Cell identification method and device based on electromagnetic wave scattering spectrum

Abstract

Provided is a high-sensitivity and high-precision cell identification method for distinguishing between normal cells and abnormal cells in a cell-containing sample collected from a patient. A cell-containing sample collected from a patient is irradiated with electromagnetic waves, then the electromagnetic wave scattering spectrum of cells in the sample is measured, and normal cells and abnormal cells are identified according to the characteristics of the electromagnetic wave scattering spectrum. In the identification process, a learning model which learns electromagnetic wave scattering spectrums of normal cells and abnormal cells is used. The sample can be a living cell or a sample fixed and dyed by ethanol or formalin. The electromagnetic wave scattering spectrum may be a Rayleigh scattering spectrum, a Mie scattering spectrum, or a combination of both.

Description

Technical Field

[0001] This invention relates to a method and apparatus that uses cell-containing samples collected from patients to easily detect degenerated abnormal cells (also known as aberrant cells) from normal cells in the samples with high sensitivity and accuracy. Specifically, this invention relates to a method and apparatus for identifying tumor cells that are difficult to identify using an optical microscope. Background Technology

[0002] One method for examining malignant tumors is pathological examination, which involves observing tissue sections from suspected patients using an optical microscope. In this diagnostic process, after a clinician makes a preliminary diagnosis of a suspected malignant tumor, if abnormal features are found, the pathologist receives pathological specimens, such as histological or cytological specimens, provided by the clinician. The pathologist then diagnoses the type and extent of the tumor and develops a treatment plan based on this diagnosis. Therefore, pathological diagnosis plays a crucial role in medical practice. However, compared to other developed countries, Japan has a relatively insufficient number of pathologists, who must handle an extremely large workload of diagnosing specimens. Furthermore, when using an optical microscope for pathological diagnosis, it is often difficult to distinguish between tumor cells and normal cells, and the diagnostic results rely to some extent on the pathologist's subjective experience. In response to these challenges, in recent years, AI (artificial intelligence) based on machine learning, which analyzes images from optical microscope observations of pathological specimens for assisted diagnosis, has gradually become a research hotspot for addressing these problems.

[0003] Even experienced pathologists cannot distinguish some tumor cells from normal cells when observing specimens using only an optical microscope. The differences between tumor cells and normal cells are not only reflected in the micrometer-level cell morphology or organelle structures observable under an optical microscope, but also in the microscopic structural disorder of the cytoskeleton and intracellular organelles at the micrometer or even smaller (nanometer) level, as well as protein denaturation. Research suggests that combining optical microscopic images with light scattering spectroscopy to detect and analyze this nanometer-level information could potentially improve the diagnostic accuracy of pathological specimens. Furthermore, acquiring high-density information at a speed exceeding that of optical microscopy could potentially enable high-speed, high-precision AI analysis.

[0004] Before analysis, appropriate sample preparation and accurate detection procedures are essential. In cytological diagnosis, the collected sample is typically first smeared onto a glass slide and then immediately fixed with ethanol for dehydration. If the sample is already dry, it needs to be rehydrated and then fixed with 95% ethanol. After fixation, Papanicolaou staining or other similar staining methods are performed. In pathological diagnosis, the collected sample is usually fixed with formalin and prepared into paraffin-embedded tissue blocks. The tissue blocks are then sectioned and stained with hematoxylin and eosin (HE). When a tumor is suspected, an immunostaining procedure corresponding to multiple tumor markers is used, and a pathologist performs high-precision identification. By observing the stained samples, tumor cells can be identified based on their characteristics, including enlarged nuclei, multinucleation, and abnormal intranuclear microstructure. However, many types of tumor cells are extremely difficult to distinguish accurately based on these differences alone. Taking the cytological diagnosis of the rare malignant mesothelioma as an example, even professional physicians often find it difficult to distinguish mesothelioma cells from reactive mesothelioma cells and adenocarcinoma cells, so a more suitable discrimination method is needed.

[0005] Because the preparation of pathological specimens (including a series of procedures such as fixation and staining) requires a great deal of effort and time, it is desirable to differentiate tumors during surgery or immediately after sampling, before fixation and staining. As a method, for example, Patent Document 1 discloses a method for detecting abnormal cells (tumor cells) by directly irradiating living cells with a laser without staining, and performing spectral analysis on the induced emission light emitted by the living cells. Specifically, the aim is to detect abnormal cells by irradiating cells with a visible laser and performing spectral analysis (spectroscopy) on the induced emission light of the biological sample. Although this invention relates to the application of light scattering spectroscopy, in embodiments of this invention, the scattered light component of the laser irradiating the living cells is filtered out by a filter; therefore, the light scattering targeted by this invention is not the object of detection. Because the spectrum of stimulated emission light has low precision in distinguishing intracellular states, some types of tumor cells cannot be detected.

[0006] Alternatively, another method allows for staining of live cells during surgery or immediately after collection without fixation. For example, Non-Patent Literature 1 discloses a technique for fluorescent staining using 5-aminolevulinic acid (5-ALA). 5-ALA is specifically absorbed by tumor cells and metabolized (converted) into the fluorescent molecule protoporphyrin IX, which accumulates within the tumor cells. The technique of distinguishing tumor cells by detecting the fluorescence signal of protoporphyrin IX is widely used in skin cancer treatment in Europe and America. In Japan, it is also used as an intraoperative fluorescent diagnostic reagent for brain tumors and bladder cancer to facilitate complete tumor resection. However, this method of identifying tumor cells solely based on the presence or absence of protoporphyrin IX fluorescence cannot match the precision of tumor cell identification compared to pathological specimens stained with various pigments, and it cannot distinguish tumor cell types. Furthermore, the staining method using 5-ALA to detect the metabolism of live tumor cells cannot be applied to the staining of fixed specimens; therefore, it cannot be used for the method targeted by this invention, which enables highly sensitive, accurate, and convenient detection of normal and abnormal cells (tumor cells) in cellular pathological specimens collected from patients.

[0007] Furthermore, existing research has explored various technical approaches to distinguish normal cells from abnormal cells (tumor cells) by directly irradiating them with electromagnetic waves such as light and analyzing their scattering spectra, without requiring staining or other treatments on the living cells. However, the accuracy of this method has not yet surpassed that of pathological tissue section observation.

