Morphometric genotyping of cells using optical tomography to detect tumor mutation load

Abstract

A method for developing one or more morphometric classifiers to identify tumor mutation load (TMB) is disclosed. The method provides a non-invasive method of characterizing TMB that is responsive to and independent of the size of a tumor in the early stage of tumor development. The method allows for cancer treatment for a specific characterization of the cancer the patient may suffer from, thereby achieving more efficient cancer management with much less side effects.

Description

[0001] This application is a divisional application of Patent Application No. 201980014269.6 filed in the PCT International Application PCT / US2019 / 012352 entitled "Morphometric Genotyping of Cells Using Optical Sectioning to Detect Tumor Mutation Burden" entered the Chinese National Phase on January 4, 2019. TECHNICAL FIELD

[0002] The present invention relates to optical sectioning of cells and sub-cellular scales. More specifically, the present invention relates to a system and method for developing one or more morphometric classifiers to identify tumor mutation burden (TMB). BACKGROUND

[0003] Alterations in nuclear morphology have been a major histopathological biomarker for cancer detection for the past 140 years. Direct links between chromatin organization in the nucleus and cellular functions in DNA replication, translation, and protein expression levels have been demonstrated in many published studies 1-3 . In particular, chromatin organization, which underlies the 3D architecture of the nucleus, is thought to be a major factor influencing regional and global mutation rates in human cancer cells 4-5 .

[0004] Current methods of treating various cancers involve targeting immune system checkpoint inhibitors CTLA4, PD-1, and PD-L1, which are proteins involved in allowing tumors to evade immune system responses 6 . While immune therapy has achieved durable responses in many solid tumors, only a fraction of patients truly benefit. For example, the following are response rates to single-agent PD-1 / PD-L1 inhibition: melanoma 40% 7、8 , non-small cell lung cancer (NSCLC) 25% 9、10 , renal cell carcinoma 19% 11 . Furthermore, current immune therapies have a strong risk of adverse side effects 12-14 . Therefore, there is a need for reliable biomarkers that can reliably predict which patients will benefit from immune therapy to reduce the unnecessary burden of inflammation and immune-related adverse reactions on patients. Several biomarkers have been identified that help predict patient response to immune therapy 15 . One of these biomarkers is the expression of PD-L1, which is necessary for therapeutic response, but due to tumor heterogeneity and measurement of expression levels, is not sufficient to determine response 16-21 . Recently, it has been found that mismatch repair (MMR) deficiency and tumor mutation burden (TMB; the number of somatic, coding, base substitution, and indel mutations per megabase of genomic DNA) are good predictors of response to immune therapy. 15、22-27MMR deficiency leads to genomic and microsatellite instability (MSI) and high TMB, resulting in the expression of neoantigens, which makes tumor cells more susceptible to attack by cytotoxic T cells. 23 Challenges in detecting TMB from solid tumors or ctDNA include the inability to perform biopsies, sensitivity of early disease detection, and lack of consistency in data obtained using paired tissues and different NGS (next generation sequencing) platforms. 28、29 As described below, a faster, less invasive, and less expensive method uses the VisionGate Cell-CT technology to detect low TMB versus high TMB in cancer cells based on the potential of structural biomarkers to impart morphometric changes in TMB, which can be quantified in sub-micron spatial range under 3D light microscopy. The established link between chromatin organization and genomic mutation rate indicates that structural biomarkers exist in the nucleus that can be used to detect genomic instability and / or more specific types of genomic alterations in cancer.

[0005] The ability to detect MMR deficiency and TMB levels by measuring structural biomarkers is supported by numerous studies. Studies have reported histopathological differences in microsatellite mutated colorectal cancers due to MMR deficiency. Typically, these tumors are mucinous and poorly differentiated, composed of relatively large, round, and regular cells with abundant amphophilic cytoplasm. 30-32 Additionally, Alexander et al. 33 reported that colon cancers with MSI exhibited signet ring cells and cribriform formation. Gisselsson et al. 34 A study by Gisselsson et al. 35 reported that HeLa cells with knocked-out Rad9 expression treated with alkylating agents resulted in abnormal nuclear morphology. Debes et al. 36 indicated that transfection of p300 into prostate cancer cells in culture could induce quantifiable nuclear changes such as diameter, perimeter, and absorbance. The p300 gene, along with the highly homologous CREB-binding protein (CBP) gene, is mutated in > 85% of microsatellite instability (MSI)+ colon cancer cell lines 37 and loss of heterozygosity at the p300 locus was observed in advanced intestinal-type gastric cancers. 38

[0006] Another structural biomarker associated with tumor progression is the centrosome. MMR defects and genomic instability are closely associated with an increased number of structurally abnormal centrosomes. 39、40 Centrosomes play a vital role in many microtubule-mediated processes, such as determining the shape and polarity of a cell. 41-44 .

[0007] As noted above, morphometric changes based on defects in MMR and MSI have been implicated in multiple types of cancer. While the data provided below establish the ability of the Cell-CT™ platform to perform morphometric genotyping of lung adenocarcinoma cell lines with different driver mutations, the utility of the technology should be available to identify MMR defects and mutational burden in multiple cancers.

[0008] In a related development, Nelson has employed optical tomography techniques for 3D imaging of biological cells, as disclosed in U.S. Patent No. 6,522,775, issued February 18, 2003, entitled "Apparatus and Method for Imaging Small Objects in a Flow Stream Using Optical Tomography," the entire disclosure of which is hereby incorporated by reference. Further significant developments are taught in U.S. Patent No. 7,738,945 to Fauver et al., issued June 15, 2010, entitled "Method and Apparatus for Pseudo-Projection Formation for Optical Tomography" (Fauver '945) and U.S. Patent No. 7,907,765 to Fauver et al., issued March 15, 2011, entitled "Focal Plane Tracking for Optical Microtomography" (Fauver '765), the entire disclosures of Fauver '945 and Fauver '765 are hereby incorporated by reference. Early lung cancer detection technology has been fully developed and commercialized by VisionGate, Inc. of Phoenix, Arizona, to provide measurement advantages that have proven to be a significant improvement over the operating characteristics of conventional morphological cytological analysis.

[0009] Processing in such optical tomography systems begins with sample collection and preparation. For diagnostic applications of lung disease, a patient sample can be collected non-invasively in a clinic or at home. In a clinical laboratory, the sample is processed to remove non-diagnostic material, fixed and then stained. The stained sample is then mixed with an optical gel and the suspension is injected into a microcapillary. As the cells are rotated 360 degrees relative to the image collection optics in the optical tomography system, images of objects (e.g., cells) in the sample are collected. The resulting images comprise a set of extended depth of field images from different viewing angles, referred to as "pseudo projection images." The pseudo projection image set can be mathematically reconstructed using back projection and filtering techniques to produce a 3D reconstruction of the target cells. Having equal or approximately equal resolution in all three dimensions is an advantage of 3D tomographic cellular imaging, particularly for quantitative feature measurement and image analysis. On the basis of the teachings therein, VisionGate, Inc. of Phoenix, Arizona has developed an early lung cancer detection technology to provide the measurement advantage that can potentially greatly improve the operational characteristics of conventional morphological cytological analysis. Published clinical data 45、46 shows that non-invasive sputum analysis using the Cell-CT™ platform can detect early stage lung cancer with high sensitivity (92%) and specificity (95%).

[0010] The 3D reconstructed digital images can then still be used for analysis in order to make quantitative measurements of subcellular structures, molecules or molecular probes of interest. Objects such as biological cells can be stained or labeled with at least one absorptive contrast agent or labeled molecular probe, and the measured amount and structure of this biomarker can yield important information about the disease state of the cell, including but not limited to various cancers, such as lung cancer, breast cancer, prostate cancer, cervical cancer, stomach cancer and pancreatic cancer, and various stages of dysplasia.

[0011] However, until the disclosure herein, there has not been a reliable method for detecting TMB using optical tomography. By providing herein methods and systems for identifying TMB in target cells, patients can benefit from treatment with an immunomodulator (e.g., iloprost) to reduce the risk of developing lung cancer. SUMMARY

[0012] This summary is provided to introduce some concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features of the claimed subject matter, nor is it intended to be used to determine the scope of the claimed subject matter. The present invention presented in this disclosure describes a method for developing one or more morphometric classifiers to identify tumor mutation burden (TMB). TMB has been found to be important for the triage of cancer patients for appropriate cancer treatment. The method provides a non-invasive method of characterizing TMB, which is responsive to tumors at an early stage of tumor development and is independent of the size of the tumor. Thus, the present invention is of significant interest to the practice of developing cancer treatments targeted to the specific characterization of the cancer that a patient can have, thereby allowing more effective cancer management with fewer side effects. BRIEF DESCRIPTION OF DRAWINGS

[0013] While the novel features of the invention are set forth particularly in the appended claims, a better understanding of the organization, and content of the invention, as well as other objects and advantages thereof, will be obtained from the following detailed description and drawings, in which:

[0014] Figure 1 A functional overview of the lung cancer test for sample analysis is schematically shown.

[0015] Figure 2 The basic system components of a 3D optical tomography system used in the lung cancer test system are schematically shown.

