Method for diagnosing and analyzing cells

Abstract

To provide a more automatable, consistent and comprehensive method for providing a diagnosis of the state of cells present in a patient sample.SOLUTION: The detection of pathological abnormalities in a tissue sample and / or a plurality of cells is a highly specialized and time-consuming task, usually performed by a group of selected clinicians and technicians. Described herein are methods for more automatable, consistent and comprehensive cell sample analysis to provide rapid and reliable detailed classification, for example, but not limited to cancer diagnosis, particularly diagnosis of the state of the cells present in the sample.SELECTED DRAWING: Figure 1A

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority to U.S. Provisional Application No. 63 / 051,117, filed July 13, 2020, the entire contents of which are incorporated herein by reference. Incorporation by Reference

[0002] All publications and patent applications mentioned in this specification are herein incorporated by reference in their entirety to the same extent as if each individual publication or patent application was specifically and individually indicated to be incorporated by reference.

[0002]

[0003] The methods described herein relate to detecting and determining the state of cells and their environment within a sample, particularly in the field of cancer diagnostics, using optical imaging methods to enhance sample collection, improve diagnostic accuracy and speed, and reduce costs. [Background technology]

[0003]

[0004] Detecting pathological abnormalities in tissue samples and / or multiple cells is a highly specialized and time-consuming task typically performed by a select group of clinicians and technicians. High costs are incurred due to human resources and the time delay between procuring a biopsy and providing a diagnosis to the patient and / or the physician responsible for determining next steps. Additional costs are also incurred if timely treatment is delayed. Furthermore, the analyses performed rely to some extent on subjective observation and interpretation and may therefore be irreproducible across practitioners. Therefore, it would be highly useful to provide a more automatable, consistent, and comprehensive method for analyzing patient-derived samples to provide a rapid, reliable, detailed classification of the cellular status present in a patient sample, particularly, but not limited to, cancer diagnosis. Summary of the Invention

[0004]

[0005] Described herein are methods that provide a more automatable, consistent, and comprehensive method for analyzing patient-derived samples to provide a rapid, reliable, and detailed classification, particularly, but not limited to, cancer diagnosis, of the state of cells present in the patient sample. Analytical methods that combine data from both space-dependent and time-dependent interferometric images can significantly improve the accuracy of the assignment of the state of the cells under examination.

[0005]

[0006] The method described herein also has the potential to transform how cancer is diagnosed and treated. This is the direct or indirect goal of rapid on-site evaluation (ROSE), frozen sectioning, confocal laser endoscopy (CLE), ultrasound, and other imaging modalities, but none of these diagnostic approaches have the potential to nondestructively identify, in real time, key indicators of tumor presence, tumor type, potential tumor response to specific treatments, or potential recurrence after surgery. The data obtained from the images obtained with the method described herein includes real-time data, detailed images, dynamic cellular information, and viable tissue for further analysis, all of which reveal a wealth of valuable clinical, biological, and structural information unavailable from any other single method. Furthermore, due to its nondestructive nature, it can also be used in parallel with other methods, such as CLE, ultrasound, and traditional tissue processing methods. Finally, its nondestructive nature opens up a wealth of potential downstream applications not achievable with current methods, such as frozen sectioning, which, by their very nature, alter the tissue sample and may add sampling bias by not providing complete information about the sample's surface.

[0006]

[0007] Thus, in a first aspect, there is provided a method for determining the state of a plurality of cells. The method includes the steps of acquiring time-dependent interferometric images and space-dependent interferometric images of a plurality of cells suspected of containing cancer cells; submitting the time-dependent interferometric images and space-dependent interferometric images to a multi-layer algorithmic analysis, thereby combining data associated with each pixel of the images of each of the plurality of cells; and automatically assigning a state to at least one cell of the plurality of cells, wherein the state is selected from a normal cell state or a cancer cell state.

[0007]

[0008] In some variations, the method may further include training the multi-layer algorithmic analysis by analyzing a portion of the data from the time-dependent interferogram and / or a portion of the data from the space-dependent interferogram. In some variations, the time-dependent interferogram and the space-dependent interferogram may be spatially registered.

[0008]

[0009] In some variations, the method may further include submitting the image of the plurality of cells further comprising the detectable label to a multi-layer algorithmic analysis. In some variations, the method may further include distinguishing structural features of the plurality of cells and reducing interference in a time-dependent interference image of the plurality of cells.

[0009]

[0010] In some variations, the multi-layer algorithmic analysis may include a pre-trained convolutional neural network.

[0010]

[0011] In some variations, the method may further include automatically assigning a state to the subset of cells, whereby the region in which the subset of cells is located is annotated as normal or cancerous. In some variations, the subset of structural features from the space-dependent interferometric image is submitted to an artificial intelligence analysis, e.g., a deep learning algorithm, which can assign a state to the region in which the subset of cells is located.

[0011]

[0012] In another aspect, a method is provided for performing a biopsy on a subject requiring a biopsy, the method including the steps of imaging a region of tissue to identify a region of interest, inserting a biopsy needle into the region of interest, excising a first tissue sample from the region of interest, acquiring a set of time-dependent and space-dependent interference images of the first tissue sample, and determining the number of target cells present in the first tissue sample. In some variations, the steps of imaging the region of tissue, inserting the biopsy needle, excising the first tissue sample, acquiring a set of time-dependent and space-dependent interference images, and determining the number of target cells present can be performed in a biopsy procedure room.

[0012]

[0013] In some variations, obtaining the set of time-dependent interference images and space-dependent images may further include processing the images to obtain images of intracellular metabolic activity of a plurality of cells within the first tissue sample.

[0013]

[0014] In some variations, the method may further include assigning a state to one or more cells of the plurality of cells, wherein the one or more cells having the assigned state are target cells. In some variations, the assigned state may be a disease cell state. In some variations, the disease cell state may be a cancer cell state.

[0014]

[0015] In some variations, determining the number of cells of interest includes submitting the image of intracellular metabolic activity to processing by a multi-layer algorithm to thereby assign a state to one or more cells. In some variations, assigning a state to one or more cells includes preselecting the level of metabolic activity observed in the one or more cells. The method may include comparing the detected value with a predetermined threshold value.

[0015]

[0016] In some variations, the method may further include obtaining a second tissue sample from the area of ​​interest if the number of cells of interest in the first tissue sample is insufficient for analysis.

[0016]

[0017] In some variations, imaging the region of tissue may include contrast-enhanced optical imaging, label-free optical imaging, radioactive imaging, ultrasound imaging, or magnetic imaging. In some variations, inserting the biopsy needle may include guided insertion.

[0017]

[0018] In another aspect, a method for determining a state of a plurality of cells is provided, the method comprising the steps of acquiring images of intracellular metabolic activity of a plurality of cells suspected of containing cancer cells, the images comprising time-dependent interference images; automatically assigning a state to at least a subset of the plurality of cells, the state being selected from a normal cell state or a cancer cell state; and assigning a cancer stage state to the subset of the plurality of cells.

[0018]

[0019] In some variations, the method may further include obtaining spatially dependent interference images of the plurality of cells, distinguishing structural features of the plurality of cells, and reducing interference in the images of intracellular metabolic activity of the plurality of cells.

[0019]

[0020] In some variations, automatically assigning a state to the subset of the plurality of cells includes submitting an image of intracellular metabolic activity of the plurality of cells to a deep learning algorithm, whereby the level of metabolic activity observed in the subset of the plurality of cells is compared to a preselected threshold. In some variations, if the level of metabolic activity exceeds the preselected threshold, the cells of the subset of cells may be assigned a cancerous state.

[0020]

[0021] In some variations, the method may further include automatically assigning a state to the subset of cells, whereby the region in which the subset of cells is located is annotated as normal or cancerous. In some variations, a subset of structural features from the spatially dependent interferometric image is submitted to a deep learning algorithm, which can assign a state to the region in which the subset of cells is located.

[0021]

[0022] In some variations, assigning a cancer stage state to a subset of the plurality of cells may include at least one of determining a level of differentiation of the plurality of cells, determining a level of cellular organization of the plurality of cells, determining the presence of biomarkers, and determining the cancerous / non-cancerous status of other plurality of cells obtained from the same subject. In some variations, the cancer stage is a cancer stage.

[0022]

[0023] In some variations, the time-dependent interferogram may include a set of images taken over a period of about 1 second to about 5 seconds, hi some variations, the time-dependent interferogram may include a set of images taken at a rate of about 50 fps to about 500 fps.

[0023]

[0024] In another aspect, a method is provided for determining the effect of a molecule and / or biological agent on a cell, the method comprising the steps of acquiring images of intracellular metabolic activity of a plurality of cells, the plurality of cells including at least one diseased cell and at least one non-diseased cell, the images comprising time-dependent interference images; contacting the plurality of cells with a molecule and / or biological agent and subsequently acquiring a plurality of images over a period of time, the plurality of images comprising the intracellular metabolic activity of the plurality of cells; and determining the effect of the molecule and / or biological agent on the at least one diseased cell compared to the effect on the at least one non-diseased cell. and / or determining the effect of the biological agent.

[0024]

[0025] In some variations, the method may further include obtaining spatially dependent interference images of the plurality of cells, distinguishing structural features of the plurality of cells, and reducing interference in the images of intracellular metabolic activity of the plurality of cells.

[0025]

[0026] In some variations, the step of acquiring multiple images over a subsequent period of time may be carried out for about 1 hour to about 3 days after contacting the multiple cells with the molecule and / or biological agent.

[0026]

[0027] In some variations, determining the effect of the molecule and / or biological agent on at least one diseased cell compared to the effect on at least one non-diseased cell may include determining the level of metabolic activity for the at least one diseased cell and the at least one non-diseased cell over a subsequent period of time.

[0027]

[0028] In some variations, the level of metabolic activity in diseased cells may be increased compared to the level of metabolic activity in non-diseased cells, hi some variations, the level of metabolic activity in diseased cells may be decreased compared to the level of metabolic activity in non-diseased cells.

[0028]

[0029] In some variations, the level of metabolic activity may remain the same as for non-diseased cells.

[0029]

[0030] In some variations, the methods may further include identifying off-target activity of the molecule and / or biological agent on non-diseased cells.

[0030]

[0031] In some variations, the molecule may comprise a biomolecule or organic molecule. In some variations, the biomolecule may comprise a protein, nucleic acid, sugar, or cellular expression product. In some variations, the organic molecule may comprise an organic compound having a molecular weight of less than about 2000 Da. In some variations, the biological agent may be a virus, phage, bacterium, or fungus.