[0008] For example, studies such as Non-Patent Literature 2 and Non-Patent Literature 3 report instances of detecting tumor cells by analyzing backscattered light observed when reflected from the surface of living tissue under visible light irradiation. In Non-Patent Literature 2, the characteristic of enlarged tumor cells was detected by analyzing the angle dependence of the intensity of Mie scattering light from backscattered light based on the size of the living cell nucleus. Similarly, Non-Patent Literature 3 shows that the presence of enlarged cell nuclei in tumor cells can be detected by analyzing the wavelength dependence of the intensity of Mie scattering light from backscattered light from living cells. As these examples show, for simple cells with a near-spherical shape, the changes in the direction or spectral morphology of Mie and Rayleigh scattering caused by changes in the size of the cell nucleus can be theoretically predicted based on physical principles. However, for changes in the direction or spectral morphology of Mie and Rayleigh scattering caused by more complex intracellular morphology, it is currently impossible to clearly predict them by physical means, and it is also difficult to distinguish the types of tumor cells based on this.

[0009] Patent Document 2 discloses a cell type identification method that uses flow cytometry to measure a fluid containing a mixture of multiple cells and, based on machine learning artificial intelligence technology that does not require explicit physical prediction, analyzes optical images or spectral time-series signals obtained when the relative position of the illumination pattern and the observed object is changed in a microchannel. This method can use machine learning-based AI technology to distinguish between normal and abnormal cells in a large number of cells. However, its application is limited to situations where cells flow at a constant rate in flow cytometry and time-series signals can be obtained; it is difficult to apply to cells attached to and fixed on a substrate, such as in pathological specimens. On the other hand, the present invention is characterized by not requiring time-series signals. The invention described in Patent Document 2 fails to achieve the purpose of the present invention, namely, it cannot easily, sensitively, and accurately detect the normal and tumor cells contained in a static sample collected from a patient.

[0010] As mentioned above, studies that use living cells and employ various measurement methods to distinguish between normal and abnormal cells primarily aim to accurately determine the extent of tumor tissue during surgery. However, the aforementioned methods fail to utilize more precise identification techniques based on tumor markers expressed in various tumors; therefore, they can only detect the presence or absence of tumor cells or tissue containing tumor cells.

[0011] This invention aims to provide a method for differentiating normal cells from abnormal cells in living cells. Specifically, for fixed and stained pathological specimens prepared to adapt to existing pathological diagnostic procedures, the method uses nano- to sub-nanometer-level morphological characteristics such as cell micromorphology, nucleus morphology, and intranuclear microstructure to measure the light scattering spectrum of the pathological specimen and combine it with machine learning technology to analyze it, thereby achieving the differentiation between normal cells and abnormal cells.

[0012] In other words, the object of this invention is to provide a simple, highly sensitive, and highly accurate method for detecting normal and abnormal cells in a cellular sample collected from a patient. The sample used includes pathological specimens collected from patients and subjected to fixation and staining processes, following existing pathological diagnostic procedures, as well as samples from the pre-fixation and staining stages. The aim of this method is to transform the current diagnostic approach, which relies on the experience and skill of pathologists, into a detection method that is independent of personal experience and skill and possesses high sensitivity and accuracy. In particular, this invention aims to provide a method for identifying tumor cells that are difficult to distinguish using an optical microscope.

[0013] [Patent Documents]

[0014] [Patent Document 1] Japanese Patent Application Publication No. 58-118948

[0015] [Patent Document 2] Japanese Invention Patent No. 6959614

[0016] [Non-patent literature]

[0017] [Non-Patent Literature 1] Shunichiro Ogura, et al., "Photodynamic therapy targeting dormant cancer cells using 5-aminolevulinic acid", Journal of the Japanese Society for Laser Medicine, Vol. 43, No. 4, pp. 238-248 (2023).

[0018] [Non-patent document 2] A. Wax, et. al., "Cellular organization and substructure measured using angle-resolved low-coherence interferometry", Biophysicaljournal, 82(4), pp.2256-2264 (2002).

[0019] [Non-patent document 3] V. Backman, et. al., "Polarized light scatteringspectroscopy for quantitative measurement of epithelial cellular structures in situ", IEEE Journal of Selected Topics in Quantum Electronics, 5(4), pp.1019-1026 (1999). Summary of the Invention

[0020] The technical problem that the invention aims to solve

[0021] This invention aims to provide a method and apparatus for detecting normal and abnormal cells in cell-containing samples collected from patients with high sensitivity and accuracy. In particular, the purpose of this invention is to provide a method and apparatus for identifying tumor cells that are difficult to distinguish using an optical microscope.

[0022] To address the aforementioned problems, the cell identification method of this invention involves irradiating a cell-containing sample (cell sample) collected from a patient with electromagnetic waves to obtain the electromagnetic wave scattering spectrum of the sample, and then distinguishing normal cells from abnormal cells based on the information obtained from this spectrum. Light scattering is affected by the reflection or refraction of structures with dimensions approximately equal to or smaller than the wavelength of light (100 nm to 1 μm). When the size of the scattering structure is approximately the same as the wavelength (micrometer scale), this scattering is called Mie scattering; when the scattering structure is smaller than the wavelength (submicrometer scale), this scattering is called Rayleigh scattering. The key feature of this invention is the use of spectral information (spectral shape) from Rayleigh and Mie scattering to identify abnormal cells. Although optical images also contain information from Rayleigh and Mie scattering, they do not contain information beyond the diffraction limit. By analyzing the light scattering spectrum, a feature of this invention, information about submicrometer-scale intracellular structures that cannot be observed by an optical microscope beyond the diffraction limit can be obtained. Its novelty lies in extracting information from the light scattering spectrum using mathematical methods rather than based on physical and chemical interpretations. When cell samples are exposed to light, not only do the micrometer-scale cell and nucleus morphology affect the direction and spectral pattern of light scattering, but the submicrometer-scale cell morphology, nucleus morphology, and intranuclear microstructure also alter the direction and spectral pattern of light scattering. Furthermore, when cells are stained, as in pathological specimens, to emphasize abnormal cell structural features or the types of proteins they contain, light scattering is strongly influenced by the pigment. These factors are undoubtedly reflected in the light scattering spectrum of pathological specimens. However, clearly correlating the simple morphological features observed in the spectrum with cells and intracellular structures, and conducting theoretical analysis based on physical principles, is difficult due to its complexity.

[0023] To address this problem, this invention proposes a method for predicting pathological diagnostic results by analyzing light scattering spectra using machine learning. In the machine learning process, the system is not based on physical logic, but rather extracts feature quantities from the light scattering spectrum numerically and performs statistical analysis on these features to train a prediction model (also known as a learning model). Inputting other light scattering spectra into this prediction model yields a prediction process for normal or abnormal cells; this analytical method is known as AI analysis. The feature quantities obtained through numerical methods may not conform to the spectral characteristics perceived by human vision, and in some cases depend on subtle spectral differences imperceptible to the naked eye. This fundamentally distinguishes this method from traditional physics-based theoretical analysis.