[0016] Figure 3 (A) to (C) of show single perspective views of 3D images of adenocarcinoma cells.

[0017] Figure 4 Cilia on lung columnar cells are shown.

[0018] Figure 5 An ROC curve of sensitivity versus specificity for an abnormal cell classifier is shown.

[0019] Figure 6 An example of a classification cascade for identifying specific mutations associated with different cancer types is schematically shown.

[0020] Figure 7 Results of experimental studies showing the area under the ROC (aROC) and the sensitivity and specificity to target cells are listed in a table.

[0021] Figure 8 A flowchart schematically showing an example of a method for developing one or more morphometric classifiers to identify tumor mutation burden (TMB) is shown.

[0022] Figure 9 A flowchart schematically showing an example of a method for developing one or more morphometric classifiers to use tumor mutation burden (TMB) as ground truth.

[0023] Figure 10 A flowchart schematically showing an example of a method of treating a malignant tumor in a human subject using immunotherapy.

[0024] In the drawings, like reference numerals refer to like elements or components throughout. The sizes and relative positions of elements in the drawings attached hereto are not necessarily drawn to scale. For example, the shapes and angles of the various elements can have been exaggerated or distorted in order to illustrate certain features more clearly. Additionally, certain elements can have been arbitrarily moved or repositioned for the sake of illustration and clarity. DETAILED DESCRIPTION

[0025] The following disclosure describes methods of developing one or more morphometric classifiers to identify tumor mutation burden (TMB). Several features of the methods and systems according to exemplary embodiments are set forth and described in the drawings. It will be understood that methods and systems according to other exemplary embodiments can include different or additional processes or features than those shown in the figures. Exemplary embodiments are described herein with respect to optical tomographic cell imaging systems. However, it should be understood that these examples are for the purpose of illustrating principles and that the present invention is not limited thereto.

[0026] The present invention provides an early lung cancer detection system that detects TMB using samples processed by optical tomographic systems that produce equidistant, sub-micron resolution 3D cell images that are then processed by automated feature extraction and classification algorithms to identify abnormal cells with high accuracy. Because abnormal cells are rare and there are many non-diagnostic contaminants, only a system that can perform cell detection with high sensitivity and very high specificity can effectively manage lung cancer detection while ensuring adequate sample.

[0027] Definitions

[0028] In general, the following terms as used herein have the following meanings, unless the context indicates otherwise:

[0029] The use of the word "a" or "an" when used in conjunction with the term "comprising" in the claims or the specification means one or more than one, unless specifically stated otherwise. The term "about" means the recited value plus or minus the range of measurement error, or, if no measurement method is indicated, plus or minus 10%. The use of the term "or" in the claims is used to mean "and / or" unless explicitly indicated to refer to alternatives selected from the group or alternatives mutually exclusive. The terms "comprising," "having," "including," and "containing" (and variations of these terms) are open-ended and, unless otherwise noted, allow for addition of other elements, in the claims.

[0030] Throughout this specification, the use of "one example" or "an example," "one embodiment," "an embodiment," "one implementation," or "an implementation" means that a particular feature, structure, or characteristic described in connection with the example is included in at least one embodiment of the disclosure. Thus, the appearances of the phrase "in one embodiment" or "in one example" or other variations thereof appearing throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular

[0031] "Sufficiency" refers to the content of the sample and defines the limit at which the target cells determine whether enough cell mass has been analyzed.

[0032] As used herein, "calcitriol" is the synthetic (man-made) active form of vitamin D3 (cholecalciferol).

[0033] "Capillary" has its generally accepted meaning and is intended to include transparent microcapillary tubes and equivalents with an internal diameter of generally 500 microns or less, although larger diameters can also be used.

[0034] "Cell" refers to a biological cell, such as a human, mammalian, or animal cell.

[0035] "Cell-CT TM"Platform" refers to the optical tomographic system manufactured by VisionGate, Inc. of Phoenix, Arizona, which incorporates the teachings of the Nelson and Fauver patents cited above and improvements on those teachings. The Cell-CT™ platform is an automated high-resolution 3D tomographic microscope and computing system for imaging flowing cells. In contrast to conventional optical imaging methods, the Cell-CT™ platform can compute 3D cell images with equal spatial resolution in all dimensions (isotropic resolution), making measurements independent of orientation. In addition, the focal plane blurring and view orientation dependence that conventional microscopes typically have are eliminated, providing information content that automatically identifies various types of cells and unambiguously identifies rare abnormal cells in a predominantly normal cell population.

[0036] "CellGazer" refers to a software-based utility program for facilitating the viewing of 2D and 3D images of cells provided by Cell-CT. The result of the cell review is a detailed differential diagnosis of cell types, followed by a LuCED test to determine the final outcome of the case being processed.

[0037] As used herein, "chimeric antigen receptor (CAR)" refers to an artificial T cell receptor (also known as chimeric T cell receptor, or chimeric immunoreceptor) is an engineered receptor that grafts arbitrary specificities onto immune effector cells.

[0038] As used herein, "CIS" has its generally accepted meaning of carcinoma in situ, also known as intraepithelial neoplasm.

[0039] "Depth of field" is the length along the optical axis over which the focal plane can be moved before unacceptable image blurring of a specified feature occurs.

[0040] "Enrichment" refers to the process of extracting target cells from an original sample. This process results in an enriched pellet, whose cells can then be more efficiently imaged on the Cell-CT system.

[0041] As used herein, "immunotherapy" applies to the field of oncology and refers to a method of ameliorating, treating or preventing a malignant tumor in a human subject, wherein the method's action assists or enhances the immune system in eradicating cancer cells, including administration of cells, antibodies, proteins or nucleic acids that activate an active (or effect a passive) immune response to destroy cancer cells. It also includes the combination therapy of biological adjuvants (such as interleukins, cytokines, Bacillus Comette-Guerin, monophosphoryl lipid A, etc.) with conventional therapies for treating cancer, such as chemotherapy, radiation therapy or surgery, acting through activation of the immune system to prevent or destroy the growth of cancer cells and ex vivo and adoptive immunotherapy (including therapies using autologous and / or heterologous cells or immortalized cell lines) acting through activation of the immune system.

[0042] As used herein, "Iloprost" is an immunomodulator comprising a synthetic analog of the prostacyclin PGI2.

[0043] "LuCED® test" refers to the early lung cancer detection test performed using the Cell-CT™ platform, which was developed by VisionGate, Inc. of Phoenix, Arizona, and incorporates the teachings of the Nelson and Fauver patents mentioned above and improvements of those teachings.

[0044] "LuCED® process" refers to the mechanism of 3D cell reconstruction, classification of abnormal cells, and pathological confirmation.

[0045] "LDCT" refers to low-dose computed tomography (CT) radiographic scanning.

[0046] "Object" refers to a single cell, human cell, mammalian cell, item, thing or other entity.

[0047] "Pseudo-projection" includes a single image representing a sampling volume greater than the original depth of field of the optical system, wherein the pseudo-projection image so formed includes an integration from a fixed viewpoint focal plane image volume. The concept of pseudo-projection is taught in Fauver '945.

[0048] "Sample" refers to the complete product obtained from a single test or procedure on a single patient (e.g., a sputum, biopsy or nasal swab submitted for analysis). A sample can be comprised of one or more objects. The results of the sample diagnosis become part of the case diagnosis.

[0049] "ROC" has its accepted meaning of receiver operating characteristic.

[0050] "Specimen" refers to the complete cell preparation ready for analysis, including all or part of an aliquot or sample.

[0051] As used herein, "subject" refers to a human patient.

[0052] "Cells of interest" refers to cells from a sample whose characteristics or counts are of particular interest. For example, in the LuCED test, the cells of interest are normal bronchial epithelial cells. A minimum number of samples must be enumerated during the course of the test for the sample to be considered adequate.

[0053] "Threshold" as used in the context of image processing includes a decision boundary value for any measurable characteristic of a feature. Thresholds can be predetermined or set according to instrument specifications, acceptable error rates, statistical data, or other criteria according to accepted pattern recognition principles.

[0054] "Tumor Mutation Burden" (TMB) refers to the number of somatic, coding, base substitutions and indel mutations that occur per megabase pair of genomic DNA.

[0055] "TNM stage" is used in the context of lung cancer in its generally accepted meaning and refers to the tumor, node, metastasis (TNM) stage defined by medical associations such as The International Association for the Study of Lung Cancer (IASLC).

[0056] Vorinostat is also known as suberoylanilide hydroxamic acid and is commonly used as a histone deacetylase (HDAC) inhibitor in Barrett's esophagus.

[0057] "Voxel" as used in the context of image processing is a volume element on a 3D grid.

[0058] SUMMARY

[0059] REFERENCES Figure 1 FIG. 1 schematically shows an overview of the functionality of a lung dysplasia and cancer test system for sample analysis. The test system 5 includes equipment and methods for sample collection 10 followed by a test for early lung cancer detection 12, such as the LuCED® test. The early lung cancer test 12 also includes equipment and methods for sample staining and enrichment 14, 3D cell imaging 20, 3D cell classification 22, and clinician review of abnormal candidate cells 25.