[0031]

[0032] In another aspect, a method is provided for detecting cancer cells in a set of interference images, including a space-dependent interference image and at least one time-dependent interference image of a region of tissue comprising a plurality of cells, the method comprising: performing an analysis of the set of images, the method comprising: automatically defining regions in the pair of images representing cellular boundaries of the plurality of cells; automatically defining regions in the pair of images representing intracellular regions of the plurality of cells; automatically comparing intensities of pixels in the time-dependent interference image of the intracellular region of a selected cell of the plurality of cells with intensities of pixels in regions adjacent to the selected cell; automatically assigning a state label to the pixels of the intracellular region of the selected cell, consisting of underactive, normally active, or overactive, and summing the plurality of state labels in the intracellular region of the selected cell, thereby defining the cell as healthy or overactive; and defining each overactive cell of the plurality of cells as cancerous.

[0032]

[0033] In some variations, the analyzing step may be performed by a multi-layer algorithmic analysis. In some variations, the pair of interferograms may be spatially registered.

[0033]

[0034] In some variations, the method may further include determining from the spatially dependent interference image that a subset of the plurality of cells represents a cell type that is not of interest, thereby excluding the subset of the plurality of cells from further analysis.

[0034]

[0035] The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

[0035]

[0036] The novel features of the invention are set forth with particularity in the following claims. A better understanding of the features and advantages of the present invention will be obtained by reference to the following detailed description that sets forth illustrative embodiments, in which the principles of the invention are utilized, and the accompanying drawings of which: [Brief explanation of the drawings]

[0036] [Figure 1A]

[0037] FIG. 1 is a schematic diagram of a method for determining the state of a plurality of cells, according to some embodiments of the present disclosure. [Figure 1B]

[0038] FIG. 1 is a schematic diagram of a method for determining the state of a plurality of cells, according to some embodiments of the present disclosure. [Figure 2]

[0039] FIG. 1 is a schematic diagram of an imaging system for use in methods according to some embodiments of the present disclosure. [Figure 3A]

[0040] 1A-1C are photographic representations of spatially dependent interference images of tissue samples, according to some embodiments of the present disclosure. [Figure 3B] 1A-1C are photographic representations of spatially dependent interference images of tissue samples, according to some embodiments of the present disclosure. [Figure 4A]

[0041] 1A-1C are photographic representations of time-dependent interferograms in accordance with some embodiments of the present disclosure. [Figure 4B] 1A-1C are photographic representations of time-dependent interferograms in accordance with some embodiments of the present disclosure. [Figure 4C]

[0042] 1A-1C are photographic representations of time-dependent interferograms in accordance with some embodiments of the present disclosure. [Figure 4D] 1A-1C are photographic representations of time-dependent interferograms in accordance with some embodiments of the present disclosure. [Figure 4E]

[0043] FIG. 4B shows a photographic representation of a selected region from FIGS. 4B and 4D and a graphical representation of metabolic quotient coding of elements of the selected region. [Figure 5]

[0044] 1A-1C are diagrams of photographic representations of time-dependent images according to some embodiments of the present disclosure. [Figure 6A]

[0045] 1 is a schematic diagram of currently available biopsy procedures. [Figure 6B]

[0046] 1 is a schematic illustration of a biopsy procedure according to some embodiments of the present disclosure. [Figure 7]

[0047] FIG. 1 is a schematic diagram of a cell classification method. [Figure 8]

[0048] FIG. 1 is a diagram of a workflow for using the non-negative matrix factorization (NMF) algorithm on DCI images of breast tissue. [Figure 9A]

[0049] FIG. 1 shows the results of NMF analysis for baseline signal, fibers, noise, cells, and motion artifacts. [Figure 9B] FIG. 1 shows the results of NMF analysis for baseline signal, fibers, noise, cells, and motion artifacts. [Figure 9C] FIG. 1 shows the results of NMF analysis for baseline signal, fibers, noise, cells, and motion artifacts. [Figure 9D] FIG. 1 shows the results of NMF analysis for baseline signal, fibers, noise, cells, and motion artifacts. [Figure 9E] FIG. 1 shows the results of NMF analysis for baseline signal, fibers, noise, cells, and motion artifacts. [Figure 10]

[0050] 1 is a diagram of a workflow for streamlining classification and segmentation to obtain reliable classification along with course segmentation of DCI breast specimens. [Figure 11]

[0051] FIG. 1 shows some example filters of the deepest convolutional layer of one approach, and an example input using the GradCAM method. DETAILED DESCRIPTION OF THE INVENTION

[0037]

[0052] As used herein, the "sensitivity" of an assay is a measure of how often the assay correctly produces a positive test result for a sample having the condition being tested, e.g., the "true positive rate." A sensitive assay does not often produce false negatives.

[0038]

[0053] As used herein, the "specificity" of an assay is a measure of how often the assay successfully produces negative test results for samples that do not have the condition being tested, e.g., the "true negative rate." A highly specific assay will not often produce false positive results.

[0039]

[0054] Detecting pathological abnormalities in tissue samples and / or multiple cells is a highly specialized and time-consuming task typically performed by a select group of clinicians and technicians. Costs stemming from human resources and the time delay between procuring a biopsy and providing a diagnosis to the patient and / or the physician responsible for determining next steps are high. Furthermore, the analysis performed may be irreproducible by the individual performing it, relying to some extent on subjective observation and interpretation. Furthermore, using currently used standard biopsy procedures, the majority of biopsies are returned to the clinician as inconclusive, e.g., pending due to insufficient sample for analysis. A lack of available sample leads to reordering the biopsy procedure, often resulting in a delay of at least one week and incurring additional costs for the biopsy team and the patient. Therefore, it would be highly useful to provide a more automatable, consistent, and comprehensive method for analyzing patient-derived samples to provide a rapid, reliable, and detailed classification, particularly for, but not limited to, cancer diagnosis, of the state of the cells present in the patient sample.

[0040]

[0055] Spatially dependent interferometry for optical tomography, such as full-field optical coherence tomography (FF-OCT), which uses different reference and object arm lengths in an interferometric imaging device, has been reported to be useful for cancer diagnosis by revealing structural differences between cancerous and normal tissues, an example of which is shown in Figure 3A and 3B. However, images can be dominated by strong backscattering signals from highly transparent cells, such as collagen fiber networks or myelinated axons in brain tissue, which mask weak backscattering signals from highly transparent cells. Therefore, additional imaging techniques and methods for processing the resulting data are needed to provide improved cellular imaging that can be deployed in tissue biopsy and diagnosis, treatment guidance, and the development of new therapies.

[0041]

[0056] Applicant has discovered that new image processing methods are possible by using an interferometric imaging system configured to variably employ a selected optical distance between the reference arm and the sample arm (e.g., either a different optical distance or the same optical distance). The resulting images can be "cut" through tissue while reducing and / or eliminating interfering backscattered signals, e.g., "virtually" scanned at different depths within the tissue sample. Importantly, as described above, imaging techniques that involve more than just spatially dependent interferometric imaging can be used to obtain more highly resolved and contrasty images of cells. By analyzing and displaying statistical parameters associated with the time-series characteristics of each image pixel over several seconds (e.g., over a period ranging from about 1 second to about 5 seconds, at an acquisition rate of about 100 frames per second (fps) to about 500 fps), images acquired when the lengths of the reference arm and object arm are equal (e.g., time-dependent interferometric imaging such as dynamic cellular full-field optical coherence tomography (DC-FFOCT), also known as dynamic cellular imaging (DCI)), more detailed information can be analyzed. These improved images can show cellular details more clearly, including more information about the presence of other cells, such as various subtypes of immune cells, cell size and shape, cellular details such as nuclei and cytoplasm, and the proportion of cancer cells. Details of the intracellular metabolism of cells can be obtained, and a metabolic index (MI) can be calculated from the data, allowing for the precise localization of cells exhibiting metabolic activity indicative of cancer, as shown in Figures 4A and 4B.

[0042]

[0057] Furthermore, Applicant has discovered that using an analytical method that combines data from both spatially and time-dependent interferometric images significantly improves the accuracy of assigning the state of cells under examination, resulting in a level of accuracy nearly equivalent to or even greater than that of human expert diagnosis based on standard hemolysin and eosin (H&E) staining. This improvement in accuracy frees human experts from more complex analyses, reducing costs and improving the effectiveness and reproducibility of rapid on-site evaluation (ROSE) during biopsy procedures, which is at least partially attributable to the large intra- and inter-observer discrepancies between cytopathologists evaluating ROSE. Furthermore, this method allows ROSE to be performed in settings where a cytopathologist is not available, thereby enabling clinicians to provide better care to patients regardless of geographic location.

[0043]

[0058] These techniques can also be used for treatment guidance after a patient's first therapeutic intervention, such as after a first round of chemotherapy, and to determine the patient's post-treatment status (including cancer staging). These methods can also be used to evaluate potential therapies by determining the effect a putative molecular or biological agent may have on multiple cells. The methods described herein can further be used to determine the appropriate dosage of a therapeutic to be delivered to a patient based on patient-specific samples or during the therapeutic development process to assist in obtaining regulatory approval.

[0044]

[0059] The ability to combine information about (1) structural cellular details, such as morphology and phenotype, (2) metabolic activity, such as cells exhibiting higher than normal metabolic activity that can be classified as cancerous or otherwise diseased, and (3) spatial relationships for each identifiable cell type or phenotype can be powerfully used to diagnose disease states, identify appropriate therapeutic interventions, and evaluate novel therapeutics and treatments. For example, biopsy samples can be analyzed to determine whether immune cells, such as T lymphocytes (T cells), dendritic cells (CD4), and natural killer cells (NK cells), are present within a cell sample in the same region as diseased cells, such as cancer cells. For example, determining that T cells are recruited near diseased cells can provide guidance regarding the success of current therapeutic interventions or the development of potential therapeutics if they can enhance T cell responses.

[0045]

[0060] The image acquisition, processing, and analysis methods described herein can be used to replace the standard method of endoscopic biopsy, including subsequent frozen section analysis. Methods such as confocal laser endomicroscopy (CLE) have been proposed as real-time in vivo assessment tools, but such methods are more relevant for identifying lesions for subsequent histological confirmation. CLE results cannot provide a definitive diagnosis; tissue confirmation is still required.

[0046]

[0061] The method described herein answers an unmet need. While endoscopic biopsy remains the gold standard for diagnosing cancers such as lung cancer, frozen samples of potentially cancerous tissue must be examined to ensure adequate and representative (i.e., tumor-bearing) tissue has been obtained. This is an old technique established in the 1950s, when surgeons requested intraoperative information from patients while they were still in the operating room. Frozen tissue samples are also used during tumor resection to determine whether clean margins were obtained, i.e., whether the entire tumor was removed. It has long been recognized that information from frozen section diagnosis of cancer is often flawed and, harmfully, has a higher rate of misdiagnosis than permanent sections, which require at least a day for processing, sectioning, staining, and interpretation. As a result, a compromise has been reached between the two solutions. Today, the most common frozen section diagnosis is "sufficient tissue for diagnosis, with final diagnosis deferred for analysis of permanently preserved sections." This allows the surgeon to close the tissue extraction site but provides no other information. It is also somewhat uncertain, leading to multiple biopsies being performed until the pathologist is satisfied that sufficient tissue has been obtained. Given the poor histologic appearance of frozen section tissue, a definitive diagnosis is often impossible; indeed, dynamic features such as cellular characteristics (tumor, immune stroma, etc.) are impossible to assess, or even reliably identify, due to poor image quality.

