C. Cells were washed once with cold phosphate buffer (PBS) containing 1% FCS, counted and used experimentally. Three separate breast epithelial cancer cell lines were pooled in equal proportions for the experiments described below.

  Normal mammary epithelial cells were stained with a fluorescein-conjugated monoclonal antibody of HLA class I MHC cell surface protein by culturing the cells with appropriate predetermined mAb dilutions at 4 ° C. for 30 minutes. Breast cancer is known to down-regulate class I MHC expression, but as a preventive measure normal cells were fixed with 1% paraformaldehyde to limit passive changes to cancer cells. Bound breast cancer cell lines were also fixed with 1% paraformaldehyde and added to the normal breast cell population. Before operating the ImageStream ™ instrument, DRAQ5 (BioStatus. Ltd, Leicestershire, UK), a DNA binding dye that can be excited with a 488 nm laser and emits in the red band, was added to the sample. Labeling normal breast epithelial cells with anti-class I MHC mAbs can identify normal cells in a mixture of normal and cancer cells, thereby providing an objective “truth” that It is easier to identify image features that are distinguished from epithelial cancer cells.

  Normal peripheral blood was obtained from AllCells (San Diego, Calif.). Whole blood was cultured with FITC-conjugated anti-CD45 mAb, which is expressed at a level in all peripheral white blood cells. The red blood cells were then lysed by culturing whole blood with a Becton Dickinson FACSLyse ™ for 3 minutes at room temperature. Cells were washed with PBS, counted and fixed with 1% paraformaldehyde. Breast epithelial cancer cells were generated as described above, fixed with 1% paraformaldehyde, and added to peripheral blood cells. DRAQ5 was then added as a nuclear dye and the cells were analyzed on an ImageStream ™ instrument.

  An image file containing image data of the cell mixture described above (usually a mixture of breast epithelial cells and breast cancer cells, and a mixture of normal peripheral blood cells and breast cancer cells) was analyzed using the IDEAS ™ software package and Results were obtained.

  After performing spectral compensation on the data file, an initial visual inspection was performed comparing normal breast epithelial cells (anti-HLA-FITC positive) with cancer cells (anti-HLA-FITC unstained). FIG. 6 shows an image of normal (ie, healthy) breast epithelial cells, and FIG. 7 shows how quantification of fluorescence channel data functions as a marker of disease state ( That is, it is a figure which shows the image of a disease) cell. In each figure, each horizontal line contains four simultaneously acquired images of a single cell.

  The images in columns 61a and 71a correspond to the side scatter images (ie dark field images) of the blue laser, the images in columns 61b and 71b correspond to the green HLA-FITC fluorescence images, and in columns 61c and 71c. The images correspond to bright field images, and the images in columns 61d and 71d correspond to red nuclear fluorescence images. As mentioned above, a preferred imaging system can collect six different types of images of a single cell (dark field image, bright field image, and four fluorescence images) simultaneously, and in FIGS. Two of the fluorescent channels are not used. 6 and 7 have been modified to simplify their duplication. In a full-color image, the background in FIGS. 6 and 7 is black, the images in columns 61a and 71a are blue, the images in columns 61b and 71b are green, and the images in columns 61c and 71c are gray on a gray background. It is a scale image and the images in columns 61d and 71d are red.

  When the full-color images of FIGS. 6 and 7 are compared visually, the image of normal breast epithelial cells (green fluorescent channel) in column 61c in FIG. 6 is clear, and the image of cancer cells in column 71c (green) in FIG. It will be readily apparent that the fluorescent channels are almost indistinguishable. None of the dark field images (columns 61a and 71a) are particularly strong, and the dark field image (column 61a) of the normal breast epithelial cells in FIG. 6 is significantly more than the dark field image (column 71a) of the cancer cells in FIG. It will also be clear that the strength is high. Yet another routine observation that can be easily made is that the average intensity of the red fluorescence image (column 71d) of the cancer cells of FIG. 7 is much higher than the average intensity of the red fluorescence image of FIG. 6 (column 61a). Other qualitative observations show that normal cells are highly heterogeneous, generally larger and less intense in the nucleus. Subsequent analysis calls for quantifying these differences and finding other parameters that may be discriminating. A screen capture of the corresponding IDEAS ™ analysis is shown in FIG.

  FIG. 8 shows a number of different photometric and morphological descriptors that can be used to automatically distinguish an image of healthy breast epithelial cells from an image of breast cancer cells, and the method steps shown in FIG. FIG. 3 illustrates an example graphical user interface used to execute.