[0024] In other words, this invention solves the problem of interpreting light scattering spectra data in a way that is too complex to be theoretically explained based on physics, by using numerical analysis that is not based on physical logic, namely machine learning.

[0025] Invention Effects

[0026] This invention provides a highly sensitive and precise method for differentiating normal cells from abnormal cells. This method can easily identify tumor cells that are difficult for pathologists to distinguish using conventional optical microscopes, eliminating reliance on the pathologist's experience and skill level, thereby improving diagnostic accuracy. Attached Figure Description

[0027]

Figure 1

[0028]

【 Figure 2 [This refers to the composition of MDCK cell samples.]

[0029]

【 Figure 3 The image shows a comparison of bright-field and dark-field images of MDCK cells with and without cytochalasin D (a) and with cytochalasin D (b).

[0030]

【 Figure 4 (a) is the normalized light scattering efficiency (SES) spectrum of MDCK cells immersed in culture medium immediately after the start of the measurement and 60 minutes later. n710 (a) is a comparison of (b) and (c) is a normalized light scattering efficiency (SES) spectrum of cells collected every 10 minutes from immediately after the addition of cytochalasin D to the cells until 60 minutes later. n710 The changing trend of ).

[0031]

【 Figure 5 The light scattering efficiency spectrum (SES) of cells contained in specimens A to F collected from normal patients is shown.

[0032]

【 Figure 6 These are bright-field and dark-field images of cells contained in specimens A through F collected from normal patients.

[0033] Among them, (a) represents cells whose SES spectra show a peak around 700 nm; and (b) represents cells whose SES spectra show a flat feature.

[0034]

【 Figure 7 [Image] is the light scattering efficiency spectrum (SES) of cells contained in G, H, and I samples from suspected mesothelioma patients.

[0035]

【 Figure 8 [Images] are bright-field and dark-field images of cells contained in G, H, and I samples from suspected mesothelioma patients.

[0036]

【 Figure 9 (a) shows the loadings of the first and second principal components obtained through principal component analysis of the light scattering spectrum, and (b) is a scatter plot of the percentage of the first and second principal components contained in the light scattering spectrum (score).

[0037]

【 Figure 10 [ ] is a concept graph for binary classification, representing (a) k-nearest neighbor algorithm, (b) decision tree algorithm, and (c) random forest algorithm.

[0038]

【 Figure 11 [Illustration diagram showing the allocation of learning and testing data for samples A to I used to verify the classification accuracy (correctness) of the machine learning model for light scattering spectra.]

[0039]

【 Figure 12 [Illustration 1] is a structural block diagram of a cell identification device.

[0040]

【 Figure 13 [ ] is the light scattering efficiency spectrum (SES) of the cells contained in specimens A to I after normalization. n710 ).

[0041]

【 Figure 14 The contribution rate and loading results are from the first principal component (PC1) to the tenth principal component.

[0042]

【 Figure 15 [This refers to the SES of cells] n710 Scatter plot of the scores of the nth principal component and the mth principal component.

[0043]

【 Figure 16 The image above is an illustration of the malignancy level of tumor cells.

[0044]

【 Figure 17 [This is an auxiliary diagram for pathological diagnosis.]

[0045]

【 Figure 18 [ ] is the light scattering efficiency spectrum (SES) of two types of live cells after normalization treatment. n710 ).

[0046]

【 Figure 19 [This refers to SES containing two types of cells.] n710 Scatter plot of the scores of the nth principal component and the mth principal component.

[0047]

【 Figure 20 The diagram above illustrates the scattering spectroscopy measurement method. Detailed Implementation

[0048] This invention relates to a cell identification method, characterized by irradiating a cell-containing sample collected from a patient with electromagnetic waves and obtaining the light scattering spectrum of the generated scattered light. Based on the characteristic differences in the light scattering spectra of normal and abnormal cells, the method extracts the feature quantities contained in the spectrum numerically, and constructs a learning model accordingly to achieve the identification of unknown cells.

[0049] Cells are isolated from the external environment by the cell membrane, and the cytoplasm contains various organelles, primarily the nucleus. It is known that the size or microstructure of many organelles undergoes characteristic changes in normal and abnormal cells. This invention aims to capture characteristic quantities of light scattering spectra, particularly those of abnormal cells, including various tumor cells that play an important role in pathological diagnosis. However, this invention is not limited to this; it also provides a method for detecting and analyzing various intracellular and extracellular microstructures using light scattering spectroscopy.

[0050] Cellular and intracellular structural degeneration can be triggered by a variety of factors, including natural aging, chemical or physical disturbances, stimulation by various physiologically active substances, infection, or cell carcinogenesis. Therefore, early detection of cellular abnormalities by capturing morphological changes in the cell nucleus and other organelles is of great significance. Cellular degeneration, especially morphological and nuclear degeneration, plays a crucial role in various pathological diagnoses. For example, tumor cells may be characterized by cell clumping, enlarged and multinucleated nuclei, enlarged nucleoli, and narrowed cytoplasm. Furthermore, tumor cells may also exhibit abnormal polymerization of cytoskeletal proteins such as actin and keratin.

[0051] Furthermore, many cancerous malignant tumors exhibit increased mobility, crossing the basement membrane due to epithelial cell degeneration (epithelial-mesenchymal transition), leading to mesenchymal infiltration or metastasis to other organs. However, at this time, actin filaments maintaining the cytoskeleton undergo reorganization, forming filamentous pseudopodia in the cell membrane, further increasing cell mobility. Therefore, by observing morphological changes at the micron to submicron level, such as cell morphology, nuclear shape and size, organelles, and the cytoskeleton, cell degeneration caused by tumors and other lesions can be effectively detected. This invention employs a cell identification method that involves irradiating cell samples collected from patients with electromagnetic waves to obtain the electromagnetic wave scattering spectrum of the cells in the sample. A learning model trained on the scattering spectrum characteristics of normal and abnormal cells is then used to distinguish between normal and abnormal cells. Samples can be live cells or stained samples fixed with ethanol or formalin.

[0052] Light scattering is widely used in research to acquire cellular characteristics. Light scattering phenomena can be characterized by a particle size parameter (α; α = πd / λ) defined by the ratio of the scatterer diameter (d) to the observed electromagnetic wavelength (λ), and the relative refractive index of the scatterer. When α << 1, Rayleigh scattering predominates, and the intensity of the scattered light increases with decreasing wavelength. When α approaches 1, Mie scattering is dominant, and the intensity of the scattered light remains constant, independent of α. Furthermore, when α reaches 10⁻⁶ or greater, diffraction occurs due to interference of the scattered light, exhibiting complex response characteristics to the scattering angle and the value of α.