[0047]

[0062] In contrast, images obtained as described herein are of very high quality, e.g., low noise and free of artifacts due to freezing or fixation. When processed and analyzed using these methods, images can generate dynamic cellular imaging data in three dimensions, which is not the case with frozen sections, cytology preps, or CLE and its variants, such as needle-based confocal laser endomicroscopy (nCLE, Mauna Kea Technologies). The resulting dynamic cellular data, including shape, size, motility, and high-resolution live cell relationships with other cellular and tissue components, are not available using other methods. For example, the widespread use of immunological agents such as PD1 and CTLA inhibitors has placed a premium on identifying the immune cell repertoire within and surrounding patient tumors.

[0048]

[0063] The analytical and diagnostic methods described herein have the potential to address this unmet need in real time in a manner not possible with stained tissue sections and beyond the capabilities of liquid biopsies, as discussed above. Liquid biopsies aim to provide such information, but contend with high background "noise" due to non-tumor contributions to the signal and lose spatial correlation, both of which are not true for the methods described herein.

[0049]

[0064] However, the use of the image acquisition and processing methods described herein does not preclude the subsequent use of the ROSE method. On the contrary, the present method may improve currently performed ROSE and provide a better alternative. In such cases, because the imaging method described herein does not alter the tissue sample, high-quality tissue samples remain available for additional research. This is not true for cytology preps or frozen sections that are subsequently fixed in formalin. CLE methods further limit subsequent research due to the lack of tissue for evaluation. For all these reasons, the methods described herein can be inserted between treatment and subsequent tissue processing, allowing virtually unlimited opportunities for subsequent tumor characterization. For example, in tumor genomic profiling, a portion of the biopsy tissue must be set aside to avoid performing assays on significantly inferior formalin-fixed, paraffin-embedded (FFPE) tissue, thereby reducing the amount of tissue remaining for other analyses. On the other hand, using the methods described herein, the entire tissue sample remains viable after imaging and analysis, and the high quality of the FF-OCT images allows for easy identification of the most representative regions for further analysis.

[0050]

[0065] Thus, the methods described herein can provide a dramatic enhancement of the information that can be obtained from any type of biopsy (especially from small tissue samples obtained by minimally invasive biopsies) in real time, without destroying any part of the specimen and compromising subsequent detailed analyses, such as DNA sequencing and protein analysis, which are increasingly common in the management of cancer patients. The methods described herein may have the potential to provide highly important information not available by other means. Some examples include non-tumor cell infiltration, a parameter of growing interest because it predicts patient response to expensive immunotherapy and the distribution of tumor cells within the biopsy. This is crucial in cancer surgery. Surgeons often perform a biopsy of a sufficient size of tissue from the tumor bed to document "clean margins" with normal tissue surrounding the tumor and complete the tumor resection. Positive margins indicate insufficient surgical resection and the need for additional tissue removal. Unfortunately, frozen sections (or sections prepared by the next day) can be difficult to detect. Permanent sections (PDX) are two-dimensional assays of three-dimensional tissue, allowing tumors to spread to and beyond the tissue margins. As a result, the true clinical scenario becomes apparent only when a patient's tumor recurs locally, providing conclusive evidence that it has not been completely removed. The ability to image specimens in three dimensions and identify tumor cells can be an extremely useful adjunct to surgical margin analysis and the detection of residual tumor. Yet another example is the use of so-called PDX (patient-derived xenografts). Pharmaceutical and biotechnology companies are increasingly relying on human tumors directly implanted into permissive hosts (PDX xenografts) to more realistically assess response to various targeted therapies. Because most biopsies are frozen or fixed (killing tissue cells), PDX tissue is scarce and therefore useless for this purpose. As valid, validated tumor tissue (which can be nondestructively identified with current methods) becomes more widely available after biopsy, the method described herein can be used to identify therapies most likely to benefit patients, replacing current methods of tumor marker-based prediction and other indirect measurement methods that often fail to predict response.

[0051]

[0066] Methods for Imaging and Classifying Cells: Deep Learning Analysis Methods. As mentioned above, the combination of data contained within spatially dependent optical coherence (FF-OCT) images and time dependent optical coherence (DCI) images obtained using a system such as system 200 described below can provide much more detailed information about the state of a region of cells, whether used for diagnosis, for example, from a biopsy sample, or for screening-type assays, for example, from multiple cells, as shown in FIG. 1A. The FF-OCT image 110 and the DCI image 120 can be spatially registered to the multiple cells under analysis.

[0052]

[0067] As shown in Figure 3A, a spatially dependent interference image 300 (FF-OCT image) shows structural details of the sample, with region 310 being a small portion of image 300. Figure 3B shows a magnified version of image 310. Figure 4A shows a time-dependent interference image 410 (DCI image), which is the same region shown in region 310. Figure 4A was processed as described herein to generate a colorimetric metabolic index (MI) image, in which cells are coded into red, green, and blue channels according to their metabolic rate (in terms of color) and level of activity (in terms of brightness). Simply put, the MI representation displays cells as brighter if they are more metabolically active and red if they have a faster metabolic rate. A subregion 420 in Figure 4A is magnified and shown in Figure 4B, making the details of individual cancer cells 405 more readily apparent. In this type of image representation, the color gradient from blue to green to red defines increasing rankings of movement, e.g., faster metabolic activity, and pixel brightness represents greater intensity of activity. Figure 4C shows the same region 410 as Figure 4A. Unlike Figure 4A, a subregion 430 has been selected. Figure 4D shows a magnified region 430, where details of individual immune cells 415 are more readily apparent. Figure 4E compares the metabolic index between cancer cells 405 and immune cells 415 in regions 420 and 430, respectively, and shows that each cell type has a different color intensity distribution for each color (e.g., red (R), green (G), blue (B)) and can therefore be distinguished.

[0053]

[0068] In other embodiments, variations of this analysis can be performed, as shown in FIG. 1B. As discussed above, color representations, e.g., MI, can provide visual information to a user viewing an image and can be used in deep learning-assisted analysis for diagnosis. However, time series contain much more data than the MI color representation would suggest. In the variation shown in FIG. 1B, the deep learning analysis can be provided with all the data contained in a data cube (including coordinates and time) associated with a series of time-dependent images, e.g., DCI image stack 120′, which can include any n images, where n is between about 2 and about 10. 3 images, or about 10 to about 10 6 DCI images can also be acquired at various depths, allowing the entire thickness of the sample to be examined.

[0054]

[0069] In any case, deep learning analysis 130 of images 110 and 120 determines the spatial location of each The combined data can be extracted and classified from each pixel of the aligned images, enhancing detail and improving diagnostic accuracy without relying on laborious, highly specialized human visual analysis. More accurate diagnoses 140 can be obtained. These methods can therefore improve speed and accuracy, reducing workflow dependency on highly skilled clinicians and freeing up other, more advanced analyses.

[0055]

[0070] Deep learning is a class of machine learning algorithms that use multiple layers to progressively extract high-level features from raw inputs. The term "deep" in deep learning derives from the use of multiple layers in the network. The network can have any number of layers of limited size, which allows for practical applications and optimized implementations that require, for example, significant reductions in computing resources and computing time. The layers can also be heterogeneous and can deviate significantly from biologically informed models. Deep learning methods can be supervised, semi-supervised, or unsupervised machine learning methods. As used herein, deep learning methods can include any of a variety of architectures known in the art, such as deep neural networks, deep belief networks, recurrent neural networks, and convolutional neural networks, which can lead to analytical results that may exceed the performance of human experts. Deep learning analytical methods can be convolutional neural networks (CNNs), and in some variations, pre-trained CNNs. The CNN may be any suitable CNN, including but not limited to AlexNet, VGG16, VGG19, SqueezeNet, GoogLeNet, Inception v3, DenseNet-201, MobileNetV2, ResNet-18, ResNet-50, ResNet-101, Xception, InceptionResNetV2, NASNet-Large, NASNet-Mobile, ShuffleNet, etc. In some embodiments, the deep learning analysis method may be AlexNet, GoogLeNet, ResNet-101, or ShuffleNet. The deep learning analysis may automatically assign a state to at least one cell of the plurality of cells.

[0056]

[0071] In some variations, the deep learning method may be a method for determining the state, e.g., cancer state, of multiple cells when the multiple cells may be suspected of containing cancer cells. The state may be automatically assigned and selected from a normal cell state or a cancer cell state for an individual cell, multiple cells, or a region containing multiple cells. In some variations, determining whether the multiple cells are cancerous may further include determining the stage of the cancer cells. In some embodiments, determining the stage of the cells may include determining the stage, for example, the method is performed on a sample of cells from a patient who has already been treated with a first type of cancer therapy.

[0057]

[0072] In another variation, the deep learning method can be a method for determining the effect of a molecule or biological agent on a plurality of cells, which may include both diseased and non-diseased cells. For example, at least some of the plurality of cells may be infected with a virus, a bacteria, have metabolic dysfunction, secretory function, etc. Upon treatment with the molecule or biological agent, the effect of the molecule / biological agent can be analyzed to determine whether the diseased cells have recovered to a non-diseased state, whether they have been eliminated, e.g., inactivated, to prevent further spread of disease, whether the non-diseased cells have maintained a non-diseased state, or whether the non-diseased cells have been adversely affected, e.g., an off-target effect, by the molecule / biological agent. The result of the analysis, e.g., diagnosis 140, can include any of these determinations. For example, a plurality of cells, including one or more cancer cells, can be contacted with a molecule or biological agent capable of targeting the cancer cells for killing. Images acquired over a selected period of time can be presented to the deep learning method to determine whether the molecule / biological agent is effective in killing the cancer cells. When cancer cells are killed, the metabolic activity of such cells decreases and then disappears, which can be identified through deep learning analysis. In other applications, such as evaluating potential antivirals, virally infected cells may exhibit an increased metabolic rate due to viral replication. After contacting multiple cells with a molecular / biological agent, images obtained over a selected time period after administration can be submitted to deep learning analysis. The analysis can determine whether the infected cells exhibit a decrease in metabolic activity to metabolic levels associated with non-diseased cells, e.g., whether viral infection has been reduced or eliminated. Alternatively, the analysis can determine whether the infected cells are killed before administration of the molecular / biological agent, e.g., whether the cells exhibit no metabolic activity. Furthermore, the analysis can determine whether administration of the molecular / biological agent prevents additional cells, e.g., neighboring cells, from becoming newly infected. For example, neighboring cells may exhibit an increase in metabolic activity over the imaging period after administration, which may indicate newly infected cells.