  The analysis shown in FIG. 8 proceeded from the dot plot 81 in the upper left of the figure. Single cells were first identified based on dot plot 81, defined as bright field area to aspect ratio. A gate (not shown separately) around a population containing single cells estimated based on the criteria that the area is large enough to exclude debris and the aspect ratio is greater than -0.5. Set and exclude cell doublets and clusters. The accuracy of gating was tested by examining random cells inside and outside the gate using the click-on-dot visualization function.

  Normal breast cells were then differentiated from breast cancer cells using an anti-HLA-FITC marker that was applied only to normal cells. A solid yellow histogram 85a of the FITC intensity is generated and shown on the right side of the dot plot 81. A gate 83 was set around FITC positive (normal breast epithelial cells) and FITC negative (breast epithelial cancer cells), resulting in a subpopulation of 2031 normal cells and 611 cancer cells. These subpopulations were then used to identify features that quantitatively distinguish between normal and cancer cells based on differential histograms. FIG. 8 has been modified to simplify its duplication. In a full-color image, in order to make cells and data easy to see, the background of each frame including cells is black, and the background of each dot plot and histogram is black. With this modification, a uniform distribution of dots 81a was obtained, but such a uniform distribution was not found in the full-color image.

  The remaining 10 histograms (ie, histograms 85b to 85k) in FIG. 8 are differential histograms of normal cells 87a (displayed in green in a full color image) and cancer cells 87b (displayed in red in a full color image). Different quantitative characteristics are shown. The ten identification features are grouped into the following five distinct classes: It is scattering intensity, scattering texture, morphology, nuclear intensity, and nuclear texture. The differential histograms 85b, 85c, and 85d show the difference between the two populations using the following three scattered intensity features that are different but correlated. It is the “scattered average intensity” (intensity sum divided by cell area), “scatter intensity” (intensity sum minus background), and “scatter spot subtotal” (partial maximum intensity sum). All three scattered intensity features showed good discrimination, but the “scattered average intensity” (histogram 85b) was the most selective.

  Derivative histograms 85e and 85f can be obtained using either an intensity profile gradient measure ("scatter slope RMS"; histogram 85e) or a more selective pixel intensity variability ("scatter frequency"; histogram 85f). Was quantified.

  The differential histograms 85g, 85h and 85i plotted cell area (bright field, histogram 85g), nuclear area (from DNA fluorescence image, histogram 85h), and cytoplasmic area (cell / nuclear area, histogram 85i). Cancer cell lines were generally small in bright field, confirming qualitative observation from cell images. The nuclear area of cancer cell lines was proportionally smaller than that of normal cells (eg, the nucleus / cell area ratio was not discriminatory), but the cytoplasmic area was significantly smaller in cancer cells.

  Finally, differential histograms 85j and 85k plotted nuclear mean intensity (histogram 85j) and nuclear frequency (non-uniform chromaticity, histogram 85k), respectively. As with scattering, both of these features had some discriminatory power.

  Multispectral / multimodal images collected by the ImageStream ™ instrument in this engineering experiment and analyzed using the IDEAS ™ software package enable normal epithelial cells in dark field scatter, morphology, and nuclear staining to A number of significant differences were shown between epithelial cancer cells. Although it is well understood that cells adapted for tissue culture have undergone a selection process that may have altered their cellular characteristics, these data are based on the identified morphological and photometric characteristics identified above. Has been shown to provide an automated sorting device that separates normal cells from transformed epithelial cells and possibly other types of cells.

  In other experimental studies, image data collected from a mixture of normal peripheral blood cells and breast cancer cells was analyzed. As shown in FIG. 5, human peripheral blood cell classification uses a flow imaging system configured to acquire multiple images of each cell simultaneously, and an automatic image analysis program (an exemplary imaging system is ImageStream ™ instrument, an exemplary image analysis program is an IDEAS ™ software package). Using expression of CD45 combined with analysis of dark field light scattering properties, cells can be separated into the following five individual populations based on the image data collected by the flow imaging system. It is lymphocytes, monocytes, neutrophils, eosinophils, and basophils. This separation of human peripheral blood into individual subpopulations is illustrated in greater detail in FIG. 9 and includes exemplary relative abundance data for various subpopulations.