[0060] If sputum is used, it is typically collected by spontaneous coughing at the patient's home or by induction at a clinic. Other types of sample collection can be performed under clinical conditions, such as a biopsy. The test sample is processed to remove contaminants and non-bronchial epithelial cells, such as by reducing the volume of white blood cells and oral squamous cells. The enriched sample is processed on the Cell-CT™ platform, which digitally images cells in true 3D at equal intervals, sub-micron resolution, as described by Nelson and Fauver, supra. Cancer-related biological features are measured on the 3D cell images and combined into a score for identifying a small number of cells with cancerous features. These cells are then optionally displayed for manual cytologist review using a review station, such as the CellGazer™ review station developed by VisionGate, Inc. of Phoenix, Arizona. The review station provides a visual display that allows the cytologist to view the cell images in 2D and 3D, establishing a definitive normal or abnormal status for particular cell candidates. Three-dimensional (3D) cell classification 22 can be performed using the techniques disclosed below.

[0061] The cell imaging system 20 includes a process implemented by computer software, such as executed by a personal computer interfaced with the optomechanical apparatus, to correct for motion that occurs during image capture. Most cell images emerge from filtered back-projection in a well-reconstructed manner. A computer-implemented algorithm can identify poorly reconstructed cells, which can thus be culled from subsequent processing. One example of such a method for detecting poor reconstruction is taught by Meyer et al. in U.S. Patent No. 8,155,420, issued April 10, 2012, entitled "System and Method for Detecting Poor Quality in 3D Reconstructions," the disclosure of which is incorporated herein by reference.

[0062] Early attempts to develop lung cancer screening procedures were based on sputum cytology, which showed poor sensitivity (approximately 60% on average) but excellent specificity (Schreiber and McCrory (2003) Chest 123 (1 ed.): 115). This experience led some to conclude that sputum was of no value in detecting lung cancer. Careful analysis of sputum involving paraffin embedding (Booking A, Biesterfeld S, Chatelain R, Gien-Gerlach G, Esser E., Diagnosis of bronchial carcinoma on sections of paraffin-embedded sputum. Sensitivity and specificity of an alternative to routine cytology, Acta Cytol. 1992; 36(1): 37-47) showed that samples actually contained abnormal cells in 86% or more of cancer patients. Collection during morning coughs over three consecutive days yielded the best results. Further analysis revealed the presence of abnormal cells in sputum stratified by all relevant clinical factors, including tumor histological type, size, stage, and location (Neumann T, Meyer M, Patten F, Johnson F, Erozan Y, Frable J et al., Premalignant and Malignant Cells in Sputum from Lung Cancer Patients, Cancer Cytopathology, 2009; 117(6): 473-481). Based on these sample characteristics, the currently published lung cancer detection test uses spontaneous cough sputum. Initial assessments showed satisfactory results using Cytoyt (Hologic, Marlborough, MA) or the well-known Saccomanno method for sputum fixation. The issue of sample adequacy is also important for sputum cytology. Attempts to increase the amount of sputum have also yielded varying degrees of success. Sputum induction can increase sputum production to help obtain a sufficient whole sample.

[0063] Examples of sample enrichment and preparation

[0064] A lung cancer detection test applicable to TMB detection Examples In this example, the sample underwent three processing stages before analysis: 1) cell isolation and cryopreservation; 2) enrichment by fluorescence activated cell sorting (FACS); and 3) embedding the enriched cells into an optical oil that matches the refractive index of the optical components of the optical computed tomography imaging system.

[0065] Cryopreservation and FACS enrichment (FACS is an example)

[0066] Sputum was treated with the mucolytic dithiothreitol (DTT) (Fisher Scientific, Waltham, MA). In one example, for long-term preservation, the sample was filtered through a 41 μm nylon mesh and maintained at -80°C in 15% dimethyl sulfoxide (DMSO) (Fisher Scientific, Waltham, MA). After filtration, aliquots of up to 100 μL were taken for lung cancer detection assay analysis. First, sputum cells were stained with hematoxylin (Electron Microscopy Sciences, Hatfield, PA) for downstream lung cancer detection assay imaging. Then, the cells were treated with an antibody mixture containing fluorescent conjugates selected for enriching bronchial epithelial cells and depleting contaminated inflammatory cells (neutrophils and macrophages). An anti-cytokeratin-FITC conjugate mixture (Cell Signaling, Danvers, MA) targeted cytokeratins expressed in both normal and malignant epithelial cells. Anti-CD45-APC conjugates (Mylteni, Bergisch Gladbach, Germany) were used as negative selectors for inflammatory cells. Cells were also stained with DAPI (Life Technologies, Grand Island, NY) prior to sorting. For FACS enrichment, a DAPI-positive mother gate was created to exclude bimodal cells and debris, followed by exclusion of high-side-scatter events primarily in oral squamous cells. Subsequently, sub-gates for high cytokeratin (high FITC) and low CD45 (low APC) were plotted. The cell populations in these sub-gates were enriched target epithelial cells and sorted using an optical computed tomography system (e.g., Cell-CT® optical computed tomography system) for more efficient downstream lung cancer detection assays.

[0067] Encapsulation of enriched cells

[0068] Following FACS enrichment (or any other enrichment process), the cells are dehydrated in ethanol and then suspended in xylene. The cells are then transferred to and embedded in a suitable volume of optical medium, which is a viscous oil with a refractive index matched to the optical tomography system. Once embedded, the cells are injected into a disposable cartridge for imaging on the optical tomography system.

[0069] Now for reference Figure 2, showing the basic system components of a 3D optical tomographic imaging system used in the lung cancer test system. The cell imaging system 20 is an automated high resolution 3D tomographic microscope and computing system for imaging cells in flow. It includes an illumination source 90 optically coupled to a condenser 92 that optically cooperates with an objective 94 to scan an image of an object 1 contained in a capillary 96. The image is obtained by scanning the volume occupied by the object with an oscillating mirror 102 and transmitted through a beam splitter 104 to a high speed camera 106. The high speed camera produces a plurality of pseudo projection images 110. A set of pseudo projection images for a plurality of axial tube rotational positions is produced for each object.

[0070] Although the test system is not limited to any one contrast method, in one example, the lung cancer detection test is specifically targeted to cell morphology based on traditional use of hematoxylin stain. In the lung cancer detection test application, the optical tomographic system computes 3D cell images in a manner that has the same resolution in all dimensions (i.e., isotropic resolution), making the measurements independent of direction. In addition, the focal plane blurring and view orientation dependence that conventional microscopes typically have are eliminated, providing information content that automatically identifies various types of cells and unambiguously identifies rare abnormal cells in a predominantly normal cell population. The output of the optical tomographic system identifies about 0.5% of all cells as abnormal candidates for verification using a CellGazer™ (VisionGate, Phoenix, AZ) workstation, which is an imaging software tool that enables human viewing of images without focal plane and directional blurring.

[0071] Optical tomographic system imaging is performed in a small amount of liquid suspension. For the lung cancer detection test, these cells are from the enriched epithelial cell population described above. Because the optical tomographic system can separate tightly coherent objects, a narrow focused core of single file cell flow is not required, despite the requirement of standard flow cytometry.

[0072] Example operation of a lung cancer test system is described in the above-referenced Nelson and Fauver articles and other patents, including U.S. Patent No. 8,254,023 issued August 28, 2012 to Watson et al. entitled "Optical Tomography System with High-Speed Scanner," which is also incorporated herein by reference. In operation, a stained nucleus of a biological cell 1 is suspended in optical medium 112 and injected into a capillary tube 96 with an inner diameter of, for example, 62 μιη. The capillary system is designed to be disposable, thus eliminating the possibility of cross-contamination between samples. Pressure 114 applied to the fluid moves the object 1 into the imaging position, and 3D data is then acquired as the tube is rotated. The mirror 102 is actuated to sweep the focal plane across the object, and the images are integrated by the camera to produce pseudo-projections from each single view angle. A glass holder that connects the capillary tube 96 to the optical tomography system is not shown. The holder has a hole in the middle that is slightly larger in diameter than the outer diameter of the capillary tube and a glass flat (not shown for simplicity of illustration) to optically couple with the objective and condenser lenses. The capillary tube, loaded with cells embedded in a transport medium, is threaded through the holder. The transport medium that holds the cells, the glass capillary tube, the capillary holder, the oil that interfaces with the mirror, and the mirror itself are all made of materials with the same optical index. Thus, as the capillary tube is rotated 360 degrees, the light rays pass through the optics of the optical tomography system, the capillary tube, and the cells without refraction as the cells can rotate to allow a set of 500 pseudo-projections to be captured. Since the cells are suspended in a liquid medium, they are prone to a small amount of motion as the pseudo-projection images 110 are acquired.