[0058]

[0073] Multi-layer algorithmic analysis can analyze and classify a wide variety of features. Features may be extracted from algorithmic analysis, for example, as a result of analysis, or features may be extracted from image preprocessing. When features are extracted from preprocessing, they are derived from metrics calculated prior to algorithmic analysis, and subsequent metric classification or metric grouping by algorithmic analysis better separates and segments the dataset. Any suitable set of features can be used in this method. In some analyses, such as diagnosing whether an area of ​​cells is cancerous, features used in deep learning analysis can determine increased local cell density, a range of cell numbers per unit volume / area, the number of mitotic cells (e.g., altered area and shape), cell motility, increased backscatter signal in cancer cells (possibly due to increased trafficking to support faster, uncontrolled cancer cell growth), altered extracellular matrix, and / or collagen fibers (e.g., evidence of a more disrupted environment compared to healthy tissue). In other applications, such as determining the effects that molecules or biological agents may have on cells, similar features can also be used as part of deep learning methods. In the latter type of analysis, the features can also include a determination of the extent of cell death over the image acquisition period.

[0059]

[0074] In some variations, the method may further include training a multi-layer algorithmic analysis, where a portion of the data from the time-dependent interferogram and / or a portion of the data from the space-dependent interferogram may be used to train the deep learning method. In some variations, the training may be based on expert analysis. The portion of the data reserved for training the untrained algorithm may be about 20%, about 30%, about 33%, about 40%, about 45%, about 50%, or any percentage therebetween of the total amount of data.

[0060]

[0075] In some other variations, the deep learning method may also present images of multiple cells, which may include labeling with a detectable label, which may be colorimetric, radiometric, fluorescent, or the like. For example, the labeling may be hematoxylin and eosin (H&E), which labels nucleic acids and proteins within the cells, an example of which is shown in FIG. 5. H&E-stained image 410 is the same cellular region as regions 310, 410 in FIGS. 3A-B and 4A-B. In some other examples, a dye such as Sirius Red (Sigma Aldrich Cat. No. 365548) can be used to image collagen and analyze fiber orientation and density. In another non-limiting example, any cell surface marker, such as programmed death-ligand 1 (PD-L1), can be labeled, which can aid in diagnosis and / or guidance of anti-cancer treatments or the determination of the effectiveness of molecular / biological agents. Deep learning methods can combine data obtained from such labeling to further analyze and classify the cells in the image.

[0061]

[0076] In some other variations, the deep learning method may further include distinguishing structural features of the plurality of cells and reducing interference in a time-dependent interference image of the plurality of cells.

[0062]

[0077] Method for Imaging and Classifying Intracellular Metabolic Activity: Machine Learning Analysis Method. The imaging and classification method 700 shown in FIG. 7 ends with a diagnosis and includes acquiring images of one or more regions of a tissue sample or a plurality of cells, as shown in box 750. The images can have an isotropic resolution of about 1 micron to about 5 microns. In some variations, the images can have a resolution of about 1 micron. For each region or field of view (FOV), at least one spatially dependent interferometry (FF-OCT) image and at least one time-dependent interferometry (DCI) image are acquired, as shown in box 710.

[0063]

[0078] The image can then be analyzed, as shown in box 720, to define, for example, segmented cells from other features visible in the image. Machine learning segmentation software can be used, such as ilastik, an open-source image classification tool (ilastik.github.io) based on a random forest classifier. Labels are manually drawn on the DCI image in a user interface. Each pixel neighborhood can be characterized by a series of common nonlinear spatial transformations applied to each channel (R, G, or B) of the DCI image. Image transformations can be selected empirically or using predetermined values ​​to obtain an image with the best contrast.

[0064]

[0079] Similar processing and segmentation is performed on the spatially dependent optical coherence (FF-OCT) images of each FOV using grayscale images. This process can be performed to identify and extract fiber-type structural features from tissue images. The classes identified and segmented are fiber, interfiber, and cellular.

[0065]

[0080] Based on the previous segmentation, as shown in box 730, general features can be extracted from individual ROIs as metrics for classifying samples (for each individual field of view) as normal or pathological. For example, cancer is a disease resulting from the loss of control over cell division and regeneration processes, leading to uncontrolled proliferation. Observations include increased local cell density, cell number, the number of mitotic cells (with altered area and shape), increased cell motility, and increased backscatter signal in cancer cells, potentially due to increased trafficking to support faster growth. As a result, the local environment, including the extracellular matrix and collagen fibers, can change, often exhibiting a more disorganized environment compared to healthy tissue.

[0066]

[0081] Multidimensional classification can also be included, as shown in box 740. To improve performance, machine learning classifiers can be applied in multidimensional space to enable comparison with various algorithms. For example, MATLAB®'s Matlab Toolbox Classification Learner allows for comparison of many algorithms. In some variations, a linear SVM (support vector machine) can be selected for analysis. When used as a non-probabilistic binary linear classifier, new examples are assigned to one of the categories, but SVM can also be used in probabilistic classification mode. A combination of FF-OCT and DCI features can be used as features, and when several values ​​are obtained (e.g., for cell diameter), both the mean and STD of the values ​​for each image can be included. Analysis can also include externally assigned phenotypes, such as histological assignments for each tissue.

[0067]

[0082] Thus, in some embodiments, a method is provided for detecting cancer cells in a pair of interference images, including a space-dependent interference image and a time-dependent interference image of a region of tissue comprising a plurality of cells, comprising: defining regions of the pair of images representing cellular boundaries of the plurality of cells; defining regions of the pair of images representing intracellular regions of the plurality of cells; and comparing the intensities of pixels in the time-dependent interference image of the intracellular region of a selected cell of the plurality of cells with those adjacent to the selected cell. automatically comparing the intensities of pixels in adjacent regions with the intensities of pixels in the subcellular region of the selected cell; automatically assigning a state label consisting of underactive, normally active, or overactive to the pixels in the subcellular region of the selected cell; summing the state labels in the subcellular region of the selected cell to thereby define the cell as healthy or overactive; and defining each overactive cell of the plurality of cells as cancerous. In some variations, the pair of interference images may be spatially registered. In some variations, the method may further include determining from the spatially dependent interference images that a subset of the plurality of cells represents a cell type that is not of interest, and the subset of the plurality of cells may be excluded from further analysis.

[0068]

[0083] Improved Biopsy Methods. Rapid On-Site Evaluation (ROSE) is one method for improving the likelihood of obtaining an adequate and clinically relevant biopsy sample, and a schematic diagram of a typical current workflow 600 is shown in FIG. 6A. The steps performed in the top two rows are typically performed in a radiology or operating room, while the steps in the bottom row are typically performed in a histopathology lab and / or pathology office. As currently practiced, imaging is performed in step 601 to guide tissue extraction. The imaging may be contrast-enhanced optical imaging, label-free optical imaging, radioactive imaging, ultrasound imaging, or magnetic imaging. In step 602, a biopsy needle may be guided or steered to the suspected site of disease using an endoscope, and tissue is excised in step 603. The sample is then transferred to a cytopathologist, who examines the sample and determines, while the patient is still in the biopsy procedure room, in step 604 whether sufficient cells are present and whether those cells represent tissue prompting the need for a biopsy. While the cytopathologist performs this analysis, the patient remains in the procedure room, awaiting a decision from the cytopathologist. If sufficient and representative cells are present, the biopsy procedure ends in step 605, and closure of the biopsy site may be performed. If insufficient and / or unrepresentative cells are found in the sample, the interventional radiobiologist or radiologist returns to the patient and excises more tissue in an additional excision in step 606, which combines with the initially acquired sample to form a sufficiently large and representative sample. Only then is the biopsy procedure complete, the patient is released, and the biopsy procedure room is freed for the next patient. The biopsy sample (or combined biopsy samples) is then transferred to a pathologist or cytopathologist to prepare slides in step 607, from which the patient's final diagnosis is derived in step 608. However, this involves a cytopathologist at two points in the process. Thus, although ROSE has currently been shown to provide a higher yield of appropriate final diagnoses, it is not widely implemented due to a lack of available cytopathologist resources. Furthermore, the results of cytopathologists are also subjective and prone to inter- and intra-observer variability.The ROSE method, which incorporates and relies on indirect tools such as CLE, can also be prone to false positives and false negatives due to its subjective nature and, furthermore, does not provide tissue specimens for final downstream analysis.

[0069]

[0084] In contrast, radiologists and interventional radiologists can perform biopsies without having to rely on cytopathologists or cytotechnologists to perform ROSE if DCI and / or FF-OCT imaging as described herein is used at the time of biopsy, as shown in workflow 650 in Figure 6B. The steps shown in the top row of the upper row are performed by the interventional radiologist or radiologist in the radiology suite or operating room, while the steps in the bottom row are performed by the pathologist or cytopathologist in the histopathology laboratory and / or pathology office. The immediate analysis of the excised material using DCI and / or FF-OCT imaging in step 609 can be performed directly in the biopsy procedure room, reducing the time required per patient in the procedure room, eliminating the cost and time commitment of a pathologist / cytopathologist, and even allowing the pathologist / cytopathologist to be away from the clinic where the biopsy is performed.

[0070]

[0085] Thus, there is provided a method for performing a biopsy on a subject requiring a biopsy, comprising imaging a region of tissue to identify a region of interest and inserting a biopsy needle into the region of interest. The method includes excising a first tissue sample from the region of interest, obtaining a set of time-dependent interference images and space-dependent interference images of the first tissue sample, and determining the number of target cells present in the first tissue sample. Imaging the region of tissue may include contrast-enhanced optical imaging, label-free optical imaging, radioactive imaging, ultrasound imaging, or magnetic imaging. In some embodiments, inserting a biopsy needle may include guided insertion. In some embodiments, determining the number of target cells may include quantified counts of target cells. In other embodiments, determining the number of target cells may include estimating the number of cells or annotating multiple cells without counting the number of cells present.

[0071]

[0086] Imaging the area of ​​tissue, inserting the biopsy needle, removing the first tissue sample, obtaining a set of time-dependent and space-dependent interference images, and determining the number of target cells present may be performed within a biopsy procedure room.

[0072]

[0087] Obtaining the set of time-dependent interference images and space-dependent images may further include processing the images to obtain images of intracellular metabolic activity of a plurality of cells within the first tissue sample.

[0073]

[0088] The method can further include assigning a state to one or more cells of the plurality of cells, wherein the one or more cells having the assigned state are target cells. The assigned state can be a disease cell state. In some embodiments, the disease cell state can be a cancer cell state. In other embodiments, the assigned state can be a cell type identification, such as an immune cell, including, but not limited to, a T cell, a NK cell, etc.