  FIG. 9 is a diagram schematically illustrating the separation of human peripheral blood cells into various subpopulations based on photometric characteristics. Classification accuracy was determined using population specific monoclonal antibody markers, and backgating marker positive cells versus CD45 plots in scatter, and morphological analysis of related images. The x-axis of the graph shown in FIG. 9 corresponds to anti-CD45-FITC intensity and the y-axis corresponds to dark field scattering intensity.

  The technique disclosed herein (use of the above-described flow imaging instrument system exemplified by the ImageStream ™ instrument and the image analysis software exemplified by the IDEAS ™ software package) converts epithelial cancer cells into normal PBMC In order to determine whether they can be distinguished from each other, an artificial mixture of tumor cells and normal PBMC was prepared as described above. The cell mixture was labeled with anti-CD45-FITC mAb and fluorescent DNA binding dye to distinguish PBMC subpopulations as generally described above. A comparison of bivariate plots of scatter versus CD45 on PBMC samples showing spikes of normal peripheral blood mononuclear cells and cancer cells is shown in FIGS. 10A and 10B.

  FIG. 10A is a graph showing the distribution of normal peripheral blood mononuclear cells (PBMC) based on image data collected from a cell population, and does not include breast cancer cells. FIG. 10B is a graphical representation of the distribution of normal PBMC and breast cancer cells based on image data collected from cell populations containing both cell types, and how the distribution of breast cancer cells is distinguished from the distribution of normal PBMC cells. It shows what is possible. In this analysis, the cancer cells 101a are well outside the normally defined PBMC population 101b, as confirmed by visual inspection of the outlier population.

  FIG. 11A shows how the cytoplasmic area distribution of breast cancer cells is distinguishable from that of normal PBMC cells and was obtained from image data collected from a cell population containing both cell types. FIG. 11B schematically shows the distribution of normal PBMC and breast cancer cells based on the measurement of cytoplasmic area, and FIG. 11B shows how the distribution of the scattering frequency of breast cancer cells can be distinguished from the distribution of the scattering frequency of normal PBMC cells. FIG. 4 schematically shows the distribution of normal PBMC and breast cancer cells based on measurements of scatter frequency obtained from image data collected from cell populations containing both cell types.

  Cancer cell 111a can also be distinguished from normal PBMC 111b using some of the morphological and photometric features identified in FIG. 8 (eg, nuclear area, cytoplasmic area, scattering intensity, and scattering frequency). it can. FIG. 11A graphically shows the distribution of normal PBMC and breast cancer cells based on measurements of cytoplasmic area obtained from image data collected from cell populations containing both cell types, with normal distribution of cytoplasmic area of breast cancer cells. It shows how it can be distinguished from the distribution of cytoplasmic area of PBMC cells. FIG. 11B graphically illustrates the distribution of normal PBMC and breast cancer cells based on the measurement of scatter frequency obtained from image data collected from cell populations containing both cell types, with normal distribution of scatter frequency of breast cancer cells. It shows how it can be distinguished from the scattering frequency distribution of PBMC cells. These features were initially identified to distinguish between normal breast cells and breast cancer cells, but showed a high level of discrimination between breast epithelial cancer cells and PBMC. It is important that normal epithelial cells are more clearly distinguished from PBMC and are distinguished from epithelial cancer cells using these parameters.

  10A, 10B, 11A and 11B have been modified to simplify replication. In a full-color image, the background of each frame containing dot plots is black to make cells and / or data easier to see, and dots representing PBMC cells and cancer cells are represented in different colors.

  The results described above result in individual images (ie, images separated into subpopulations corresponding to cancer cells and healthy cells using one or more of the above identified photometric and / or morphological parameters). This was confirmed by visual inspection. Image gallery data was generated from the PBMC data showing the spikes described above.

  FIG. 12 is a diagram showing a composite image of cells generated by combining bright field and fluorescence images of breast cancer cells. Includes representative images from a population of cancer cells obtained using overlay synthesis of bright field and DRAQ5 DNA fluorescence (red with image analysis software). FIG. 13 is a diagram showing representative images of five different PBMC populations that can be defined by scatter data obtained from image data of cell populations. It includes images of five peripheral blood mononuclear cell populations defined using dark field scatter, CD45 (green), and DRAQ5 (red) for nuclear morphology. Note that the two figures have different size scales. It should be understood that FIG. 12 has been modified to simplify its duplication. In the full-color image, the background of FIG. 12 is black, each frame containing cells is brown / grey, and the nucleus of each cell is red. FIG. 13 is similarly modified to simplify its duplication. In the full-color image, the background in FIG. 13 is black, the outer periphery of each cell is green, and the nucleus of each cell is red.