[0073] Thus, the cell images in the pseudo-projections must be registered to a common center so that the cell features enhance each other during the reconstruction process. U.S. Patent No. 7,835,561 entitled "Method for Image Processing and Reconstruction of Images for Optical Tomography" discloses a correction technique for the pseudo-projections. U.S. Patent No. 7,835,561 is incorporated herein by reference. The corrected set of pseudo-projections is processed using a filtered back-projection algorithm similar to that used in conventional X-ray CT to calculate the tomographic 3D cell reconstruction. Pseudo-projection images 110 taken at three angular positions (0g, 90g, and 180g) are shown. The light source 90 provides illumination at a wavelength of 585 nm to optimize image contrast based on the hematoxylin absorption spectrum. In the reconstruction, the 3D pixels or voxels are cubic with a size of approximately 70 nm in each dimension. The size of the reconstructed volume will vary as the image acquisition volume will be cropped around the object. Typically, the volume on the sides is about 200 to 300 pixels.

[0074] Reference is now made to Figure 3 Figures (A) through (C) showing perspective views of 3D images of adenocarcinoma cells. Figure 3 Figure (A) of （13） Since the gray scale values in the 3D image are associated with various cellular features, a look-up table was created that maps cellular structures to color and opacity values to generate cell images in the center (as shown in Figure (B) of Figure 3 and on the right (as shown in Figure (C) of Figure 3 In color reproductions of these images, the semi-transparent white 402 represents cytoplasm, the opaque blue 404 represents the nucleus, the semi-transparent green 406 represents loose chromatin and karyoplasm, and the condensed chromatin and nucleolus are represented by the opaque red 408. In international patent regulations where only black and white drawings are provided, these colors are identified by the border (shown as a dashed border) identified by the corresponding reference numbers 404, 406, and 408.

[0075] Reference is now made to Figure 4 showing cilia on lung columnar cells. The imaged normal bronchial epithelial cells have single cilia chains with a diameter of about 250 nm. This further demonstrates the resolution of the 3D cell imaging system.

[0076] Reference is now made to Figure 5 showing a ROC curve for the abnormal cell classifier. The ROC curve 700 is a plot of the sensitivity to dysplastic cells on the vertical axis 701 versus the specificity on the horizontal axis 703. The point 707 represents the area where the dysplastic cell classifier performs with a sensitivity of 75% at near 100% specificity. The classifier was built using a data set containing cells indicative of abnormal lung processes, including moderate to severe dysplasia and certain atypical cell conditions. The training of the classifier was performed using a set of approximately 150 known dysplastic cells and approximately 25,000 known normal cells. The single cell ROC curve 700 demonstrates the accuracy, which shows near perfect detection of dysplastic cells. The accuracy of the classifier is generally represented as the area under the ROC curve (AROC). When the AROC is 1, perfect discriminative power is obtained. The LuCED AROC value is 0.991. For single cell detection, the selected operating point can provide a sensitivity of 75% and a specificity of 100%. The cell classification is related to the detection of cases, as shown in the following table. For example, if an abnormal cell is encountered during the LuCED analysis, the probability of case detection will be 0.75 or 75%. If two abnormal cells are encountered by LuCED, then the probability of case detection will be (1-(1-0.75) 2= 0.9375 or close to 94% case sensitivity, etc.

[0077] 1 cell - 75% case sensitivity,

[0078] 2 cells - 94% case sensitivity, and

[0079] 3 cells - 98% case sensitivity.

[0080] Reference is now made to Figure 6 , which shows an example of a classification cascade used to train classifiers adapted to recognize specific mutations associated with different cancer types. Training is performed to produce a series of binary classifiers to isolate the desired cells, including a first classifier 602, a second classifier 604, a third classifier 608, a fourth classifier 609, a fifth classifier 611, and a sixth classifier 615.

[0081] In one example, the first classifier 602 is trained to separate malignant cells from other normal cells. The first classifier 602 groups all data from the malignant cell lines and assigns it to one class (e.g., a group of malignant cells). This group of malignant cells, plus normal cells as a negative control, are used to train the first classifier to separate normal cells from malignant cells. This step is particularly important since malignant cells are rarely seen in sputum. During the training process, only a small fraction of the cells in the sputum are manually inspected. Since the manual inspection is part of the process, it can be assumed that only the abnormal cells that arise from the process are truly malignant cells, which can then be subtyped using the classifiers described below.

[0082] The second classifier 604 separates malignant subtypes. Any organ system has different types of tissue associated with it. For example, lung tissue includes squamous epithelial and adenocarcinoma tissue of the bronchus. Small cell lung cancer (SCLC) cells from neuroendocrine glands are also sometimes apparent. Thus, a classifier is needed to separate the particular cancer subtype in which the desired driver mutation occurs. This is done by first separating small cell lung cancer from adenocarcinoma and squamous carcinoma, and then separating adenocarcinoma from squamous carcinoma. Further separation of the desired mutation subtype in adenocarcinoma is done in steps. The grouping of cell lines for the training set selected for this example is given in Table 1 below. Separation of specific driver mutations is determined according to morphological factors in the third through sixth classifiers 608, 609, 611, and 615.

[0083]

[0084] Still referring to Figure 6In one example, the mutation-driven stepwise separation begins with the first classifier 602 in which a set of cells is separated into normal and malignant or dysplastic categories. Any cells identified as malignant are further processed in the second classifier 604 which separates SCLC: NCI-H69 type cells from other malignant cells and passes to the third classifier 608. The third classifier 608 separates Adeno: SW900 from other adenocarcinoma type cells and passes to the fourth classifier 609 other cells. The fourth classifier 609 separates Adeno: ALK+, NCI-H2228 cell type from other remaining cell types and passes the remaining cell types to the fifth classifier 611. The fifth classifier 611 separates Adeno: Wild type, A549 from EGFR+ adenocarcinoma cell types and passes the EGFR+ adenocarcinoma subtypes to the fifth classifier 615. The sixth classifier 615 separates Adeno: T790M, NCI-H1975 from Adeno: EGFR-p.E746_A750del.

[0085] Those skilled in the art will recognize that this is merely one example of the application of the present invention and that other cell types and mutation drives can be used to construct and train classifiers including TMB classifiers according to the methods described herein. The present invention is in no way limited to this example. Classifier decisions are achieved by establishing decision boundary values for any measurable property of a feature during the classifier training process. Thresholds can be selected or set according to instrument specifications, acceptable error rates, statistical data or other criteria according to accepted pattern recognition principles.

[0086] Experimental results

[0087] Reference is now made to Figure 7 , which summarizes the results of an experimental study. Table 650 indicates the area under the ROC (aROC) 652 for the target cells classified by the classifiers trained according to the training methods described above as well as the sensitivity 654 and specificity 656. Specificity relates to the misidentification of malignant cells by a classifier intended to separate a particular driver mutation. For example, the specificity of identifying cells from small cell lung cancer (SCLC) tumors is 99.98%. A small number of cells referred to as SCLC actually came from some other cell line listed in Table 650. Since only 0.02% of the malignant cells were misidentified, the positive predictive value can be calculated to identify SCLC as PPV = TP / (TP + FP) = 100 * 0.748 / (0.748 + 0.002) = 99.7.

[0088] The excellent discrimination between normal and abnormal cells in the evidence for the LuCED® process, coupled with the published evidence showing that morphological changes in malignant cells are associated with the genomic characteristics of the cells, suggests that genetic mutations responsible for driving the cancer process can be identified by purely morphological means 52 .

[0089] In the present disclosure, the concept of morphometric genomics is extended and used to detect cancer drivers into the domain of tumor mutational burden. Cell-CT TM The system is used to generate a morphometric basis for tumor mutational degree, the idea being to provide a non-invasive means of characterizing TMB. The LuCED® algorithm detects cancer with the same sensitivity regardless of the histology, stage, and size of the tumor 46 . Thus, a Cell-CT based TMB measurement would have the potential to non-invasively characterize TMB regardless of the histology, stage, and size factors.

[0090] Reference is now made to Figure 8 , which shows a diagram of an embodiment of a method for developing one or more morphometric classifiers to identify tumor mutational burden (TMB). The method includes the following actions:

[0091] Obtaining selected clones from transduced cells830;

[0092] Analyzing MLH1 expression of selected clones to screen for clones with comparable levels to the parental cell line832;

[0093] Expanding selected clones in culture medium842;

[0094] Harvesting selected clones843;

[0095] Determining TMB levels of harvested clones850;

[0096] Analyzing selected clones on a 3D microscope optical tomography system852; and

[0097] Comparing selected clones to a set of control cell lines854.

[0098] In one example, the set of control cell lines includes the parental NCI-H23 and clones expressing scrambled shRNA. Further, the analysis action is to generate a plurality of morphometric biofeatures per cell. TMB data can be determined by genomic analysis using whole exome sequencing or targeted exome sequencing using NGS or targeted genome.