[0053] In this invention, Rayleigh scattering spectrum, Mie scattering spectrum, or a combination of both can be used as the scattering spectrum. Furthermore, more preferably, the scattering spectrum can use either Rayleigh scattering-generated forward scattering spectrum or Mie scattering-generated forward scattering spectrum, or a combination of both. Moreover, the difference in electromagnetic wave scattering spectra between normal and abnormal cells is due to an increase or decrease in scattering resulting from changes in cell structure that are less than or equal to changes in the wavelength of visible light.

[0054] The wavelength of the electromagnetic waves used for observation in this invention is preferably 100 nm to 1 μm, and particularly preferably visible light in the wavelength range of 400 to 750 nm. The electromagnetic waves can also be used in the ultraviolet to infrared range. Furthermore, in applications of this invention, electromagnetic waves in the X-ray to ultraviolet wavelength range can be used to obtain the electromagnetic scattering spectrum of smaller regions of cells in a sample, and a learning model obtained by learning the electromagnetic scattering spectral characteristics of normal and abnormal cells can be used to distinguish between normal and abnormal cells.

[0055] The present invention is characterized by its ability to detect the structural characteristics of scatterers with high sensitivity and precision by observing the scattered light from a single cell. Specifically, the scatterers are selected from structures that constitute the cell, such as the cell nucleus and cytoskeleton; therefore, the value of the α parameter ranges from 0.1 to 150. Since the size of the cell nucleus ranges from several micrometers to 10 micrometers, when the wavelength of the electromagnetic wave to be observed is around 500 nm, the value of α is in the range of approximately 20 to 120. In this case, the intensity of the scattered light largely depends on the scattering angle, and most scattering phenomena exhibit forward scattering in the same direction as the incident light. Using the method of the present invention, the scattered light from the cell is observed by detecting the scattered light scattered in any direction at the incident angle of the electromagnetic wave irradiating the cell, thereby obtaining the electromagnetic wave scattering spectrum of the cells in the sample. A learning model obtained by learning the electromagnetic wave scattering spectrum characteristics of normal and abnormal cells is then used to distinguish between normal and abnormal cells, thus realizing a cell identification method.

[0056] It is known that when the cross-sectional area of ​​the scatterer is larger than the wavelength of the observed electromagnetic wave, the scattering efficiency exhibits complex variations with the value of α. Furthermore, when the scatterer possesses the relative refractive index of cellular components in media such as water, the scattering efficiency significantly depends on the size of the scatterer and the wavelength of the electromagnetic wave. In non-patent literature 4 (M. Xu et al., “Unified Mie and fractal scattering by cells and experimental study on application in optical characterization of cellular and subcellular structures”, Journal of biomedical optics, 13(2), 024015-024015 (2008).), researchers compared the measured light scattering spectra with calculated values ​​by irradiating cells with visible light and varying the scattering angle within the range of 1.1–165°. Additionally, the scattered light produced by cells can be represented as the superposition of the nuclear scattering component, the homogeneous cytoplasmic scattering component, and the scattering component of tiny organelles other than the nucleus. The results of measuring the light scattering spectra at various scattering angles will be compared with calculated values ​​to illustrate this. Accordingly, under forward scattering conditions with a scattering angle of 1.1°, the intensity of scattered light increases with increasing wavelength; while in backscattering, due to phase reversal, the intensity of scattered light decreases with increasing wavelength. Analysis of the contributions of various components of scattered light reveals that scattering from the cell cytoplasm contributes as background, while, in addition to the fractal "wave" components from organelles, the scattering components generated by the cell nucleus account for a significant proportion. As mentioned above, research shows that by irradiating cells with visible light, the light scattering spectrum of cells can be accurately measured within a scattering angle range of 1.1–165°. In the method of this invention, the electromagnetic wave scattering spectrum of cells can be obtained under both forward and backscattering observation conditions, and a machine learning model obtained by learning the electromagnetic wave scattering characteristics of normal and abnormal cells can be used to distinguish between normal and abnormal cells.

[0057] In Non-Patent Literature 4, the wavelength dependence of scattered light intensity becomes increasingly significant with increasing cell nucleus size. In forward scattering, the scattered light intensity increases significantly with wavelength, but the opposite trend is observed in backscattering. Non-Patent Literature 4 also investigated the scattering behavior during acetic acid-induced cell degeneration, showing that organelles denatured by acetic acid exhibit whitening, with a significant increase in scattering intensity, and the contribution of the "fluctuation" component increases with decreasing wavelength.

[0058] Therefore, this invention can capture and analyze scattered light from any angle, whether from front or back scattering, thereby accurately identifying the differences between normal and abnormal cells that are difficult to distinguish using traditional methods.

[0059] In embodiments of the present invention, the light scattering spectrum obtained by measurement is represented as two-dimensional data showing the relationship between scattered light intensity and wavelength. In this invention, to collect and analyze a large number of light scattering spectra obtained from single cells, principal component analysis is used as a method for extracting features from the two-dimensional data. Principal components are extracted by diagonalizing the covariance matrix of the scattered light intensity at each wavelength. The first principal component is the eigenvector with the largest eigenvalue, maximizing the variance of the scattering intensity. In addition to the first principal component, the second principal component, with the second largest eigenvalue, is used to analyze the features of the cell's scattered light to create a predictive model for distinguishing the differences in feature quantities between normal and abnormal cells. Specifically, the predictive model employs the k-Nearest Neighbors algorithm (kNN), the Decision Tree algorithm, and the Random Forest algorithm (RF). Regarding accuracy evaluation, the trained predictive model predicts whether cells not used for training are normal or abnormal based on the cell's light scattering spectral features, and the effectiveness and accuracy of the analysis method are evaluated based on the accuracy rate.

[0060] This invention provides a cell identification method based on AI (artificial intelligence) analysis. It measures the light scattering spectrum of individual cells in a cell-containing sample collected from a patient, analyzes the characteristic quantities of the obtained light scattering spectrum using the aforementioned method, and uses a learning model constructed from previously obtained light scattering spectra of normal and abnormal cells to determine whether the cells in the sample are normal or abnormal. Furthermore, when performing this type of AI analysis, it is preferable to use methods such as... Figure 12 The cell identification device 20 shown includes: a scattering spectrum acquisition unit 21, which irradiates a cell-containing sample collected from a patient with electromagnetic waves using an electromagnetic wave scattering spectrum measuring device 31 to acquire a scattering spectrum 32; a learning model 22, which has pre-learned the scattering spectra of normal cells and abnormal cells; and an identification unit 23, which uses the learning model to identify normal cells and abnormal cells. The identification results obtained by the identification unit 23 in the cell identification device 20 are output, for example, to a display device 33, such as a monitor, capable of classifying and displaying normal cells and abnormal cells (tumor cells) in numerical form. The cell identification device 20 is a computer system, whose memory contains programs for the learning model 22, the scattering spectrum acquisition unit 21, and the identification unit 23, and the corresponding programs are executed by a processor.