  The test described above optically transforms a subpopulation of normal epithelial cells from a subpopulation of transformed cells by analyzing multispectral / multimodal image data from a mixed population of cells where image data is collected simultaneously. It is important to demonstrate the possibilities that can be identified. The tests described above also demonstrate the possibility of detecting epithelial cancer cells in the blood by analyzing multispectral / multimodal image data from a mixed population of cells where image data is collected simultaneously. is doing.

  Due to the relatively high operating speed of the exemplary imaging system (about 100 cells / second or about 350,000 cells / hour) with respect to applying the concepts described herein to the concept of a particular disease state, And because the amount of image information collected in each cell is relatively large (high resolution bright field image, dark field image, and four fluorescence images), the concept disclosed herein is for detecting chronic lymphocytic leukemia. And particularly suitable for monitoring.

  In such applications, a 360 nm UV laser is incorporated into a simultaneous multispectral / multimodal imaging system, and the optics of the imaging system are optimized for imaging performance that is diffraction limited to the 400-460 nm (DAPI radiation) spectral band. The exemplary imaging system (ie, ImageStream ™ instrument) used in the experimental tests described above uses a solid state, 200 mW, 488 nm laser for fluorescence excitation. Such laser wavelengths excite a broad range of fluorescent dyes, but DAPI cannot be excited and are not optimal for cell cycle analysis that binds stoichiometrically to DNA. In addition, the beam is configured to have a narrow width, improving the overall sensitivity at the expense of increased measurement error per cell. A feasibility study using propidium iodide as a DNA dye shows that an imaging system using a 488 nm laser can generate a cell cycle histogram with a G0 / G1 peak coefficient of variation of about 5%.

  To generate high resolution cell cycle histograms to detect changes in DNA content associated with CLL, DAPI optimized for 360 nm UV laser is used instead. The beam is configured to have a relatively wide illumination intersection (about 100 microns) and under typical operating conditions, DAPI excitation consistency is within 1% between cells. Overall, the cell cycle histogram CV is considered to be about 2-3%. Furthermore, the optics of the exemplary instrument used in the experimental tests described above are diffraction limited to 460-750 nm and do not cover the DAPI spectral emission band. Thus, in order to measure detailed nuclear features of diagnostic values such as morphological cavities and heterogeneous dyes, such optics are configured to achieve diffraction limited imaging performance in the 400-460 nm spectral band. Replaced by optics.

  In particular, in order to use the concepts disclosed herein to detect changes in DNA content associated with CLL, other masking algorithms and features that define the morphology of normal sample nuclei other than those described above may be used. It is desirable to provide embedded image processing software.

  The set of morphological features available in the exemplary image processing software described above does not include features of boundary contours that quantify the nuclear lovicity, the number of invaders, and similar parameters. Such a feature is very useful in the analysis of leukemia because it captures the qualitative observation of many nuclear forms used by conventional hematopathologists. Incorporation of such algorithms and features allows for improved automatic classification of normal cells, precursors, and transformed cells.

  The boundary contour masking algorithm and associated features used in the experimental tests described above are eosinophils, neutrophils, monocytes in approximately one third of each type of cell as a function of orientation relative to the imaging plane. Improve cell classification of basophils, and lymphocytes. Cells in one of the two preferred directions (in 6 possible directions) do not benefit from the algorithms and features previously used. In order to improve cell classification, by increasing the statistical resolution within the position where the distribution of normal cells is expected, it leads to a first pass classification device between normal cells and transformed cells. The boundary contour algorithm and features can be expanded to consistently classify normal white blood cells regardless, thereby improving the ability to flag abnormal cells that do not fall into the expected location. Such a sorter can characterize the morphological differences observed between normal and transformed lymphocytes to further improve discrimination, generally using the techniques described above. .

  In order to configure the imaging analysis software to detect changes in DNA content associated with CLL, an automatic classifier is incorporated into the software package. The automatic classifier incorporates at least one or more of the following photometric and / or morphological parameters. It is DNA content, nuclear morphology, heterogeneous pigment, N / C ratio, granularity, CD45 expression, and other parameters. As described above, the classification device is configured to classify the image data corresponding to the blood cell population, and classifies the population into the following subpopulations. It is lymphocytes, monocytes, basophils, neutrophils, and eosinophils.