[0074]

[0089] Determining the number of cells of interest can include submitting the image of intracellular metabolic activity to processing by a multi-layer algorithm to thereby assign a state to the one or more cells. Assigning a state to the one or more cells can include comparing the level of metabolic activity observed in the one or more cells to a preselected threshold.

[0075]

[0090] The method may further include obtaining a second tissue sample from the area of ​​interest if the number of cells of interest in the first tissue sample is insufficient for analysis.

[0076]

[0091] Methods for analyzing the effects of molecular / biological agents on cells. Using methods for analyzing images from DCI and / or FF-OCT images, the effects of molecular and / or biological agents on a plurality of cells can be determined. In particular, the effects of molecular and / or biological agents can be performed on a mixture of diseased and non-diseased cells. The plurality of cells can further include more than one type of cell, thereby allowing for early identification of off-target effects of therapeutic agents on other types of cells typically found near the diseased cell type.

[0077]

[0092] The plurality of cells may be held in an imaging vessel that allows for cell maintenance, e.g., medium exchange, gaseous environment exchange, and nutrient addition. A series of time-dependent interferometry (DCI) images can be acquired over a period of time to observe the effect of the molecular and / or biological agent, and the images can be processed to provide a metabolic index (MI) image that indicates coded high, medium, or low intracellular activity. The effect of the molecular and / or biological agent can be compared between diseased and non-diseased cells to determine whether the molecular and / or biological agent is effective in treating the disease in the diseased cells. For example, at least some of the plurality of cells may have a viral infection, a bacterial infection, metabolic dysfunction, secretory function, or the like. Upon treatment with the molecular or biological agent, the effect of the molecular / biological agent can be analyzed to determine whether the diseased cells have recovered to a non-diseased state, been eliminated, e.g., inactivated, to prevent further spread of disease, whether the non-diseased cells maintain a non-diseased state, or whether the non-diseased cells have been adversely affected by the molecular / biological agent, e.g., turned off. It can be determined whether the target effect is being experienced. In many disease states, such as cancer or viral infection, diseased cells may have higher levels of intracellular metabolic activity, which can be imaged as described herein, and analysis can determine whether the level of metabolic activity in diseased cells is reduced compared to the metabolic activity of non-diseased cells upon contact with a molecule and / or biological agent. Analysis can also allow for observation of whether the molecule and / or biological agent adversely affects the metabolic activity of non-diseased individuals, for example, whether it reduces or eliminates metabolic activity, which may be seen as an off-target activity of the molecule and / or biological agent.

[0078]

[0093] For example, a plurality of cells, including one or more cancer cells, can be contacted with a molecule or biological agent capable of targeting cancer cells for killing. Images obtained as described herein over a selected period of time after administration can be analyzed to determine whether the molecule / biological agent is effective in killing the cancer cells. When cancer cells are killed, their metabolic activity decreases and then disappears, which can be identified in a set of images. In other applications, such as evaluating potential antiviral agents, virally infected cells may have an increased metabolic rate due to viral replication. After contacting a plurality of cells with the molecule / biological agent, images obtained over a selected period of time after administration can be analyzed to determine whether the infected cells exhibit a decrease in metabolic activity to a metabolic level associated with non-diseased cells, e.g., whether viral infection has been reduced or eliminated. Alternatively, analysis can determine whether the infected cells observed prior to administration of the molecule / biological agent are killed, e.g., whether the cells exhibit no metabolic activity. Furthermore, analysis can determine whether administration of the molecule / biological agent prevents additional cells, e.g., neighboring cells, from becoming newly infected. For example, adjacent cells may show increased metabolic activity over the imaging period following administration, which may indicate newly infected cells.

[0079]

[0094] This method can be used to guide therapy in patients who have already completed a first course of treatment to identify appropriate next-line therapies. This method can be used to screen or assay therapeutic agents during discovery or development in preclinical research.

[0080]

[0095] The observation period during which imaging is performed can be about 1 hour, about 4 hours, about 8 hours, about 12 hours, about 24 hours, about 36 hours, about 48 hours, about 72 hours or more.

[0081]

[0096] In some variations, a spatially dependent interference image can be obtained, where structural details of the cells can be distinguished from subcellular details. The effect of structural details can be used to further modify the time dependent interference image to remove interference and provide a clearer picture of the intracellular metabolic activity of the cells.

[0082]

[0097] The molecule may be a biomolecule or an organic molecule. In some variations, the biomolecule may be a protein, a nucleic acid, a sugar, or an expression product of a cell. In some variations, the organic molecule may be an organic compound having a molecular weight of less than about 2000 Da. In some other variations, the biological agent may be a virus, a phage, a bacterium, or a fungus.

[0083]

[0098] Imaging System and Image Processing Method An embodiment of an imaging system 20 suitable for implementing methods according to the present invention is shown schematically in FIG.

[0084]

[0099] The imaging system 20 comprises an interferometric device 200 , an acquisition device 208 , and at least one processing unit 220 .

[0085]

[0100] The interferometric device 200 is, on the one hand, illuminated by a spatially incoherent, low-coherence-length light source 201 through each elementary surface of the reflecting surface 205 of the reference arm of the interferometric device. The device is adapted to generate optical interference between, on the one hand, a reference wave obtained by reflection of light emitted by the same light source and, on the other hand, an object wave obtained by backscattering of light emitted by the same light source by each voxel of a slice of the sample 206 in the depth direction of the sample, the sample 206 being arranged on the object arm of the interference device, said voxels and said elementary plane corresponding to the same point of the imaging field.

[0086]

[0101] The light source 201 is a temporally incoherent or low coherence length (in practice, in the range of 1 to 20 micrometers) and spatially incoherent light source, such as a halogen lamp or an LED. According to one or more exemplary embodiments, the light source 201 can form part of the imaging system 20, as in the example of Figure 2, or can be an element external to the imaging system, which is adapted to operate with the light waves emitted by the light source.

[0087]

[0102] The acquisition device 208 allows the acquisition of at least one two-dimensional interference signal resulting from the interference between the reference wave and the object wave.

[0088]

[0103] The processing unit 220 is configured to perform at least one step of processing at least one two-dimensional interference signal acquired by the acquisition device 208 and / or at least one step of image generation according to at least one of the imaging methods according to the present specification to generate at least one image of the sample slice.

[0089]

[0104] In one embodiment, the processing unit 220 is a computing device that may include a first memory CM1 (not shown) for storing digital images, a second memory CM2 (not shown) for storing program instructions, and a data processor capable of executing the program instructions stored in this second memory CM2, and in particular controlling the execution of at least one step of processing at least one two-dimensional interference signal acquired by the acquisition device 208 and / or the execution of at least one step of image calculation according to at least one of the imaging methods described herein.

[0090]

[0105] The processing unit may also be fabricated in integrated circuit form including electronic components suitable for implementing the function(s) described herein for the processing unit. Processing unit 220 may also be implemented by one or more physically separate devices.

[0091]

[0106] The acquisition device 208 may be, for example, an image sensor of the CCD (Charge Coupled Device) or CMOS (Complementary Metal Oxide Semiconductor) camera type, which is capable of acquiring images at high speed, for example at a frequency of 100 Hz. Depending on the dynamics of the sample under investigation, and more particularly the dynamics of the movements within the sample, cameras operating from a few Hz to a few KHz can be used.

[0092]

[0107] According to one embodiment, the interferometer 200 includes a beam splitter element 202, for example a non-polarizing splitter cube, making it possible to form two arms. One of the arms, hereinafter referred to as the "reference arm", has a flat reflective surface 205, for example a mirror. The other arm (hereinafter referred to as the "object arm") is intended to receive, during operation, a three-dimensional diffuse sample 206 of a slice for which it is desired to generate a tomographic image at at least one depth according to one of the methods herein.

[0093]

[0108] In the example of Figure 2, the interferometer is of the Linnik interferometer type and comprises two identical microscope lenses 203, 204 arranged in each arm. Thus, the reflecting surface 205 is arranged at the focus of the lens 204 in the reference arm, and the sample 206 is arranged at the focus of the lens 203 in the object arm. Other types of interferometers can be envisaged for the implementation of the method according to the present invention, in particular Michelson, Mirau, Fizeau and other such types of interferometers.

[0094]

[0109] At the output of the interferometer 200, there is an optical system 207, e.g., an achromatic doublet, whose focal length is adapted to allow proper sampling of the sample 206 by the acquisition device 208, allowing the planes located at the focal points of the two lenses at the output of the interferometer to be conjugated in one coplanar plane. The acquisition device 208 is positioned at the latter plane to acquire the interference signal generated by the interferometer. To avoid limiting the resolution allowed by the microscope lenses 203 and 204, the focal length of the optical system 207 is selected in accordance with the Shannon criterion. The focal length of the optical system 207 is, for example, several hundred millimeters, typically 300 mm. Glass plates 209 and 210 are provided in each arm, if necessary, to compensate for dispersion.

[0095]

[0110] Time-Dependent Interferometry Images. For time-dependent interference images, such as those obtained by dynamic cellular full-field optical coherence tomography (DC-FFOCT), also known as dynamic cellular imaging (DCI) described herein, the images are acquired as follows: Because the light source 201 has a small coherence length, interference between the light reflected by the reflecting surface 205 (reference wave) and the light backscattered by the sample 206 occurs only if the optical paths of the two arms are equal within the coherence length. Therefore, at a specific depth in the sample, called a coherence slice, interference occurs between the reference wave and the light backscattered by each voxel of the slice located in a plane perpendicular to the optical axis of the object arm, where a voxel is an elementary volume defined by the coherence slice. The light backscattered by each voxel represents the amplitude of the coherent sum of the waves backscattered by all diffusing elementary structures present in this voxel. These images do not rely on a large depth of field, allowing the use of microscope objectives with high numerical apertures. High lateral resolution, ranging from 0.5 to 1.5 microns, can be achieved.

[0096]

[0111] Interference signals resulting from optical interference between the reference wave and the waves backscattered by different voxels are acquired in parallel at an instant t by the acquisition device 208. The result is an interference image S corresponding to the state of interference at a given instant t of the coherence slice. An interference image element or image pixel located at a given position (x, y) is defined with respect to a two-dimensional coordinate system associated with the acquisition device 208 and exhibits a value S(x, y, t) corresponding to the intensity of the interference signal acquired at the position (x, y) at the instant t as a result of interference between the wave backscattered by the voxel at the corresponding position in the sample and the reference wave reflected by the primary surface of the reflecting surface 205 of the reference arm at the corresponding position.

[0097]

[0112] The images may further display the intracellular metabolism of the imaged cells by analyzing and displaying a time series of each image pixel along several seconds (e.g., for a period ranging from about 1 second to about 5 seconds, at an acquisition rate of about 100 frames per second (fps) to about 500 fps). In some variations, the image time series may be about 3 seconds at 300 fps.