[0099] Reference is now made to Figure 9, showing a flowchart of an example of a method for developing one or more morphometric classifiers to use tumor mutation burden (TMB) as a ground truth. The method can be used in conjunction with the methods described above with respect to Figure 8 Low TMB and high TMB can be used as ground truths for developing cell classifiers for each cell in an isogenic cell line and determining the ROC area for each cell classifier 930. A score that matches the ground truth is defined for each cell in the isogenic cell line 932. A classifier for producing a score that closely matches the ground truth is trained by using an adaptive boosting logistic regression algorithm to define a set of projection axes used by a logistic function to produce a score from 0 to 1. The adaptive boosting logistic regression algorithm can be iterated by using successive trials that weight each observation by the difference between the ground truth and the current score to adaptively converge to a solution that progressively transforms a set of more extensive cell characterizations into a solution. Alternatively, analyzing the selected clones can include using a random forest algorithm to produce a classifier using non-parametric assumptions for the feature distribution. Further, an evaluation of the classifier discrimination can be improved by pruning a set of possible feature trees to optimize discrimination 950. The area under the ROC curve aROC is used to judge classifier efficacy, where the area under the receiver operating characteristic curve or aROC is computed by integrating the ROC curve, which represents the overall performance of the binary classifier output in terms of classification sensitivity and specificity 952. A threshold is established to be used with the classifier score to create a binary output that is associated with the ground TBM with high precision 954. A numerical score can be produced that represents the probability that a cell belongs to a target class. The target cells can be separated from non-target cells by further binarizing the score by applying a threshold to the distribution 1030. In one useful example, a threshold can be determined to provide a precision of 0.95 or higher to separate cells with low TMB from cells with high TMB.

[0100] Reference is now made to Figure 10 , showing a flowchart of an example of a method of treating a malignant tumor in a human subject using immunotherapy. The method includes the following actions:

[0101] analyzing a 3D image of a cell based on a pseudo-projection obtained from a sample obtained from the subject 1030;

[0102] operating a biological sample classifier to identify normal or abnormal cells from the sample 1032;

[0103] determining a TMB score from each abnormal cell 1042;

[0104] applying a predetermined threshold to the TMB score 1052;

[0105] When cancer is discovered, surgery should be performed to remove the cancerous lesion;

[0106] When the TMB score exceeds a predetermined threshold, the human subject is assisted as a candidate for immunotherapy by administering an immunomodulator to the subject for a predetermined period of time to assist the subject's immune system in destroying the cancer cells 1054. The immunomodulator can advantageously be a drug selected from the group consisting of a chimeric immune receptor, a prostacyclin analogue, iloprost, a chimeric antigen receptor (CAR) of T cells, vorinostat, an HDAC inhibitor, cholecalciferol, calcitriol, and combinations thereof.

[0107] Data to support this concept is currently being developed. The approach will entail the following actions:

[0108] 1. Bailis et al. 53 (PLoS One. 2013 Oct 29;8(10):e78726) knocked out expression of the MLH1 gene in NCI-H23 lung adenocarcinoma cells to generate isogenic lines for direct comparison of MMR-proficient cells and MMR-deficient cells. The Bailis et al. paper reports that after several weeks in culture, the MMR-deficient cells exhibited microsatellite instability, a common mechanism by which cancer cells acquire high TMB. A similar approach can be used by transducing NCI-H23 cells with MLH1 shRNA lentivirus particles (Santa Cruz Biotechnology, sc-35943-V) or scrambled shRNA lentivirus particles (Santa Cruz Biotechnology, sc-108080) and selecting for integration using puromycin. MLH1 shRNA-transduced clones derived from single puromycin-resistant cells can be analyzed by Western blot analysis to screen for clones with > 90% reduction in MLH1 expression. Clones can also be derived from control scrambled shRNA-transduced cells and analyzed for MLH1 expression to screen for clones with levels comparable to the parental NCI-H23 cell line. The selected clones can be expanded in culture by weekly aliquots of harvest and fixed in an ethanol-based fixative for further analysis. The fixed cells can be analyzed using the FoundationOne assay (Foundation Medicine, Inc.) to determine TMB levels, which generates a comprehensive genomic profile of over 300 cancer-related genes. Once cell lines with different TMB levels are identified, they can be analyzed on the VisionGate Cell-CT platform and compared to control cell lines (parental NCI-H23 and scrambled shRNA-expressing clones) to verify the accuracy of the identified cells.

[0109] 2. Process on the Cell-CT™ platform to generate 704 morphometric biofeatures per cell for each isogenic cell line - this can be done using the 3D microscope optical tomography system platform, using the training techniques and functionality described herein to image process and cell classify to accomplish this.

[0110] 3. Using low TMB and high TMB as ground truth, a cell classifier can be developed for each isogenic cell line and the ROC area determined for each classifier. Typically, this process involves defining a score that matches the ground truth for the cell in question. In this application the ground truth is defined by point 2 above as high TMB and low TMB. The classification process aims to produce a score that is very close to the actual ground truth. There are several methods that can be used to accomplish this, including:

[0111] a. Adaptive boosting logistic regression 50 This method uses principal component projection to define a projection axis and then uses this projection axis through a logit function to produce a score between 0 and 1. The algorithm iterates through successive trials using the difference between the ground truth and the current score to weight each observation. This adaptive process converges to a solution that gradually transforms a more extensive set of cell characterizations into a solution.

[0112] b. Random forest 51 In this method, a classifier is generated using a non-parametric assumption about the distribution of the features. One limitation of adaptive boosting is that it assumes a distribution of the features behind the principal component processing. This is a potential problem because the features can not strictly conform to the assumed distribution, resulting in inaccurate projections. In this method, a random vector of random length is defined. The discriminability is evaluated and the set of potential feature trees is pruned to optimize the discriminability.

[0113] 4. The area under the ROC curve, aROC, can be used to judge the efficacy of a classifier. The area under the receiver operating characteristic curve or aROC is calculated by integrating the ROC curve, which represents the overall performance of the binary classifier output in terms of classification sensitivity and specificity. The term "sensitivity" refers to the ability of the classifier to correctly classify objects that have the attribute (or set of attributes) that the classifier was trained to detect as "target" or positive objects. Similarly, specificity represents the ability of the classifier to correctly classify objects that do not have the target attribute as "non-target" or "negative". Both sensitivity and specificity can range from 0 to 1, and it is desirable to have a classifier to perform, and both parameters to be as close to 1 as possible. Although the classifier generates a continuous number (score, i.e. the probability that the object belongs to the target class) as output for each object (in our case, cells), the output can be further binarized by applying a threshold to the score that separates the positive and negative classes. Once the classifier is developed, the ROC curve can be generated by calculating the sensitivity and specificity values as a function of the threshold value that is used to separate the two classes of objects. The aROC value can range from 0 to 1, and the aROC value represents the percentage of true positive objects and true negative objects that are correctly classified by the classifier. In one example, the true positive objects are cells with high TMB, and the true negative objects are cells with low TMB. Thus, aROC > 0.95 means that the classifier will correctly classify more than 95% of all cells with either low TMB or high TMB.

[0114] 5. Establish a threshold value to use with the classifier scores to create a binary output that is highly correlated with the underlying TBM.

[0115] As described in the previous section, the classifier will typically produce a numerical score that represents the probability that a cell belongs to the target class. In order to separate the target cells from the non-target cells, the score is further binarized by applying a threshold to the score distribution. Typically, scores above the threshold are considered "positive" or target cells, while objects with scores below the threshold are considered "negative" or non-target cells. Since the value of the classifier threshold can vary across the entire range of the score distribution, the final metric of choosing the appropriate numerical value is the highest accuracy of the classifier to correctly distinguish (classify) the cells. In our case, the threshold will be determined to provide an accuracy of 0.95 or higher to separate cells with low TMB from cells with high TMB. The TMB data will be determined by genomic analysis performed using whole exome sequencing or targeted exome sequencing using NGS or targeted genome. The high TMB values and low TMB values will be determined based on the TMB distribution. Typically, TMB above 80% is considered high, although other metrics can be applied depending on the distribution characteristics.

[0116] Classifier training - inputs and methods

[0117] The creation and optimization of a cell detection classifier is generally referred to as "classifier training" as the process aims to accurately diagnose cells according to a reference or ground truth. Using the classification methods described herein, cells can be classified into types including but not limited to normal, cancerous, and dysplastic. There are two main aspects of accuracy: the first is specificity (the classifier calls normal cells normal), and the second is sensitivity (the classifier calls abnormal cells abnormal). Algorithm training methods include adaptive boosting logistic regression and random forest. Those skilled in the art will be familiar with how to apply other classical training techniques to a classifier, such as template methods, adaptive processing, and the like.

[0118] The method for training the classifier ensures very good results in cases where data is used as input. First, when the inputs to the classifier training process accurately describe the clinically relevant aspects of the cells and are robust to environmental factors that can affect the results of the optical tomography system, the classifier accuracy can be ensured:

[0119] 1. The three-dimensional cell images generated by the Cell-CT™ optical tomography system have high resolution, allowing for accurate measurement of key features that support correct classification.

[0120] 2. Certain functions useful in classification only appear in 3D images. Therefore, compared to 2D imaging, the 3D feature set can not only better describe the cells, but also make the classification based on three-dimensional imaging more accurate.

[0121] 3. Three-dimensional image segmentation algorithms have been developed to separate the entire cell from the background, and the cell nucleus from the cell. The accuracy of these segmentation algorithms was verified by comparing the segmented traces to human-derived cell or nuclear envelope traces.

[0122] 4. Feature measurements describe various aspects of the cell, cell nucleus, cytoplasm, and nucleolus. In one example of a test system, 594 features were calculated for each 3D cell image, representing the shape, volume, chromatin distribution, and other more subtle morphological elements of the object. The calculation of these features has been proven to be independent of the orientation of the cell.