  Automatic differential analysis of PBMC based on multimodal images collected simultaneously from cells in the flow is performed using an imaging system similar to that described above and imaging processing software similar to that described above. PBMC are stained with FITC-conjugated anti-CD45 and a DNA-binding dye, DAPI. Peripheral blood leukocytes are generally classified by 5-part differential analysis into lymphocytes, monocytes, basophils, neutrophils, and eosinophils, as shown in FIGS. .

  Data sets from peripheral blood leukocytes from CLL patients are acquired and analyzed as described above. A taxonomy developed for normal peripheral blood leukocytes is applied to these datasets to determine CLL cell identification by comparing it to a normal profile. Various classifiers are evaluated and generally determine the segment in which CLL cells are best exemplified as described above with respect to the histogram of FIG. These include the following: It is cell size, nucleus size, nucleus-to-cytoplasm ratio, nucleus contour, nucleus texture, and cytoplasm granularity. The results are compared to a standard blood smear of a CLL patient sample to determine the accuracy of the technique.

  In addition to the normal staining protocol using anti-CD-45 as a marker, peripheral blood leukocytes are stained with monoclonal antibodies to CD5 and CD20 and DAPI prior to collecting image data. This method allows the identification of CLL cells according to accepted flow cytometry criteria. In this way, morphological criteria can be correlated with the immune phenotype.

  With a large (10,000-20,000 leukocytes) data set from multiple CLL patients, blood cells can be subpopulated (ie lymphocytes, monocytes, basophils, neutrophils, and eosinophils) ) Can be easily optimized and selected for photometric and morphological markers that can be used for classification.

  Morphological heterogeneity is observed in CLL cells. However, since it is technically difficult to consistently prepare and evaluate a peripheral blood smear of a patient, an accurate and objective understanding of the extent has not been achieved so far. By acquiring large data sets from CLL patients using the multi-mode imaging system described above, the degree of morphological heterogeneity can be objectively analyzed using an imaging processing software package become. The developed classifiers described above are applied to these datasets to evaluate the morphological heterogeneity assessed by analyzing the degree of specific classifiers (eg, nuclear size, N / C ratio, etc.). Apply to large populations of CLL cells. Based on this analysis, the classifier that most accurately identifies the maximum percentage of CLL cells is optimized so that the entire population is included by the classifier.

  As described above, when applied to CLL, the techniques disclosed herein are not used to separate cell populations into subpopulations corresponding to healthy cells and subpopulations corresponding to diseased cells. Instead, image data collected from blood cell populations separate blood cell populations into subpopulations based on blood cell types (ie, lymphocytes, monocytes, basophils, neutrophils, and eosinophils). Used to do. Since CLL is associated with an increase in the number of lymphocytes found in the blood cell population (ie, an increase in lymphocyte subpopulation), detecting an increase in lymphocytes indicates the presence of a disease state (ie, CLL). Although the preferred method described herein involves separating the blood cell population into a plurality of different subpopulations, the CLL detection method simply separates the blood cell population into lymphocyte and non-lymphocyte subpopulations. It can also be executed. Using experimental data showing the average lymphocyte subpopulation in healthy patients, detecting a lymphocyte subpopulation higher than average suggests a CLL disease state.

  In addition to initially detecting CLL disease status, the imaging and analysis methods described above are applied to longitudinally track CLL patients to determine response to treatment, clinical response stability, and disease recurrence. can do. Changes in the peripheral blood population, including both normal and residual CLL, can be followed and correlated with clinical outcomes.

<Example computing environment>
As described above, one embodiment of the present invention includes image analysis of multiple images collected simultaneously from a portion of a cell population. Reference is made to an exemplary image analysis software package. FIG. 14 and the associated description below provide a simplified and general description of a suitable computing environment for practicing the present invention, and the required image processing is generally similar to that shown in FIG. Implemented using a computing device. For those skilled in the art, the necessary image processing includes laptops and other types of portable computers, multiprocessor systems, network computers, mainframe computers, handheld computers, personal data assistants (PDAs), and processor and machine instructions. Can be implemented by many different types of computing devices, including other types of computing devices that perform multiple functions when executed on a processor.

  An exemplary computing system 150 suitable for performing the image processing required by the present invention includes an input device 152 and an output device 162, eg, a processing unit 154 that is functionally connected to a display. The processing unit 154 performs the functions of the present invention (analysis of a plurality of images acquired simultaneously from a group of objects to enable presentation of at least one feature by the group to be measured), image processing / A central processing unit (CPU 158) for executing machine instructions including an image analysis program is included. In at least one embodiment, machine instructions generally perform functions similar to those described above, with reference to the flowchart shown in FIG. Those skilled in the art will find suitable processors or central processing units (CPUs) for this purpose available from Intel Corporation, AMD Corporation, Motorola Corporation, and other suppliers.