[0098]

[0113] A time-dependent interferogram, e.g., a DCI image, is calculated from the time-series interferogram and displays the temporal evolution of the intensity among N two-dimensional interference signals for the current slice of the sample. Each pixel IB(x,y) of the DCI image located at a given location (x,y) represents the value calculated for this given location of a selected parameter.

[0099]

[0114] In some variants, this parameter is a parameter that represents the time dispersion of the intensities of the N two-dimensional interference signals considered. Such a parameter can be, for example, the standard deviation of the statistical distribution of the intensities. In this way, an overall measurement is performed that represents the time dispersion of the backscattered light intensity at a given point in the biological tissue. The value obtained for this parameter is expressed in the form of an image. This allows the tissue regions where the movement occurs to be clearly displayed.

[0100]

[0115] According to one or more embodiments of the imaging system, the pixels of the image exhibit at least one component defined in terms of a colorimetric representation space, the value of which is a function of the value of a selected parameter.

[0101]

[0116] For example, a pixel of image IB located at a given position (x, y) and / or at least one component of this pixel exhibits a value defined in terms of a colorimetric representation space, which is a function of the value calculated for the relevant parameter for the corresponding position (x, y) from the intensities SNi(x, y) of the N acquired interference signals (i = 1 to N). For example, if the colorimetric representation space used for image IB is a gray level representation, the value of pixel IB(x, y) can be equal to or a function of the value VN(x, y) within a scaling factor to obtain a gray level coded with a given number of bits. For example, in the case of a gray level image, zones of samples animated by large movements and therefore with a high value of this parameter appear in such an image with a high gray level. On the other hand, areas where no movement is detected and exhibit a parameter value of zero exhibit a very low gray level. In another variant, if the colorimetric representation space used for image IB is a representation in the RGB (red, green, blue) colorimetric representation space, then at least one of the components R, G, or B of pixel IB(x,y) at position (x,y) in image IB is equal to VN(x,y) or is a function of VN(x,y) within a scaling factor, for example, to obtain a colorimetric component coded with a certain number of bits. In the case of images analyzed for motion, in the RGB colorimetric representation space, zones of samples animated by large movements and therefore with high values ​​of this parameter appear in such images with a red colorimetric representation, zones with low movement with a blue colorimetric representation, and zones with medium movement with a green color, providing a comprehensible metabolic index (MI).

[0102]

[0117] Further details of image processing of time-dependent interferometry (DCI) images are described in "Method and System for Full-Field Interference Microscopy Imaging,” International Patent Application No. PCT / EP2016 / 057827 to Boccara et al., the entire disclosure of which is incorporated herein by reference in its entirety.

[0103]

[0118] Spatially Dependent Interferometric Images. For spatially dependent interferometric images, e.g., images acquired using full-field optical coherence tomography (FF-OCT), an imaging system such as system 200 described above is used to obtain images with varying object and reference arm lengths, e.g., different object and reference arm lengths. In some variations, a series of images can be obtained with varying object arm lengths without changing the reference arm length (in all cases, the reference arm length is different from the object arm length used). The images can be processed as described in International Patent Application No. PCT / FR01 / 03589 to Boccara et al., entitled "Method and Device for High-Speed ​​Interferential Microscopic Imaging of an Object," filed November 15, 2001, and published as WO0240937, the entire disclosure of which is incorporated herein by reference in its entirety. Example

[0119] General: A commercially available FFOCT / DCI system (LLTech, Paris, France) with an isotropic resolution of 1 micron is used for all imaging. [Example]

[0104]

[0120] Diagnosis using deep learning classification of image parameters extracted through deep learning. Tissue samples from human breast tissue from patients are imaged from at least 85% of all patients with breast cancer and at least 5% of all patients without breast cancer. A total number of fields of view (FOVs) containing one DCI image and one FF-OCT image are analyzed. Each of these images is first processed separately. Each pair of images is subjected to the deep learning method described herein, and several parameters are analyzed, including increased local cell density, a range of cell numbers per unit volume / area, the number of mitotic cells (e.g., altered area and shape), cell motility, increased backscatter signal in cancer cells (possibly due to increased trafficking to support faster, uncontrolled cancer cell growth), altered extracellular matrix, and / or collagen fibers (e.g., evidence of a more disrupted environment compared to healthy tissue). The combined DCI and FF-OCT image data is expected to yield a diagnostic accuracy of greater than 95%, with a sensitivity and specificity of greater than 95%. [Example]

[0105]

[0121] Diagnosis using deep learning classification of images. Images were captured and analyzed similarly to Experiment 1 using the multi-layer algorithm analysis described herein. A confusion table for a total of 153 samples is shown in Table 3. This confusion table shows an overall accuracy of approximately 92%, a sensitivity of 91.8%, and a specificity of 91.8%, an improvement over the methods in Examples 3 and 4.

[0106]

[0122] Table 3. Confusion table for deep learning analysis.

[0107] [Table 1] [Example]

[0108]

[0123] Diagnosis using machine learning classification of features obtained by machine learning assisted segmentation. Tissue samples from human breast tissue from 29 patients were imaged (23 patients with breast cancer and 6 patients without breast cancer) (box 710 in Figure 7). Within each tumor sample, a large number (7 to 23) fields of view (FOVs) were analyzed, resulting in a total of 298 FOVs, where one DCI image and one FF-OCT image were each analyzed separately first.

[0109]

[0124] As shown in box 720 of Figure 7, segmentation was performed as follows: The first step of the automated analysis was to perform cell segmentation in the images (box 720). For this, we used ilastik, a free and relatively intuitive machine learning tool segmentation software. ilastik is based on a random forest classifier. Labels were manually drawn on the DCI images in the user interface, e.g., a manual thresholding approach. Each pixel neighborhood was characterized by a set of common nonlinear spatial transformations applied to each channel of the DCI image, such as R, G, or B. The image transformation that empirically provided the best contrast was used, and the learning process ran for approximately 20–30 minutes.

[0110]

[0125] Similar processing and segmentation procedures were performed on FF-OCT images (grayscale images only) to extract fibers. The classes used were fiber, interfiber, and cell.

[0111]

[0126] Based on the previous segmentation, in box 730, a sample (one field of view) is selected. Common features can be extracted from individual ROIs as metrics for classifying them as normal or pathological. While some features used in DCI and FF-OCT are unreliable in the final analysis, they are potentially good indicators of cancer. Because cancer is a disease resulting from the loss of regulatory control of cell division and replication cycles, cancer results in changes in local cell density, cell number, the number of mitotic cells (altered in area and shape), cell motility, and backscattering signal in cancer cells, which in some cases increases trafficking (e.g., increased scattering density within a voxel), alterations to the local environment, including the extracellular matrix, and collagen fibers, which often indicate a more disorganized environment compared to healthy tissue, leading to uncontrolled proliferation. Therefore, all of these features were useful for feature extraction.

[0112]

[0127] To improve performance, a number of algorithms were compared using machine learning classifiers in multidimensional space using the Matlab® toolbox Classification Learner (box 740). A combination of FF-OCT features and DCI features was used as features. When several values ​​were obtained (e.g., for cell diameter), both the mean and standard deviation of the values ​​for each image were used, resulting in 42 features. Table 1 summarizes the scores obtained using various strategies. A linear SVM (support vector machine) with an external phenotype (given by the histology of each tissue) provided a diagnosis with 90% sensitivity and 86% specificity for individual images (box 750). However, taking into account the tissue and selecting tissues with a cancerous fraction of the FOV greater than 75%, the sensitivity and specificity reached 100%.

[0113]

[0128] Using this first manual threshold analysis method, a sensitivity of 100% (21 true positives compared to 0 false positives) and a specificity of 75% (2 false negatives and 6 true negatives) was obtained in the diagnosis.

[0114]

[0129] Table 1. Algorithm comparison.

[0115] [Table 2] [Example]

[0116]

[0130] Diagnosis using machine learning classification of features obtained by automatic machine learning-assisted segmentation. Imaging and analysis were performed as in Example 3, but without the initial manual threshold. Supplementary parameters after fiber and cell segmentation included, among others, FF-OCT / DCI signal intensity, DCI MI color (e.g., migration speed captured in the image). The confusion matrix obtained for a total of 524 FOV is shown in Table 2.

[0117]

[0131] Table 2. Confusion table for automated machine learning.

[0118] [Table 3]

[0119]

[0132] Combining all these parameters in the analysis resulted in an overall accuracy of 86.1%, as shown in Table 2. Averaging several areas of individual biopsies, similar to how standard histopathology processes average, provided a further improvement to 100% accuracy. [Example]

[0120]

[0133] A ROSE biopsy experiment uses time-dependent imaging and, optionally, space-dependent imaging in real time during the biopsy procedure. The patient is prepared in the biopsy procedure room for biopsy sample excision. Ultrasound imaging is used to guide a steerable biopsy needle to the site of suspected malignant cells. A first sample of cells is removed from the suspicious area, and the cells are immediately imaged using DCI and, optionally, FF-OCT. The images are processed as described herein and a metabolic index (MI) is assigned. If it is determined that there are an insufficient number of cells with a moderate to high MI, a second excision is performed to add to the cell count for that patient's analysis. After it is determined that the second excision sample provides sufficient clinically relevant cells via DCI with optional FF-OCT imaging and processing, the patient's biopsy site is closed, and the cell sample is transferred to pathology / cytopathology for permanent slide preparation. [Example]

[0121]

[0134] Screening of biological agents for the treatment of infected cells. Colonies of Pseudomonas aeruginosa Gram-negative bacteria are imaged by time-dependent interferometry (DCI) at time = 0 in a cell chamber that allows for perfusion, nutrient introduction, and waste removal. The effectiveness of a library of engineered phages with putative specificity for P. aeruginosa is investigated. Individual engineered phage populations are introduced into separate bacterial colonies, and DCI imaging is performed at 1-hour intervals up to 24 hours post-administration. Images are processed as described above to generate images with metabolic index (MI) coding. The kinetics of phage killing are observed, and the phage population with the best kinetics for bacterial cell killing is identified. [Example]

[0122]

[0135] Screening of biological agents for the treatment of infected cells. A mixed population of mammalian cells, such as mouse epithelial cells infected with Pseudomonas aeruginosa Gram-negative bacteria, is imaged by time-dependent interferometry (DCI) at time = 0 in a cell chamber that allows for perfusion, nutrient introduction, and waste removal. The effectiveness of a library of engineered phages with putative specificity for Pseudomonas aeruginosa is investigated. Individual engineered phage populations are introduced into separate mammalian cell / bacterial populations, and DCI imaging is performed at 1-hour intervals up to 24 hours post-administration. Images are processed as described above to generate images with metabolic index (MI) coding. The effect of phage killing on mammalian cell metabolism and the kinetics of phage killing are observed to determine whether the lytic effect of phage killing is harmful to mammalian cells. The phage population with the best kinetics of bacterial cell killing and the least harmful effects on mammalian cells is identified. [Example]

[0123]

[0136] Screening for molecules for the treatment of diseased cells. Spheroids, e.g., three-dimensional cultured cell compositions, containing populations of healthy and cancerous human breast cells are imaged by time-dependent interferometry (DCI) at time = 0 in a cell container that allows perfusion, nutrient introduction, and waste removal. A preselected concentration of a tyrosine kinase inhibitor is administered to the spheroids, and DCI imaging is performed at hourly intervals up to 72 hours after administration. The spheroids can also be imaged by spatially dependent interferometry (FF-OCT) at each time point. The images are processed as described above to generate images with metabolic index (MI) coding. The series of images is analyzed to determine whether the tyrosine kinase inhibitor is effective in killing cancer cells. Analysis also includes determining whether healthy cells in the spheroids are adversely affected by the administration of the tyrosine kinase inhibitor, e.g., whether they may exhibit off-target effects. [Example]

[0124]

[0137] Diagnosis using machine learning classification of features extracted using non-negative matrix factorization techniques.