[0123] 5. The diagnostic truth (the gold standard of pathology) used for classifier training is typically based on the graded cell diagnosis provided by two cytotechnologists and one cytopathologist.

[0124] Classifier training - statistical considerations

[0125] Second, in a test conducted by the inventors herein, the accuracy of the classifier training process was ensured through a rigorous process that covered three aspects:

[0126] 1. The database used to train the classifier is formulated to contain enough material to ensure that the 95% confidence interval of the binomial formula keeps the variance of the performance estimate within an acceptable range.

[0127] 2. Overtraining is a potential pitfall in the training process, where the classifier may contain too much information, making the data used for training overly specialized. This leads to an overly optimistic estimate of the classifier's performance. The risk of overtraining can be mitigated through cross-validation, which involves taking a portion of the training data and using it as test data. The limit of usable information in the classifier is reached when the performance estimate based on the training data exceeds the estimate based on the test data.

[0128] 3. Finally, as a further safeguard against overtraining, the classifier was tested using data from a second set of cells that were not part of the training process.

[0129] Overview of Abnormal Cell Classifier Training

[0130] The following factors are considered in defining the parameters that control the training of the abnormal cell classifier:

[0131] 1. Due to the scarcity of abnormal cell samples and the abundance of non-diagnostic factors, the classifier must operate with high sensitivity and very high specificity. As shown in Table 2, a high case detection sensitivity can be maintained when the sensitivity of a single-cell classifier is 75% and the sample contains more than one abnormal cell.

[0132] 2. To ensure that the workload remains within a reasonable range, the specificity target is set at 99%.

[0133] 3. 95% confidence interval for the lower binomial (21) The interval should be maintained above 70% sensitivity and 98.5% specificity.

[0134] Ultimately, a high detection rate is desired for each positive case. The sensitivity of single-cell testing translates into the detection of abnormalities, as shown in Table 2.

[0135] Table 2

[0136]

[0137] The meaning of Table 2 is important for lung cancer detection testing. The results shown in the table indicate that if abnormal cells are present in the group analyzed by the lung cancer detection test, these abnormal cells can be detected with confidence, thus allowing for highly sensitive identification of the case. The remaining issue is the presence of abnormal cells in the lung cancer detection test analysis, which is the remaining factor determining the cancer detection rate.

[0138] Development and features of classifiers

[0139] Generally, features are computed to provide numerical representations of various aspects of the 3D tomograms. The computed features are used in conjunction with expert diagnoses of the objects to develop classifiers that can distinguish between object types. For example, a data set can be computed for objects of a second type, type 2, such as normal and abnormal cells, having M 3D tomograms and N 3D tomograms computed for objects of a first type, type 1. Here "M" and "N" represent the number of type 1 and type 2 values, respectively. The data set is preferably generated by an optical tomographic system. The optical tomographic system provides 3D tomograms that include 3D images of the objects, e.g., cells. The cells typically include other features, e.g., a nucleus having organelles such as a nucleolus. The object types can include different types of cells, organelles, cells exhibiting a selected disease state, probes, normal cells, or other features of interest, e.g., features related to TMB. A set of x 3D image features is computed based on the 3D tomograms of all M + N objects. Next, a refined set of y 3D image features that best distinguish between the object types is found, where "x" and "y" represent the number of 3D image features at each stage. The refined set of y 3D image features is used to construct a classifier whose output is related to the object type. In one exemplary implementation, at stage 102, a set of 3D tomograms is assembled, where the assembled set represents substantially all of the important markers that would be used by an expert to distinguish between 3D biological object types. After a representative set of 3D tomograms is assembled, a 3D image feature set can be computed for each object that characterizes the important markers.

[0140] Features

[0141] Tomograms of biological objects, such as cells, exhibit a variety of observable and measurable features, some of which can be used as features for classification. Table 3 below provides a summary of features, i.e., important markers, that are used to facilitate classification goals.

[0142] Table 3

[0143] Features

[0144]

[0145] By way of further explanation, in one useful example, it has been found that the presence of lacunae in a 3D biological object is a useful classification feature based on a measurement criterion, including comparison to a computed or selected threshold. Another feature related to lacunae can include the number of lacunae in the object. Another feature related to lacunae includes the volume of lacunae or the number of lacunae. Another feature includes the surface area of lacunae or the number of lacunae. The shape and location of inter-nuclear lacunae can also be used as useful features. In addition, combinations of feature characterizations can also be used to construct classifiers as described above.

[0146] Similarly, it has been found that invaginations in a 3D biological object are a useful classification feature based on a measurement criterion, including comparison to a computed or selected threshold. Another feature related to invaginations can include the number of invaginations in the object. Another feature related to invaginations includes the volume of invaginations or the number of invaginations. Another feature includes the size of invaginations or the number of invaginations. The location of intra-nuclear invaginations also includes useful features. In addition, combinations of feature characterizations can also be used to construct classifiers as described above.

[0147] It has been found that invaginations in a 3D biological object are a useful classification feature based on a measurement criterion, including comparison to a computed or selected threshold. The volume, surface area, shape, location of lacunae connected to invaginations, and combinations of invagination features can also be advantageously used to construct classifiers as described above.

[0148] It has been found that nucleoli present in a 3D biological object are a useful classification feature based on a measurement criterion, including comparison to a computed or selected threshold. As described above, the volume, surface area, shape, location of objects that can be nucleoli or condensed chromatin, and combinations of the above features can also be advantageously used to construct classifiers. It has now been found that nucleolar texture features present in a 3D biological object are useful classification features. Using various sized structuring elements, fuzzy erosion techniques can be used to isolate features of various sizes within the nucleus. Fuzzy erosion techniques generally require the use of a filter to blur the image and measure the resulting fuzzy erosion by applying a labeling operation. The total 3D volume, the number of discrete components, the volume histogram, the mean volume and variance, and the shape histogram are then computed.

[0149] It has been found that distance metrics that describe the spatial relationships between nucleoli, invaginations, lacunae, and the nuclear envelope are useful classification features. For example, if the mean and variance of the three nucleoli are found, then the minimum and maximum inter-nucleolar distances can be found. Likewise, the distance between the average coordinates of a cluster of nucleoli and the mass center of the entire object can be found. Similar calculations can be made by replacing any of the above entities with the mass center of the nucleoli and the nucleus.

[0150] It has now been found that the Fast Fourier Transform (FFT) features are useful classification features. The FFT features are formed from the Fast Fourier Transform of the 3D tomogram. The FFT features represent both prominent and average representations of the FFT classification.

[0151] The application has been described herein in considerable detail, in order to comply with patent laws and to provide those skilled in the art with the information needed to apply the novel principles and to construct and use such exemplary and preferred embodiments of the application as claimed. However, it is to be understood that various modifications can be accomplished without departing from the true spirit and scope of the application.

[0152] The disclosures of the following publications are incorporated herein by reference.

[0153] 1. Reddy K., Feinberg A. Higher order chromatin organization in cancer; Seminars in Cancer Biol. 2013:23:109-115.

[0154] 2. Mattout A., Cabianca D., Gasser S., Chromatin states and nuclear organization in development - a view from the nuclear lamina. Genome Biol., 2015; 16:174.

[0155] 3. Zink, D., Fischer, A., Nickerson J. Nuclear structure in cancer cells. Nat. rev.; Cancer. 2004:4:677-687.

[0156] 4. Polak, P., Karlic, R., Koren, A., Thurman, R., Sandstrom, R., Lawrence, M., Reynolds, A. et al., Cell-of-origin chromatin organization shapes the mutational landscape of cancer. Nature, 2015; 518:360-364.

[0157] 5. Schuster-Boeckler B., Lehner, B. Chromatin organization is a major influence on regional mutation rates in human cancer cells. Nature, 2012; 488: 504-507.

[0158] 6. Pardoll DM. The blockade of immune checkpoints in cancer immunotherapy. ; Nat Rev Cancer. 2012 Mar 22; 12(4): 252-64.

[0159] 7. Robert C, Long GV, Brady B, Dutriaux C, Maio M, Mortier L, et al. Nivolumab in previously untreated melanoma without BRAF mutation. N Engl J Med 2015; 372:320-30.

[0160] 8. Weber JS, D'Angelo SP, Minor D, Hodi FS, Gutzmer R, Neyns B, et al.; Nivolumab versus chemotherapy in patients with advanced melanoma who progressed after anti-CTLA-4 treatment (CheckMate 037): a randomised, controlled, open-label, phase 3 trial. Lancet Oncol 2015; 16:375-84.

[0161] 9. Borghaei H, Paz-Ares L, Horn L, Spigel DR, Steins M, Ready NE, et al. Nivolumab versus docetaxel in advanced nonsquamous non-small-cell lung cancer. N Engl J Med 2015; 373:1627-39.

[0162] 10. Gangadhar, Tara C., Vonderheide, Robert H. Mitigating the toxic effects of anticancer immunotherapy, Nat. Rev. Clin. Oncol. 2014; 11(2): 91-99.