[0061] Hereinafter, an example of an embodiment of the present invention will be described in detail with reference to the accompanying drawings. It should be noted that the scope of the present invention is not limited to the following embodiments and illustrated examples, and many changes and modifications are possible.

[0062]

Example 1

[0063] In this embodiment, through experiments using cultured animal cells, it was found that changes in submicron-level intracellular structures that are difficult to observe under an optical microscope are causally related to light scattering spectra, and that light scattering spectra contain information about the microstructure within the cell nucleus.

[0064] By combining a dark-field inverted microscope with a spectrometer, the light scattering spectrum of a single cell in the wavelength range of 420–720 nm was measured. In this embodiment, the addition of the polymerization inhibitor cytochalasin D to animal cultured cells induced the breakage of fibrous actin in the cytoskeleton, and the changes in the single-cell light scattering spectrum during this process were systematically studied. Considering that the cell scattering spectrum includes the morphological characteristics of the white light spectrum of the light source, the light scattering efficiency spectrum (SES) obtained by dividing the measured value of the scattered light by the intensity of the scattered light from the light source was used as the benchmark for the light scattering spectrum. SES represents the wavelength dependence of the scattered light intensity when light of all wavelengths is incident at the same intensity. The absolute value of SES can fluctuate significantly between different cells due to factors such as cell state, cell illumination conditions, and objective lens adjustments for collecting scattered light. To correct for such fluctuations in the measurement, the normalized light scattering efficiency spectrum (SES) was obtained by normalizing the entire spectrum with the SES value at 710 nm. n710 It is used for machine learning analysis.

[0065] The cell line used was a canine kidney tubular epithelial cell line (MDCK; RCB0995, RIKEN Bioresource Center, Japan). These cells were cultured in cell culture dishes using Durbeco Modified Eagle Medium (phenol red-free) supplemented with fetal bovine serum (FBS) and antibiotics. When the cells were nearly confluent (80% coverage), the medium was replaced, and the cell culture dishes were placed on the sample stage. The central portion of the cell culture dish was observed using a dark-field inverted microscope (Olympus IX81 with a dark-field condenser lens). A ring-shaped spacer 11 (30 μm thick) was placed in the center of cell culture dish 10, and medium 12 was added within a 1.70 cm diameter area inside the ring, and MDCK cells 14 were seeded and cultured. During observation, a coverslip 13 was placed on top to prevent the medium from drying out.

[0066] Figure 1 A schematic diagram of the measurement system is shown. Figure 2This is a schematic diagram showing the configuration of the cell culture dish used for measurement. (Example) Figure 1 As shown in (b), the system uses a halogen lamp as the illumination source, illuminating the cells in the culture dish through a dark-field condenser lens with a numerical aperture larger than the objective lens, and collecting the scattered light generated by the cells through the objective lens. Because illumination is provided by a dark-field condenser lens with a numerical aperture larger than the objective lens, the light collected by the objective lens does not contain any component that transmits light through the cells. On the other hand, as... Figure 1 As shown in (c), in a conventional optical microscope image (bright-field image), the objective lens collects the light transmitted through the cells. The scattered light collected by the objective lens is split into two paths. One path is imaged onto a CMOS camera to obtain a dark-field image. The other path is introduced into a spectrometer via an optical fiber. The intensity spectrum of the scattered light measured by the spectrometer is divided by the spectrum of the illumination light to obtain the SES (Self-Effected Strain). In the scattered image formed on the fiber side, since the core diameter of the fiber corresponds to a region with a diameter of approximately 30 μm on the cell culture dish, selective measurement of SES at the single-cell level can be achieved.

[0067] Here, in addition to Rayleigh and Mie scattering, Raman scattering was also detected in the spectrometer. However, the intensity of Raman scattering is much lower than that of Mie and Rayleigh scattering, and below the detection limit of the spectrometer; therefore, its intensity is negligible or undetectable in this measurement. That is to say, as... Figure 20 As shown in (1), in this invention, the sample (scatterer) is irradiated with light, such as white light, covering the entire visible light spectrum, as the excitation light 41. The intensity of the scattered light, namely Mie scattering and Rayleigh scattering 43, is significantly higher than that of Raman scattering and fluorescence 44. Therefore, in the spectrometer 45, Raman scattering and fluorescence 44 can be ignored as scattering spectral components. On the other hand, as shown in FIG20 (2), in the optical system for selectively detecting Raman scattering and fluorescence 44, if light with high intensity characteristics at a specific wavelength is not used as the excitation light 46, and a powerful optical filter (excitation light cutoff filter) 47 for filtering out Mie scattering and Rayleigh scattering is not used, the main components detected by light scattering are Mie scattering and Rayleigh scattering. Mie scattering and Rayleigh scattering have the same wavelength as the irradiation light (excitation light) irradiating the cells, but the wavelengths of Raman scattering and fluorescence are different from those of the excitation light.

[0068] Figure 3 The results of observing bright-field and dark-field images of MDCK cells using a CMOS camera are presented. Figure 3Figure 3(a) shows an image of MDCK cells without the addition of cytochalasin D, and Figure 3(b) shows an image of MDCK cells with a broken cytoskeleton after the addition of cytochalasin D. For the bright-field and dark-field images in Figure 3(a), no significant difference was observed with the naked eye when observed under a microscope immediately after the addition of cytochalasin D (0 minutes) and 60 minutes later. Under these conditions, to confirm the actual breakage of actin filaments induced by the addition of cytochalasin D, we separately verified the observation of samples labeled with a fluorescent dye (SPY555-actin, SPIROCHROME, SC202) using confocal microscopy. Actin filaments, which constitute the cytoskeleton, do not absorb in the visible light region and their diameter is less than 10 nm; therefore, they are difficult to observe using an optical microscope, which explains why… Figure 3 No actin filament breakage was observed in the bright-field images on the left of (a) and (b). Figure 3 In the dark-field images on the right (a)(b), although signs of actin filament breakage may be observed, it is difficult to clearly distinguish the differences by visual inspection. However, by applying the machine learning analysis methods described later to quantify the structural changes, it is possible to classify and identify cells with broken actin filaments.

[0069] Figure 4 (a) Shows a comparison of MDCK cells and SES cells under a microscope at 0 min and 60 min in culture medium without added cytochalasin D. SES cells were observed immediately after measurement (0 min) and at 60 min. n710 The results are largely consistent. Even without the addition of cytochalasin D, there are concerns that prolonged cell placement might alter cell structure and thus affect SES (Self-Extraction Emulsion). n710 However, under these experimental conditions, no changes in cell structure as seen in the light scattering spectrum were observed for at least 60 minutes.