[0125]

[0138] 8 illustrates a workflow for using the Non-Negative Matrix Factorization (NMF) algorithm in DCI acquisition of breast tissue. In the illustrated example, referring to step 801, one DCI acquisition of, for example, 1000 images acquired at 150 Hz may be acquired for each FOV (resulting in a time stack of interferometric frames equivalent to a 1440 x 1440 x 1000 pixel data cube per FOV).

[0126]

[0139] The raw interference (time) domain can then be transformed into the frequency domain to extract the appropriate metabolic information and remove incoherent portions of the signal. This may include normalizing the frames to a constant energy to remove frame-to-frame discrepancies introduced by the acquisition. At step 802 in FIG. 8, the method may further include averaging the frames by two groups to attenuate noise (resulting in 500 frames pseudo-acquired at 75 Hz). At step 803, the method may include passing the data to the frequency domain using a pixel-wise FFT (producing 250 frequency maps in 0.15 Hz steps). Next, the method may normalize the FFT by its norm L1 and pass it to a logarithmic scale to compensate for low-frequency amplitude distortion. This results in a frequency stack that preserves both spatial and dynamic information. At step 804, the method may include flattening the frequency data cube into a 2D table with 250 rows (frequency bins) and 2,073,600 columns (1,440 x 1,440 pixels). As a result, the spectrum of each pixel is treated as an individual data point, ignoring spatial organization.

[0127]

[0140] Next, in step 805 of Figure 8, a non-negative matrix factorization (NMF) algorithm can be applied to the flattened frequency cube for each DCI FOV individually. The goal of NMF is to factorize the data matrix X of size 2073600 x 250 into two low-rank positive matrices, i.e., X ≈ W H , where H is of size k x 250 and W is of size 2073600 x k. k is the number of components selected for splitting. The two constituent matrices can be found by minimizing the error (e.g., squared Frobenius norm - sum of squares) between the original data matrix and the factorization results.

[0128]

number

[0129] To solve this optimization problem, a multiplicative update algorithm is used, iteratively updating W and H in the direction of the gradient until convergence. The factorization rank k is chosen empirically and can be set using prior knowledge of the data and trial-and-error experiments. For example, k can be chosen to be 5. After NMF factorization, the matrix H reveals the frequency signature, where the revealed components correspond to the baseline signal, fibers, noise, cells, and movement artifacts. The W matrix reveals the corresponding spatial activation of the frequency signature. Figures 9A-9E show the results for each of the five components, respectively.

[0130]

[0141] In the next step 806 of Figure 8, a unified feature vector is constructed for each FOV in preparation for classification. To do this, the H components are sorted by energy (area under the curve) and those with the minimum and maximum energy are removed, as these correspond to the noise and baseline components, respectively. The remaining three components are then concatenated to form a single feature vector that characterizes each FOV. Sorting the components by energy also ensures consistency of the feature vectors across FOVs.

[0131]

[0142] To establish a diagnosis between cancerous and normal tissues in 382 FOVs (from 47 samples), a trained classifier can be applied to the integrated feature vector extracted from NMF (step 807 in Figure 8). The classifier can be a tree-based classifier, for example, AdaBoost, XGBoost, Random Forest, ExtraTrees, GradientBoosting, or DecisionTree. Using this classification method after feature extraction by NMF, a sensitivity of 78% and a specificity of 64% were obtained in diagnosis. [Example]

[0132]

[0143] Analysis with deep learning classification, feature learning with classification and segmentation.

[0133]

[0144] Similar to Example 9, the dataset includes individual FOVs acquired from different locations for each sample, with each FOV having 1440 x 1440 pixels. Each FOV can be input to a trained convolutional network for classification and detection. In this example, a pre-trained VGG-16 neural network was used, as shown in Figure 10. The VGG-16 neural network takes each FOV as input and passes it through a stack of convolutional feature extraction layers.

[0134]

[0145] A global average pooling layer (GAP) is added after the convolutional feature extraction layer, which generates a feature vector size of 512, representing the average activation of each filter in the last convolutional layer of VGG-16. After pooling, the classifier is kept at minimal complexity by using only one hidden layer of size 1024, followed by a binary output neuron with sigmoid activation.

[0135]

[0146] To determine the interpretability and reliability of a trained model, two approaches can be followed: gradient ascent by iteratively maximizing the activation of each convolutional filter to ensure that learning is not limited to classifiers but extends to feature extractors. Thus, we can obtain synthetic inputs and obtain textures learned from the data. See 1101 in Figure 11 for examples of some filters in the deepest convolutional layer. The second approach consists of displaying class activation maps of multiple inputs using the GradCAM method, which reveals "significant" regions in the input that indicate a particular class. This provides a coarse localization of the presence of a class in the input, which can serve many purposes. It is important to ensure that the model is not biased (e.g., higher importance of the context than the actual object of interest, or, on the other hand, highly focused attention on a small part of the object). See 1102 in Figure 11 for examples of positive GradCAM (localizing cancer cells) and negative GradCAM highlighting normal lobules.

[0136]

[0147] The attention map can be converted into a segmentation mask that can serve as ground truth for training the U-Net architecture in Figure 10, which is constructed by merging networks already trained on classification tasks and adding a decoder branch.

[0137]

[0148] In this example, classification and segmentation are streamlined together to obtain reliable classification along with course segmentation of DCI breast specimens. This approach allows pathologists to provide feedback and corrections with low throughput, leading to improved expert annotation.

[0138]

[0149] When a feature or element is referred to herein as being "on" another feature or element, it may be immediately on the other feature or element, or intervening features and / or elements may be present. In contrast, when a feature or element is referred to as being "directly on" another feature or element, there are no intervening features or elements present. When a feature or element is referred to as being "connected," "attached," or "coupled" to another feature or element, it should be understood that it may be directly connected, attached, or coupled to the other feature or element, or that intervening features or elements may be present. In contrast, when a feature or element is referred to as being "directly connected," "directly attached," or "directly coupled" to another feature or element, there are no intervening features or elements present. Although described or illustrated with respect to one embodiment, features and elements so described or illustrated may be applicable to other embodiments. Those skilled in the art will also understand that references to structures or features located "adjacent" to another feature may have portions that overlap or underlie the adjacent feature.

[0139]

[0150] The terms used herein are for the purpose of describing particular embodiments only and are not intended to be limiting of the invention. For example, as used herein, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly dictates otherwise. As used herein, it will be further understood that the terms "comprises" and / or "comprising" specify the presence of stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items and may be abbreviated as " / ."

[0140]

[0151] Spatially relative terms such as "under," "below," "lower," "over," and "upper" may be used herein for ease of description to describe the relationship of one element or feature to another element or feature as shown in the figures. It is understood that the spatially relative terms are intended to encompass various orientations of the device during use or operation in addition to the orientation shown in the figures. It should be understood that, for example, if a device in the figures is inverted, elements described as "under" or "beneath" other elements or features would then be oriented "over" the other elements or features. Thus, the exemplary term "under" can encompass both an orientation of above and below. The device may be oriented otherwise (rotated 90 degrees or at other orientations), and the spatially relative descriptors used herein would be interpreted accordingly. Similarly, terms such as "upwardly," "downwardly," "vertical," "horizontal," and the like are used herein for descriptive purposes only, unless otherwise noted.

[0141]

[0152] The terms "first" and "second" may be used herein to describe various features / elements (including steps), but these features / elements should not be limited by these terms unless the context dictates otherwise. These terms may be used to distinguish one feature / element from another. Thus, a first feature / element discussed below could be referred to as a second feature / element, and similarly, a second feature / element discussed below could be referred to as a first feature / element, without departing from the teachings of the present invention.

[0142]

[0153] Throughout this specification and the claims that follow, unless the context requires otherwise, the word "comprise," and variations such as "comprises" and "comprising," mean that various components can be used together in methods and articles (e.g., compositions and devices and apparatuses that include methods). For example, the term "comprising" is understood to mean the inclusion of stated elements or steps, but not the exclusion of other elements or steps.

[0143]

[0154] As used in this specification and claims, including in the examples, unless expressly specified otherwise, all numerical values ​​can be read as if preceded by the term "about" or "approximately," even if the term "about" or "approximately" is not explicitly stated. The phrase "about" or "approximately" can be used when describing a size and / or location to indicate that the stated value and / or location is within a reasonable expected range of value and / or location. For example, numerical values ​​can include values ​​such as + / -0.1% of the stated value (or range of values), + / -1% of the stated value (or range of values), + / -2% of the stated value (or range of values), + / -5% of the stated value (or range of values), and + / -10% of the stated value (or range of values). Any numerical value given herein should also be understood to include about or approximately that value unless the context dictates otherwise. For example, if the value "10" is disclosed, "about 10" is also disclosed. Any numerical range recited herein is intended to include all subranges therein. As will be appreciated by those skilled in the art, when a value is disclosed as being "less than or equal to" a value, it is understood that "greater than or equal to" the value and possible ranges therebetween are also disclosed. For example, if a value "X" is disclosed, "less than or equal to X" and "greater than or equal to X" (e.g., where X is a number) are also disclosed. It is also understood that throughout this application, data is provided in a number of different formats, and this data represents endpoints and starting points, and ranges for any combination of the data points. For example, when a specific data point "10" and a specific data point "15" are disclosed, it is understood that greater than 10 and 15, greater than or equal to 10 and 15, less than 10 and 15, less than or equal to 10 and 15, and equal to 10 and 15, and between 10 and 15 are considered to be disclosed. It is also understood that each unit between two specified units is also disclosed. For example, if 10 and 15 are disclosed, then 11, 12, 13, and 14 are also disclosed.