[0163] 11. Byun, D., Wolchok, J., Rosenberg, L. M., Girotra, M., Cancer immunotherapy-immune checkpoint blockade and associated endocrinopathies. 2017; 13: 195-207.

[0164] 12. Abdel-Wahab, N., Alshawa, A., Suarez-Almazor, M. Adverse effects in cancer immunotherapy. Adv. Exp. Med. Biol., 2017; 995: 155-174.

[0165] 13. Hodi FS, O'Day SJ, McDermott DF, Weber RW, Sosman JA, Haanen JB, et al. Improved survival with ipilimumab in patients with metastatic melanoma. N Engl J Med 2010; 363:711-23.

[0166] 14. Motzer RJ, Escudier B, McDermott DF, George S, Hammers HJ, Srinivas S, et al. Nivolumab versus everolimus in advanced renal-cell carcinoma. N Engl J Med 2015; 373:1803-13.

[0167] 15. Voong KR, Feliciano J, Becker D, Levy B. Beyond PD-L1 testing-emerging biomarkers for immunotherapy in non-small cell lung cancer. Ann Transl Med. 2017 Sep; 5(18):376.

[0168] 16. Danilova L, Wang H, Sunshine J, Kaunitz GJ, Cottrell TR, Xu H, Esandrio J, Anders RA, Cope L, Pardoll DM, Drake CG1, Taube JM. Association of PD- 1 / PD-L axis expression with cytolytic activity, mutational load, and prognosis in melanoma and other solid tumors. Proc Natl Acad Sci U S A. 2016 Nov 29; 113(48):E7769-E7777.

[0169] 17. Kefford R., Ribas A., Hamid O., et al. Clinical efficacy and correlation with tumor PD-L1 expression in patients with melanoma treated with the anti-PD-1 monoclonal antibody MK-3475. Journal of Clinical Oncology. 2014;32([supplement: abstr 3005]).

[0170] 18. Carbognin L., Pilotto S., Milella M., et al. Differential activity of nivolumab, pembrolizumab and MPDL3280A according to the tumor expression of programmed death-ligand-1 (PD-L1): sensitivity analysis of trials in melanoma, lung and genitourinary cancers. PLoS ONE. 2015;10(6).

[0171] 19. Muro K., Bang Y., Shankaran V., et al. Relationship between PD-L1 expression and clinical outcomes in patients (Pts) with advanced gastric cancer treated with the anti-PD-1 monoclonal antibody pembrolizumab (Pembro; MK-3475) in KEYNOTE-012. Journal of Clinical Oncology. 2015;33([supplement; abstr 3]):3-3.

[0172] 20. Madore J., Vilain R. E., Menzies A. M., et al. PD-L1 expression in melanoma shows marked heterogeneity within and between patients: Implications for anti-PD-1 / PD-L1 clinical trials. Pigment Cell and Melanoma Research. 2015;28(3):245-253.

[0173] 21. Mitchell P., Murone C., Asadi K., et al. PD-L1 expression in NSCLC: analysis of a large early stage cohort: and concordance of expression in primary, node and metastasis. Journal of Thoracic Oncology. 2015; 10([supplement; S199 abstr]).

[0174] 22. Le DT, Durham JN, Smith KN, Wang H, Bartlett BR, Aulakh LK, Lu S, Kemberling H, Wilt C, Luber BS, Wong F, Azad NS, Rucki AA, Laheru D, Donehower R, Zaheer A, Fisher GA, Crocenzi TS, Lee JJ, Greten TF, Duffy AG, Ciombor KK, Eyring AD, Lam BH, Joe A, Kang SP, Holdhoff M, Danilova L, Cope L, Meyer C, Zhou S, Goldberg RM, Armstrong DK, Bever KM, Fader AN, Taube J, Housseau F, Spetzler D, Xiao N, Pardoll DM, Papadopoulos N, Kinzler KW, Eshleman JR, Vogelstein B, Anders RA, Diaz LA Jr. Mismatch repair deficiency predicts response of solid tumors to PD-1 blockade. Science. 2017 Jul 28;357(6349):409-413.

[0175] 23. Viale G, Trapani D, Curigliano G. Mismatch Repair Deficiency as a Predictive Biomarker for Immunotherapy Efficacy. Biomed Res Int. 2017;2017:4719194.

[0176] 24. Boussiotis V. A. Somatic mutations and immunotherapy outcome with CTLA-4 blockade in melanoma. New England Journal of Medicine. 2014;371 (23):2230-2232.

[0177] 25. Goodman AM, Kato S, Bazhenova L, Patel SP, Frampton GM, Miller V, Stephens PJ, Daniels GA, Kurzrock R. Tumor Mutational Burden as an Independent Predictor of Response to Immunotherapy in Diverse Cancers. Mol Cancer Ther. 2017 Nov;16(11):2598-2608.

[0178] 26. Rizvi N. A., Hellmann M. D., Snyder A. Mutational landscape determines sensitivity to PD-1 blockade in non-small cell lung cancer. Science. 2015;348(6230):124-128.

[0179] 27. Carbone DP, Reck M, Paz-Ares L, Creelan B, Horn L, Steins M, Felip E, van den Heuvel MM, Ciuleanu TE, Badin F, Ready N, Hiltermann TJN, Nair S, Juergens R, Peters S, Minenza E, Wrangle JM, Rodriguez-Abreu D, Borghaei H, Blumenschein GR Jr, Villaruz LC, Havel L, Krejci J, Corral Jaime J, Chang H, Geese WJ, Bhagavatheeswaran P, Chen AC, Socinski MA; CheckMate 026 Investigators. First-Line Nivolumab in Stage IV or Recurrent Non-Small-Cell Lung Cancer. N Engl J Med. 2017 Jun 22;376(25):2415-2426.

[0180] 28. Davis AA, Chae YK* and Giles FJ. Is Circulating Tumor DNA (Ctdna) Use Ready For Prime Time? Applications and Challenges of Ctdna in the Era of Precision Oncology. Chemo Open Access 2017, 6:2.

[0181] 29. Chae YK, Davis AA, Jain S, Santa-Maria C, Flaum L, Beaubier N, Platanias LC, Gradishar W, Giles FJ, Cristofanilli M. Concordance of Genomic Alterations by Next-Generation Sequencing in Tumor Tissue versus Circulating Tumor DNA in Breast Cancer. Mol Cancer Ther. 2017 Jul;16(7):1412-1420.

[0182] 30. Kim H, Jen J, Vogelstein B, Hamilton SR. Clinical and pathological characteristics of sporadic colorectal carcinomas with DNA replication errors in microsatellite sequences. Am J Pathol. 1994 Jul;145(1):148-56.

[0183] 31. Ward R, Meagher A, Tomlinson I, O'Connor T, Norrie M, Wu R, Hawkins N. Microsatellite instability and the clinicopathological features of sporadic colorectal cancer. Gut 2001 Jun;48(6):821-9.

[0184] 32. Greenson JK, Bonner JD, Ben-Yzhak O, Cohen HI, Miselevich I, Resnick MB, Trougouboff P, Tomsho LD, Kim E, Low M, Almog R, Rennert G, Gruber SB. Phenotype of microsatellite unstable colorectal carcinomas: Well-differentiated and focally mucinous tumors and the absence of dirty necrosis correlate with microsatellite instability. Am J Surg Pathol. 2003 May;27(5):563-70.

[0185] 33. Alexander J, Watanabe T, Wu TT, Rashid A, Li S, Hamilton SR. Histopathological identification of colon cancer with microsatellite instability. Am J Pathol. 2001 Feb;158(2):527-35.

[0186] 34. David Gisselsson, Jonas Bjork, Mattias Hoglund, Fredrik Mertens, Paola Dal Cin, Mans Akerman, and Nils Mandahl. Abnormal Nuclear Shape in Solid Tumors Reflects Mitotic Instability. Am J Pathol. 2001 Jan; 158(1): 199-206.

[0187] 35. Bai H, Madabushi A, Guan X, Lu AL. Interaction between human mismatch repair recognition proteins and checkpoint sensor Rad9-Radl-Husl. DNA Repair (Amst). 2010 May 4; 9(5):478-87.

[0188] 36. Debes JD, Sebo TJ, Heemers HV, Kipp BR, Haugen DL, Lohse CM, Tindall DJ. p300 modulates nuclear morphology in prostate cancer. Cancer Res. 2005 Feb 1;65(3):708-12.

[0189] 37. Ionov Y, Matsui S, Cowell JK. A role for p300 / CREB binding protein genes in promoting cancer progression in colon cancer cell lines with microsatellite instability. Proc Natl Acad Sci USA. 2004 Feb 3;101(5):1273-8.

[0190] 38. Koshiishi N, Chong JM, Fukasawa T, Ikeno R, Hayashi Y, Funata N, Nagai H, Miyaki M, Matsumoto Y, Fukayama M. p300 gene alterations in intestinal and diffuse types of gastric carcinoma. Gastric Cancer. 2004;7(2):85-90.

[0191] 39. Pihan GA, Purohit A, Wallace J, Malhotra R, Liotta L, Doxsey SJ. Centrosome defects can account for cellular and genetic changes that characterize prostate cancer progression. Cancer Res. 2001 Mar 1;61(5):2212-9.