  The processing unit 154 also includes random access memory 156 (RAM) and non-volatile memory 160, typically including read only memory (ROM), and some form of memory storage such as hard drives, optical drives, and the like. These memory devices are bidirectionally connected to the CPU 158. Such storage devices are well known to those skilled in the art. Machine instructions and data are temporarily loaded from non-volatile memory 160 into RAM 156. The memory also stores operating system software and auxiliary software. Although not shown separately, a power source is required to supply the power necessary to start the computing system 150.

  Input device 152 may be any device or mechanism that facilitates input to the operating environment, including but not limited to a mouse, keyboard, microphone, modem, pointing device, or other input device. Although not specifically shown in FIG. 14, the computing system 150 is logically connected to an imaging system such as that schematically shown in FIG. 1 to collect the collected images to perform the desired image processing. Data can be used in the calculation system 150. Of course, rather than logically connecting the computing system directly to the imaging system, the data collected by the imaging system can be collected using many different data transfer devices such as portable memory media or network (wired or wireless) It can be simply transferred to the computing system via The output device 162 most commonly includes a monitor or computer display designed to allow a person to visually recognize the output image.

  While the concepts disclosed herein have been described in conjunction with preferred embodiments and modifications thereof for practicing the same, those skilled in the art will recognize within the scope of the appended claims. Many other modifications can be made to the concept disclosed in. Accordingly, the scope of the concepts disclosed herein is not limited in any way by the above description, but is to be determined entirely by reference to the appended claims. The invention claiming exclusive rights is defined as follows.

1 is a schematic diagram illustrating an exemplary flow imaging system that can be used to simultaneously collect multiple images from an object in flow. FIG. It is a figure showing the image recorded (generated) by the flow imaging system shown in FIG. FIG. 5 shows a flowchart of overall method steps performed in one implementation of the concepts disclosed herein. FIG. 4 illustrates an exemplary graphical user interface used to perform the method steps shown in FIG. FIG. 4 illustrates an exemplary graphical user interface used to perform the method steps of FIG. 3 applied to the analysis of human peripheral blood. It is a figure which shows the image of a normal (namely, healthy) breast epithelial cell. FIG. 6 shows an image of breast cancer (ie, disease) cells showing how quantification of fluorescence channel data functions as a marker of disease state. FIG. 4 shows a plurality of different photometric and morphological descriptors that can be used to automatically distinguish an image of healthy breast epithelial cells from an image of breast cancer cells to perform the method steps shown in FIG. FIG. 3 illustrates an exemplary graphical user interface used. FIG. 3 schematically illustrates the separation of human peripheral blood cells into various subpopulations based on photometric characteristics. It is a figure which shows typically distribution of the mononuclear cell (PBMC) of normal peripheral blood based on the image data collected from the cell population which does not contain a breast cancer cell. Shows how the distribution of breast cancer cells is distinguishable from that of normal PBMC cells, and schematically illustrates the distribution of normal PBMC and breast cancer cells based on image data collected from cell populations containing both cell types. FIG. Measurement of cytoplasmic area obtained from image data collected from cell populations containing both cell types, showing how the distribution of cytoplasmic area of breast cancer cells is distinguishable from that of normal PBMC cells FIG. 6 is a diagram schematically showing the distribution of normal PBMC and breast cancer cells based on FIG. Shows how the distribution of scatter frequencies of breast cancer cells is distinguishable from that of normal PBMC cells and measures scatter frequencies obtained from image data collected from cell populations containing both cell types FIG. 6 is a diagram schematically showing the distribution of normal PBMC and breast cancer cells based on FIG. It is a figure which shows the synthetic | combination image of the cell produced | generated by combining the bright field and fluorescence image of a breast cancer cell. FIG. 4 shows representative images of five different PBMC populations that can be defined by scatter data obtained from image data of cell populations. FIG. 4 schematically illustrates an example computing system used to perform the method steps shown in FIG. 3.

Claims (1)

A disease state detection method for detecting a disease state by an image collected from a cell population,
(A) imaging the cell population to collect image data such that multiple images of each cell are collected simultaneously;
The plurality of images are the following two types of images:
(I) a multi-spectral image;
(Ii) including at least one of a multi-mode image;
(B) analyzing the collected image data to determine whether a subpopulation of cells suggesting the disease state is detected. A disease state detection method comprising:

Images