[0070] Subsequently, cytochalasin D at a concentration adjusted to 200 μM was added to MDCK cells to inhibit actin filament polymerization, and the cellular serous ether excision fraction (SES) was measured every 10 minutes from immediately after addition until 60 minutes later. n710 The trend of change is as follows Figure 4 As shown in (b). Figure 4 (b) shows that a change in the light scattering spectrum was observed immediately after the addition of cytochalasin D, specifically an increasing trend in light scattering intensity over time from around 500 nm to shorter wavelengths. This is attributed to the light scattering caused by the inhibition of actin filament polymerization, leading to the breakage of actin filaments of different lengths, which contributed to Rayleigh scattering. Figure 4The increase in short-wavelength side scattering intensity observed in (b) can be explained by this contribution. Therefore, this embodiment demonstrates a causal relationship between visible light scattering spectra and changes in intracellular structure, and can detect subtle intracellular structural differences that are difficult to discern in bright-field or dark-field images.

[0071] like Figure 4 As shown, the light scattering spectrum of unstained transparent cells under visible light exhibits a flat spectral line without obvious characteristics. However, it should be noted that this spectrum includes not only light scattering from actin fibers in the cytoskeleton but also light scattering from fine cellular structures, the cell nucleus, and intranuclear microstructures. Although the combined effect of these scattering signals forms the seemingly flat overall spectrum, slight irregular noise appearing on the flat spectrum may contain information related to these microstructures. Unfortunately, it is difficult to make deterministic predictions about these fluctuations based solely on physical principles. To address this, this invention proposes an innovative solution: using AI analysis technology based on machine learning, deep information originating from intracellular microstructures can be extracted from the light scattering spectrum without relying on deterministic predictions from physics.

[0072]

Example 2

[0073] This embodiment demonstrates that by following known pathological diagnostic procedures to fix and stain normal and tumor cells collected from actual patients to prepare pathological specimens, machine learning can effectively identify light scattering spectra that are difficult to interpret from a physics perspective.

[0074] Pleural effusion was collected from the upper rib margin of normal samples (A-F) and suspected samples containing pleural mesothelioma cells (G, H, I) using syringes. Cells from the pleural effusion were smeared onto glass slides, fixed by spraying, and then fixed in 95% ethanol fixative. Staining was performed using the Papanicolaou staining method; cell nuclei were stained with hematoxylin, and cytoplasm was stained first with OG-6 and then with EA-50. The stained cells were then infiltrated in a xylene bath, and the finally sealed samples were used as observation specimens. Pathologists identified the cells in each specimen, and single-cell samples identified as normal pleural mesothelial cells and other tumor cells were labeled. Light scattering spectroscopy measurements were performed based on the labeled samples.

[0075] Regarding normal cells, 23–25 pre-labeled normal cells in specimens A–F were analyzed. Regarding tumor cells, 13–27 pre-labeled tumor cells in specimens G–I were analyzed. It should be noted that the tumor cells included mesothelioma cells and adenocarcinoma cells. Furthermore, the possibility that reactive mesothelioma cells may be present in samples classified as normal cells or tumor cells cannot be ruled out.

[0076] A significant characteristic of mesothelioma cells is the higher probability of binucleated cells compared to other adenocarcinoma cells or reactive mesothelioma cells. Furthermore, mesothelioma cells and adenocarcinoma cells tend to form cell clusters, a characteristic that distinguishes them from the dispersed distribution of reactive mesothelioma cells. However, if only a simple comparison of nuclear size is made, the nuclear size of mesothelioma cells falls between that of adenocarcinoma cells and reactive mesothelioma cells; but due to the differences in nuclear size distribution, it is difficult to accurately distinguish them by observing only a single nucleus. Differentiation requires observing the distribution pattern of nuclear size across a large number of cells. Although the nucleus of mesothelioma cells is usually located in the center of the cell, while the nucleus of adenocarcinoma cells is often located at the cell periphery, nuclear location alone is still insufficient to effectively distinguish them from reactive mesothelioma cells.

[0077] The purpose of this embodiment is to verify the difference between the light scattering spectra of tumor cells and those of normal cells, rather than to identify adenocarcinoma cells, reactive mesothelial cells, or mesothelioma cells. Specifically, whether cells in a sample are mesothelioma cells or normal cells requires a comprehensive judgment from both pathological and clinical perspectives, while the purpose of this invention is to detect abnormal cells that differ from normal cells.

[0078] Figure 5 The light scattering efficiency (SES) spectra of cells in normal specimens A–F are presented. The SES characteristics of the cells in each specimen differ. While the SES of cells in specimens A and F exhibits a generally flat spectrum, cells in specimens B, C, and D show a peak in scattering efficiency around 700 nm. In specimen E, only some of the measured cells show a peak in SES at approximately 700 nm, while the SES of the remaining cells exhibits a generally flat spectrum.

[0079] exist Figure 6 In the SES measurements of cells in specimens A–F, (a) shows the bright-field image (top) and dark-field image (bottom) of cells exhibiting a scattering peak near 700 nm, and (b) shows the bright-field image (top) and dark-field image (bottom) of cells with flat SES spectra. In the cells of specimen B, it was confirmed that the cells exhibiting scattering peaks had larger nuclei; however, in the cells of specimen C, no clear correlation was observed between nucleus size and scattering peaks. This suggests that the appearance of scattering peaks may be influenced not only by increased nucleus size but also by other factors such as differences in relative refractive index caused by staining susceptibility.

[0080] Next, Figure 7 The results of SES measurements on cells contained in a suspected mesothelioma specimen (G-I) are presented. Furthermore, Figure 8 Optical microscopic images of cells in various specimens are shown (the top image is a bright field image, and the bottom image is a dark field image).

[0081] exist Figure 6 In the study, it was observed that all cells in specimen G exhibited flat SES patterns, a small number of cells in specimen H showed SES patterns with a peak around 700 nm, and a large number of cells in specimen I showed SES patterns with a peak around 700 nm. Furthermore, in Figure 8 In the cell optical microscopy images shown, the size of the cell nuclei in each cell was observed to have a certain distribution, and some nuclei were observed to be centrally located or diffusely distributed. From the above results, it can be seen that SES can clearly distinguish cells with a peak near 700 nm from cells with a flat distribution, but these results alone cannot necessarily distinguish between normal cells and tumor cells. SES obtained from cell diagnostic specimens reflects not only the microscopic structure of cells, nuclei, cytoskeleton, and intracellular organelles, but also the causal relationship of the reflection, absorption, and refraction of dye bound to cells in Papanicolaou staining. It is almost impossible to analyze all these factors contained in SES using physics-based theories. In this invention, machine learning is used as the identification method; machine learning is a numerical analysis method not based on physical logic.