[0144]

[0155] Various exemplary embodiments have been described above and are set forth in the claims. Any number of modifications can be made to the various embodiments without departing from the scope of the invention. For example, the order in which the various described method steps are performed may often be changed in alternative embodiments, and in other alternative embodiments, one or more method steps may be omitted entirely. Optional features of the various device and system embodiments may be included in some embodiments and not in other embodiments. Accordingly, the foregoing description has been provided primarily for illustrative purposes and should not be construed as limiting the scope of the invention, which is set forth in the claims.

[0145]

[0156] The examples and illustrations contained herein illustrate, by way of illustration and not limitation, specific embodiments in which the subject matter may be practiced. As noted above, other embodiments may be utilized and derived therefrom, such that structural and logical substitutions and changes may be made without departing from the scope of the present disclosure. Such embodiments of the inventive subject matter, when in fact more than one is disclosed, may be referred to herein, individually or collectively, by the term "invention" merely for convenience, without any intention to intentionally limit the scope of the present application to any single invention or inventive concept. Thus, while specific embodiments have been shown and described herein, any configurations calculated to achieve the same purpose may be substituted for the specific embodiments shown. The present disclosure is intended to cover any adaptations or variations of the various embodiments. Combinations of the above embodiments, as well as other embodiments not specifically described herein, will be apparent to those skilled in the art upon reviewing the above description.

Claims

1. 1. A method for determining the state of a plurality of cells, comprising: acquiring time-dependent and spatially-dependent interferometric images of a plurality of cells suspected of containing cancer cells; submitting the time-dependent interferogram and the space-dependent interferogram to a multi-layer algorithmic analysis, thereby combining data associated with each pixel of the image of each of the plurality of cells; automatically assigning a state to at least one cell of the plurality of cells, the state being selected from a normal cell state or a cancer cell state; A method comprising:

2. 10. The method of claim 1, further comprising training the multi-layer algorithmic analysis by analyzing one or both of a portion of data from the time-dependent interferogram and a portion of data from the space-dependent interferogram.

3. The method of claim 1 or 2, wherein the time-dependent interferogram and the space-dependent interferogram are spatially registered.

4. 4. The method of claim 1, further comprising the step of submitting an image of the plurality of cells further comprising a detectable label to the multi-layer algorithmic analysis.

5. 5. The method according to any one of claims 1 to 4, distinguishing structural features of the plurality of cells; reducing interference in the time-dependent interference image of the plurality of cells; The method further comprises:

6. 6. The method of claim 1, wherein the multi-layer algorithmic analysis comprises a pre-trained convolutional neural network.

7. 7. The method of claim 1, further comprising automatically assigning a state to the subset of the plurality of cells, whereby a region in which the subset of the plurality of cells is located is annotated as normal or cancerous.

8. 8. The method of claim 7, wherein the subset of structural features from the spatially dependent interferometric image is presented to an artificial intelligence, such as a deep learning algorithm, thereby assigning the state of the region in which the subset of the plurality of cells is located.

9. 1. A method of performing a biopsy on a subject in need thereof, comprising: imaging a region of tissue to identify a region of interest; inserting a biopsy needle into the area of ​​interest; excising a first tissue sample from the region of interest; acquiring a set of time-dependent and space-dependent interferograms of the first tissue sample; determining the number of cells of interest present in the first tissue sample; A method comprising:

10. 10. The method of claim 9, further comprising the steps of imaging the region of tissue, inserting the biopsy needle, excising the first tissue sample, obtaining the set of time-dependent interference images and the space-dependent interference images, and determining the number of target cells present. The method, wherein the step of performing is performed in a biopsy procedure room.

11. 11. The method of claim 9 or 10, wherein acquiring the set of time-dependent interference images and space-dependent images further comprises processing the images to acquire images of intracellular metabolic activity of a plurality of cells within the first tissue sample.

12. 12. The method of claim 11, further comprising the step of assigning a state to one or more cells of the plurality of cells, wherein the one or more cells having the assigned state are target cells.

13. 13. The method of claim 12, wherein the assigned state is a diseased cell state.

14. 14. The method of claim 12 or 13, wherein the diseased cell state is a cancer cell state.

15. 15. The method of any one of claims 12 to 14, wherein determining the number of cells of interest comprises submitting the image of intracellular metabolic activity to processing by a multi-layer algorithm, thereby assigning the state to the one or more cells.

16. 16. The method of claim 15, wherein assigning the state to the one or more cells comprises comparing a level of metabolic activity observed in the one or more cells to a preselected threshold.

17. 17. The method of any one of claims 9 to 16, further comprising the step of obtaining a second tissue sample from the region of interest if the number of cells of interest in the first tissue sample is insufficient for analysis.

18. 18. The method of any one of claims 9 to 17, wherein imaging the region of tissue comprises contrast-enhanced optical imaging, label-free optical imaging, radioactive imaging, ultrasound imaging, or magnetic imaging.

19. 19. The method of any one of claims 9 to 18, wherein the step of inserting a biopsy needle comprises guided insertion.

20. 1. A method for determining the state of a plurality of cells, comprising: acquiring images of intracellular metabolic activity of a plurality of cells suspected of containing cancer cells, the images comprising time-dependent interference images; automatically assigning a state to at least a subset of the plurality of cells, the state being selected from a normal cell state or a cancer cell state; assigning a cancer stage state to said subset of said plurality of cells; A method comprising:

21. 21. The method of claim 20, acquiring a spatially dependent interference image of the plurality of cells; distinguishing structural features of the plurality of cells; reducing interference in the image of intracellular metabolic activity of the plurality of cells; The method further comprises:

22. 22. The method of claim 20 or 21, wherein the subset of the plurality of cells is wherein automatically assigning a state comprises submitting the image of intracellular metabolic activity of the plurality of cells to an artificial intelligence, such as a deep learning algorithm, thereby comparing the level of metabolic activity observed in the subset of the plurality of cells to a preselected threshold.

23. 23. The method of claim 22, wherein cells of the subset of cells are assigned to a cancerous state if the level of metabolic activity exceeds the preselected threshold.

24. 24. The method of any one of claims 20 to 23, further comprising automatically assigning a state to the subset of the plurality of cells, whereby a region in which the subset of the plurality of cells is located is annotated as normal or cancerous.

25. 25. The method of claim 24, wherein a subset of the structural features from the spatially dependent interferometric image is presented to the deep learning algorithm, thereby assigning the state of the region in which the subset of the plurality of cells is located.

26. 26. The method of any one of claims 20-25, wherein assigning the cancer stage state to the subset of the plurality of cells comprises at least one of determining a differentiation level of the plurality of cells, determining a level of cellular organization of the plurality of cells, determining the presence of biomarkers, and determining the cancerous / non-cancerous area status of other plurality of cells obtained from the same subject.

27. 27. The method of claim 26, wherein the cancer stage is y cancer stage.

28. 28. The method of any one of claims 20 to 27, wherein the time-dependent interferometric images comprise a set of images taken over a period of about 1 second to about 5 seconds.

29. 29. The method of any one of claims 20 to 28, wherein the time-dependent interferometric images comprise a set of images taken at a rate of about 50 fps to about 500 fps.

30. 1. A method for determining the effect of a molecule and / or a biological agent on a cell, comprising: acquiring images of intracellular metabolic activity of a plurality of cells, the plurality of cells including at least one diseased cell and at least one non-diseased cell, the images including time-dependent interference images; contacting the plurality of cells with one or both of a molecule and a biological agent; acquiring a plurality of images over a subsequent period of time, said plurality of images comprising intracellular metabolic activity of said plurality of cells; determining the effect of one or both of the molecule and the biological agent on the at least one diseased cell compared to the effect on the at least one non-diseased cell; A method comprising:

31. 31. The method of claim 30, acquiring a spatially dependent interference image of the plurality of cells; distinguishing structural features of the plurality of cells; reducing interference in the image of intracellular metabolic activity of the plurality of cells; The method further comprises:

32. 32. The method of claim 30 or 31, wherein acquiring the plurality of images over the subsequent period of time is performed for about 1 hour to about 3 days after contacting the plurality of cells with one or both of the molecule and the biological agent.

33. 33. The method of any one of claims 30 to 32, wherein determining the effect of one or both of the molecule and the biological agent on the at least one diseased cell compared to the effect on the at least one non-diseased cell comprises determining the level of metabolic activity for the at least one diseased cell and the at least one non-diseased cell over the subsequent period of time.

34. 34. The method of claim 33, wherein the level of metabolic activity in the diseased cells is increased compared to the level of metabolic activity in non-diseased cells.

35. 34. The method of claim 33, wherein the level of metabolic activity in the diseased cells is decreased compared to the level of metabolic activity in the non-diseased cells.

36. 36. The method of any one of claims 33 to 35, wherein the level of metabolic activity remains the same as for non-diseased cells.

37. 36. The method of any one of claims 33 to 35, further comprising identifying off-target activity of one or both of the molecule and biological agent on non-diseased cells.

38. 38. The method of any one of claims 30 to 37, wherein the molecule comprises a biomolecule or an organic molecule.

39. 39. The method of claim 38, wherein the biomolecule comprises a protein, a nucleic acid, a sugar, or an expression product of a cell.

40. 39. The method of claim 38, wherein the organic molecule comprises an organic compound having a molecular weight of less than about 2000 Da.

41. 41. The method of any one of claims 30 to 40, wherein the biological agent is a virus, a phage, a bacterium, or a fungus.

42. 1. A method for detecting cancer cells in a set of interferograms comprising a space-dependent interferogram and at least one time-dependent interferogram of a region of tissue comprising a plurality of cells, the method comprising: performing an analysis of said set of images, automatically defining regions in the pair of images that represent cell boundaries of the plurality of cells; automatically defining regions of the pair of images representing intracellular regions of the plurality of cells; automatically comparing pixel intensities in the time-dependent interference image of an intracellular region of a selected cell of the plurality of cells with pixel intensities in a region adjacent to the selected cell; automatically assigning a state label to the pixels of the subcellular region of the selected cells, including under-active, normally active, or over-active; summing a plurality of state signatures in the intracellular region of the selected cell, thereby defining the cell as healthy or hyperactive; and determining each hyperactive cell of said plurality of cells as cancerous; A method comprising:

43. 43. The method of claim 42, wherein the analyzing step is performed by a multi-layer algorithmic analysis.

44. 44. The method of claim 42 or 43, wherein the pair of interferograms are spatially registered.

45. 45. The method of any one of claims 42 to 44, further comprising determining from the spatially dependent interference image that a subset of the plurality of cells represents a cell type that is not of interest, thereby excluding the subset of the plurality of cells from further analysis.

Images