[0192] 40. Robinson HM, Black EJ, Brown R, Gillespie DA. DNA mismatch repair and Chkl -dependent centrosome amplification in response to DNA alkylation damage. Cell Cycle. 2007 Apr 15;6(8):982-92.

[0193] 41. Bornens M. Cell polarity: intrinsic or externally imposed? New Biol., 3: 627- 636, 1991.

[0194] 42. Rizzolo L. J., Joshi H. C. Apical orientation of the microtubule organizing center and associated g-tubulin during the polarization of the retinal pigment epithelium in vivo. Dev. Biol., 157: 147-156, 1993.

[0195] 43. Meads T., Schroer T. A. Polarity and nucleation of microtubules in polarized epithelial cells. Cell Motil. Cytoskeleton, 32: 273-288, 1995.

[0196] 44. Whitehead C. M., Salisbury J. L. Regulation and regulatory activities of centrosomes. J. Cell. Biochem., 32-33 (Suppl.): 192-199, 1999.

[0197] 45. Meyer M., Hayenga JW, Neumann T, Katdare R, Presley C, Steinhauer DE, Bell TM, Lancaster CA, Nelson AC. The Cell-CT 3-Dimensional Imaging Technology Platform Enables the Detection of Lung Cancer Using the LuCED Sputum Test. Cancer Cytopathology, 123(9): 512-523, 2015.

[0198] 46. Wilbur D., Meyer MG, Presley C, Aye RW, Zarogoulidis P, Johnson DW, Peled N, Nelson AC. Automated 3-Dimensional Morphologic Analysis of Sputum Specimens for Lung Cancer Detection: Performance Characteristics Support Use in Lung Cancer Screening. Cancer Cytopathology, 123(9): 548-556, 2015.

[0199] 47. Fauver M, Seibel E. Rahn J. Three-dimensional imaging of single isolated cell nuclei using optical projection tomography. Optics Express, 13:4210-4223, 2005.

[0200] 48. Deans S. The Radon Transform and Some of Its Applications, New York: Dover Publishers: 2007.

[0201] 49. Rawski [Accessed March 19, 2015]; Maximum intensity projection (MIP) from Radiopaedia.org URL Obtained from http: / / radiopaedia.org / articles / maximum- intensity-projection-mip ​ .

[0202] 50. Schapire R, Freund Y. Boosting, Foundations and Algorithms, Cambridge: MIT Press, 2012

[0203] 51. Breiman L. Random Forests. Machine Learning, 45:5-32, 2001.

[0204] 52. Nelson A., M. Meyer, D. Sussman, R. Katdare, C. Presley, F. Lakers, C. Hamilton, D. Wilbur, R. Mastrangelo, M. Doherty Morphometric Genotyping Identifies Lung Cancer Cells Harboring Target Mutations; Cell-CT Platform Detects Gene Abnormalities, Journal of Thoracic Oncology, 12(11) (Supplement 2): S2278-S2279, 2017.

[0205] 53. Bailis JM, Gordon ML, Gurgel JL, Komor AC, Barton JK, Kirsch IR. An inducible, isogenic cancer cell line system for targeting the state of mismatch repair deficiency. PLoS One. 8:e78726, 2013.

Claims

1. A system for training one or more morphometric classifiers to identify tumor mutational burden (TMB), the system comprising: One or more processors; as well as A storage device that stores an instruction set, which, when executed by the one or more processors, causes the one or more processors to perform the following operations: 3D images of cells are analyzed based on pseudoprojections obtained from subject samples; Operate a biological sample classifier to identify whether the cells in the sample are normal or abnormal; Determine the TMB score for each abnormal cell; and Apply a predetermined threshold to the TMB score; When cancer is detected, surgery is performed to remove the cancerous lesions; When the TMB score exceeds the predetermined threshold, the subject is classified as a candidate for immunotherapy, which involves administering an immunomodulator to the human subject within a predetermined time period to help the human subject's immune system eliminate cancer cells.

2. The system according to claim 1, wherein, The immunomodulatory agents include drugs selected from the group consisting of: chimeric immune receptors, prostacyclin analogs, iloprost, T-cell chimeric antigen receptors (CARs), vorinostat, HDAC inhibitors, cholecalciferol, calcitriol, and combinations thereof.

3. The system according to claim 1, wherein, Low TMB and high TMB were used as underlying facts to develop biological sample classifiers for each cell in isogenetic cell lines and to determine the ROC area of ​​each cell classifier.

4. The system according to claim 3, wherein, Define a score for each cell in the isogenetic cell line that matches the underlying facts.

5. The system according to claim 1, wherein, Generate scores that closely match the underlying facts.

6. The system according to claim 1, wherein, An adaptive augmented logistic regression algorithm is used to define a set of projection axes used by a logistic function to produce scores from 0 to 1.

7. The system according to claim 6, wherein, The adaptive augmented logistic regression algorithm iterates through successive trials that weight each observation by using the difference between the underlying facts and the current score, adaptively converging to a solution that gradually transforms a broader set of cellular representations into a solution.

8. The system according to claim 1, wherein, The random forest algorithm is used to generate a biological sample classifier using nonparametric assumptions about the feature distribution.

9. The system according to claim 8, wherein, The classifier's discrimination is evaluated by pruning the potential feature tree set to optimize the discrimination.

10. The system according to claim 1, wherein, The area under the ROC curve, aROC, is used to determine the classifier's effectiveness. This is achieved by calculating the area under the receiver operating characteristic curve, or aROC, by integrating the ROC curve. The integral of the ROC curve represents the overall performance of the binary classifier output in terms of classification sensitivity and specificity.

11. The system according to claim 1, wherein, A threshold is established to be used in conjunction with the classifier score to create a binary output associated with the underlying TBM having high accuracy.

12. The system according to claim 1, wherein, Generate a numerical score representing the probability that a cell belongs to the target category.

13. The system according to claim 12, wherein, By applying a threshold to the fraction distribution to further binaryize the fractions, target cells are separated from non-target cells.

14. The system according to claim 12, wherein, The threshold was determined to provide an accuracy of 0.95 or higher for separating cells with low TMB from cells with high TMB.

15. The system according to claim 14, wherein, TMB data were determined from genomic profiling analysis using whole-exome sequencing or targeted exome sequencing using NGS or targeted genome sequencing.

16. A system for training one or more morphometric classifiers to identify tumor mutational burden (TMB), the system comprising: One or more processors; as well as A storage device for storing an instruction set that, when executed by the one or more processors, causes the one or more processors to perform the following operations: A biological sample classifier is used to identify whether cells in a sample are normal or abnormal, wherein the sample is taken from a subject and 3D images of the cells in the sample are analyzed based on pseudo-projection. Determine the TMB score for each abnormal cell; and Apply a predetermined threshold to the TMB score; When the TMB score exceeds the predetermined threshold, the subject is classified as a candidate for immunotherapy, which involves administering iloprost to the human subject within a predetermined time period to help the human subject's immune system eliminate cancer cells.

17. The system according to claim 16, wherein, Low TMB and high TMB were used as underlying facts to develop biological sample classifiers for each cell in isogenetic cell lines and to determine the ROC area of ​​each cell classifier.

18. The system according to claim 17, wherein, Define a score for each cell in an isogenetic cell line that matches the stated basic facts.

19. The system according to claim 16, wherein, Generate scores that closely match the underlying facts.

20. The system according to claim 16, wherein, An adaptive augmented logistic regression algorithm is used to define a set of projection axes used by a logistic function to produce scores from 0 to 1.

21. The system according to claim 20, wherein, The adaptive augmented logistic regression algorithm iterates through successive trials that weight each observation by using the difference between the underlying facts and the current score, adaptively converging to a solution that gradually transforms a broader set of cellular representations into a solution.

22. The system according to claim 16, wherein, The random forest algorithm is used to generate a biological sample classifier using nonparametric assumptions about the feature distribution.

23. The system according to claim 22, wherein, The classifier's discrimination is evaluated by pruning the potential feature tree set to optimize the discrimination.

24. The system according to claim 16, wherein, The area under the ROC curve, aROC, is used to determine the classifier's effectiveness. This is achieved by calculating the area under the receiver operating characteristic curve, or aROC, by integrating the ROC curve. The integral of the ROC curve represents the overall performance of the binary classifier output in terms of classification sensitivity and specificity.

25. The method according to claim 16, wherein, Establish a threshold to be used with the classifier score to create a binary output associated with the underlying TBM with high accuracy.

26. The system according to claim 16, wherein, Generate a numerical score representing the probability that a cell belongs to the target category.

27. The system according to claim 26, wherein, By applying a threshold to the fraction distribution to further binaryize the fractions, target cells can be separated from non-target cells.

28. The system according to claim 27, wherein, The threshold is determined to provide an accuracy of 0.95 or higher for separating low TMB cells and high TMB cells.

29. The system according to claim 28, wherein, TMB data were determined from genomic profiling analysis using whole-exome sequencing or targeted exome sequencing using NGS or targeted genome sequencing.

Images