[0082] By analyzing the normalized light scattering efficiency (SES) spectra of cells contained in specimens A to I n710 Principal component analysis was performed, yielding the loadings of the first and second principal components as shown in Figure 9(a). Based on these loadings, the SES of a single cell was calculated. n710 The percentages (scores) of the first and second principal components included are plotted, and a scatter plot is generated with the score of the first principal component as the horizontal axis and the score of the second principal component as the vertical axis, as shown in Figure 9(b). For example, when the scores of both the first and second principal components are in the positive value region ( Figure 9 (b) Quadrant 1), its SES n710 This can be represented by adding the loadings of the first principal component and the second principal component; if the score of the first principal component is in a positive region and the score of the second principal component is in a negative region ( Figure 9 (b) Quadrant 2), then the corresponding SES n710 This can be expressed by subtracting the second principal component loading from the first principal component loading. Additionally, SES located near the origin... n710 This indicates that the first and second principal components are almost entirely absent. Solid dots correspond to the SES of normal cells contained in normal specimens A through F. n710 The score is represented by hollow dots, which correspond to the SES of tumor cells contained in samples G to I of suspected mesothelioma. n710The scores of normal cells were mainly distributed in quadrants 3 and 4, but also in quadrants 1 and 2; while the scores of tumor cells were concentrated in quadrants 1 and 2. The results indicate that normal cells and tumor cells that could not be clearly classified by SES in Figures 5 and 7 can be numerically classified using the second principal component score.

[0083] The calculation method for each principal component score will be explained below. The score of the first principal component for patient n (n=1~N) is represented by Equation 1 below. As shown in Equation 2 below, w1(λ) is determined by weighting (eigenvector) the first principal component, such that the sum of squares of w1(λ) is 1. At the same time, as shown in Equation 3 below, w1(λ) is determined to maximize the variance V1 of the first principal component score.

[0084]

Mathematical Formula 1

[0085]

[0086]

Mathematical Formula 2

[0087]

[0088]

Mathematical Expression 3

[0089]

[0090] Furthermore, the score of the second principal component for patient n (n = 1 to N) is represented by Equation 4 below. Similar to the first principal component, the sum of squares of w2(λ) is determined to be 1 by weighting the second principal component as shown in Equation 5 below, and an orthogonality condition is imposed on w2(λ) as shown in Equation 6 below, making w2(λ) orthogonal to w1(λ), ​​and w2(λ) is determined to have the largest variance of the second principal component score under this condition.

[0091]

Mathematical Expression 4

[0092]

[0093]

Mathematical Expression 5

[0094]

[0095]

Mathematical Expression 6

[0096]

[0097] Then, in the score of the nth principal component of patient n (n = 1 to N), the orthogonality condition is added to determine w that maximizes the variance of the score of the nth principal component. n (λ). Thus, spectral data is much simpler than image data and can be analyzed with less subjectivity.

[0098] By using SES of unknown cells n710 Data and Figure 9 (b) is compared to infer whether the cell belongs to a normal cell or a tumor cell. Specifically, based on the SES measured from the unknown cell... n710 Data, calculate its value Figure 9 The loading scores on the first and second principal components shown in (a) are used to identify cell types by observing the distribution of these score points on the scatter plot in Figure 9(b). For example, if the score points of unknown cells are located in the third or fourth quadrant where a large number of normal cells are clustered, these unknown cells are likely to be normal cells; if the score points are located in the first or second quadrant where tumor cells are more concentrated, these unknown cells are likely to be tumor cells.

[0099] Binary classification methods used to distinguish between normal cells and tumor cells include representative algorithms such as the k-nearest neighbors algorithm, decision tree algorithm, and random forest algorithm. Figure 10 As shown in (a), the k-nearest neighbor algorithm is a method for determining the classification of samples by statistically analyzing the majority vote of the k nearest neighbors. Figure 10 (b) The presented decision tree algorithm divides the scatter plot into several regions (decision trees) and classifies samples within each region based on the majority voting principle. For example... Figure 10 As shown in (c), the random forest algorithm performs a primary majority vote by integrating the voting results of multiple different decision trees, and then performs a secondary majority vote on the voting results to complete the sample classification.

[0100] like Figure 11 As shown, SES of normal cells contained in one of the specimens A through F is obtained. n710 As test data for hypothetical unknown cells, the remaining normal cells were subjected to SES. n710 The SES of tumor cells contained in one specimen from specimens G to I were used as training data for constructing a predictive model of scatter plots. n710 As test data for hypothetical unknown cells, the remaining SES of tumor cells n710 The data was used as training data to construct scatter plots, and the above method was applied to perform binary classification on test data of normal cells and tumor cells. The results show that the classification accuracies of the k-nearest neighbor algorithm, decision tree algorithm, and random forest algorithm are 69.5%, 70.9%, and 72.4%, respectively. This analysis confirms the effectiveness of SES-based methods. n710 This allows for numerical classification of normal cells and tumor cells. This embodiment demonstrates that, although the light scattering spectra of normal cells and tumor cells obtained from actual patients are difficult to interpret from a physics perspective, they can be distinguished through machine learning.

[0101] In this analysis, although only the first and second principal components were considered, for example, by further considering higher-order principal components for classification, the classification accuracy between normal cells and tumor cells can be improved by considering the microscopic shape features of light scattering spectra that are imperceptible to the human eye, thus achieving high precision. Furthermore, if the same analytical method as light scattering spectroscopy is used, [the analysis can be performed on...]. Figure 3 The vector information of the dark field image and the frequency information obtained after the two-dimensional Fourier transform are processed, and this information is used as a feature quantity and incorporated into the learning model along with the light scattering spectrum, which is expected to further improve the accuracy.

[0102] Furthermore, in pathological diagnosis, the important objective is not merely to distinguish whether individual cells in a pathological specimen are normal or abnormal, but rather to diagnose the presence of abnormal cells. For example, even if the accuracy rate for identifying a single cell is only 70%, if hundreds of cells in a cytological specimen are differentiated and the results are statistically analyzed, the presence of abnormal cells in the specimen can be diagnosed with near 100% accuracy, thus achieving a high level of diagnostic capability. To improve the accuracy of pathological specimen diagnosis, not only is the accuracy rate for identifying single cells crucial, but the measurement speed and precision of light scattering spectroscopy are equally important. Moreover, if a large number of normal cells can be efficiently removed before large-scale measurements, or if the number of abnormal cells in the specimen can be predicted, the diagnostic accuracy will be significantly improved. The results of this experiment demonstrate that this method holds promise for achieving high-sensitivity and high-precision pathological diagnosis that surpasses traditional techniques.

[0103]

Example 3

[0104] In this embodiment, similar to Embodiment 2, normal cells and tumor cells collected from actual patients were used, and pathological samples were fixed and stained according to conventional pathological diagnostic procedures. The difference from Embodiment 2 is that this embodiment also considers higher-order principal components, demonstrating the results of discriminant analysis of light scattering spectra using machine learning.

[0105] Similar to Example 2, the normalized light scattering efficiency (SES) spectra of cells contained in specimens A to I were analyzed. n710 ) (refer to Figure 13 Principal component analysis was performed to obtain the loadings of the 1st to 10th principal components. Figure 14 The contribution rates and loading results for the first principal component (PC1), the second principal component (PC2), ..., and the tenth principal component are presented. Based on these loadings, the SES of a single cell is calculated. n710 The percentage composition (score) of the nth principal component (n is 1 to 10) and the mth principal component (m is 1 to 10) contained in the sample. Figure 15This shows a set of scatter plots with the score of the nth principal component on the horizontal axis and the score of the mth principal component on the vertical axis. Figure 15 A total of 100 scatter plots were presented, including those for the same principal components. Since swapping the vertical and horizontal axes of each principal component yields essentially the same information, 45 scatter plots were validated for the nth and mth principal components. Figure 15 The upper right corner of the matrix), where n < m, and it was proven that by using the scatter plot obtained by the second principal component (PC2) and the sixth principal component (PC6), the differences between normal cells and tumor cells can be clearly distinguished.

[0106] The scatter plots of the scores of the second and sixth principal components show that there are significant differences between normal cells and tumor cells. However, by changing the cell type or adjusting the analysis parameters, other components may also show significant differentiation.

[0107] As mentioned above, research shows that scatter plots of the scores from the second and sixth principal components can clearly distinguish between normal cells and tumor cells. Furthermore, these scatter plots also suggest that they may reflect the malignancy of tumor cells. Figure 16 As shown, the difference between tumor cells and normal cells is not a simple "yes / no" judgment, but rather there exists a transitional state between the two, which may characterize the malignancy of tumor cells. Cells that are clearly diagnosed as mesothelioma by pathologists and light scattering spectroscopy (including cells with depigmentation), cells whose scores are close to the origin and difficult to determine in light scattering spectroscopy, and cells that are clearly diagnosed as reactive mesothelioma by both methods can all be used as indicators of the malignancy of tumor cells: a distribution trend shifting to the upper left indicates a higher degree of malignancy, while a shift to the lower right indicates a lower degree of malignancy.

[0108] Therefore, by building up learning data, when pathologists are uncertain about the diagnostic results, they can refer to the analysis results of light scattering spectroscopy to assist in the pathological diagnosis based on image data. Furthermore, with the continuous accumulation of learning data, the discrimination accuracy based on light scattering spectroscopy is expected to further improve, ultimately enabling a fully automated diagnostic system that does not rely on pathologists. For example, ... Figure 17 As shown in (1), light scattering spectral data can be obtained from unknown cells (pathologists use cell shape information above 1 μm based on image data of unknown cells for diagnosis), and intracellular structural information below 1 μm can be obtained from the light scattering spectrum. Then, as... Figure 17As shown in (2), a score is calculated using light scattering spectral data and compared with the learning dataset to assess the malignancy of cells, thus providing auxiliary support for pathological diagnosis. Compared with image-based diagnostic methods, the light scattering spectral score calculation process is simpler and can distinguish tumor cells with higher sensitivity and accuracy. It should be noted that image data and light scattering spectral data contain different information and are expected to improve diagnostic accuracy in a complementary manner. By combining image learning data and light scattering spectral learning data, it is hoped that the accuracy of determining the malignancy of cells can be improved.

[0109]

Example 4

[0110] This implementation demonstrates the results obtained by identifying unstained live cells as pathological specimens using light scattering spectroscopy. The experiment acquired optical microscopic images of two live cell lines: Jurkat T cells and U937 mononuclear cells, as well as normalized light scattering efficiency (SES) spectra of individual cells. n710 ). Figure 18 (1) The optical microscope image shows that it is difficult to effectively distinguish between these two types of cells based solely on morphological characteristics. However, Figure 18 (2) In the normalized light scattering efficiency spectrum shown, although the upper spectrum shows morphological differences, the lower spectrum is not very distinguishable.

[0111] right Figure 18 (2) shows the normalized light scattering efficiency spectrum (SES). n710 Principal component analysis was performed to obtain the loadings from the 1st to the 10th principal components, and individual cell SES were determined. n710 The scores of the nth principal component (n is 1 to 10) and the mth principal component (m is 1 to 10) are obtained as follows: Figure 19 The scatter plot is shown. The results show that in almost all principal components, the two cell types exhibited significantly different score distributions, and all principal components up to the 10th principal component (PC10) clearly distinguished the two cell types. This embodiment demonstrates superior results compared to Examples 2 and 3 above, which used stained specimens, thus clearly revealing that the technology of this invention is also applicable to living cells. Furthermore, the ability to clearly identify unstained living cells effectively verifies the scientific rationality of the hypothesis proposed in Example 1.

[0112] Industrial application

[0113] According to the present invention, the identification process of tumor cells that are traditionally difficult to distinguish by pathologists through optical microscopy can be simplified, potentially enabling diagnosis that is unaffected by the experience and skill level of pathologists and improving diagnostic accuracy. Furthermore, through systematic measurement of light scattering spectra, comprehensive information about cell characteristics can be obtained more sensitively and rapidly than optical microscopic images, potentially driving the automation and efficiency of pathological diagnosis.

[0114] Explanation of symbols in the diagram

[0115] 1. Microscopic Scattering Spectroscopy Measurement System

[0116] 2. Cell culture dishes or specimens

[0117] 3 Dark Field Concentrating Lens

[0118] 3' Brightfield Condensing Lens

[0119] 4 Objectives

[0120] 5. Light source (halogen lamp)

[0121] 6. Spectrometer

[0122] 7CMOS camera

[0123] 10 cell culture dishes

[0124] 11 gaskets

[0125] 12 culture media

[0126] 13 coverslips

[0127] 14. Cultured Cells

[0128] 20-cell identification device

[0129] 21 Scattering Spectrum Acquisition Unit

[0130] 22 Learning Model

[0131] 23. Differentiation Unit for Normal and Abnormal Cells

[0132] 31 Electromagnetic wave scattering spectroscopy measurement device

[0133] 32 Scattering Spectrum

[0134] 33 Display device for identification results

[0135] 41, 46 Excitation Light

[0136] 42 samples (scatterers)

[0137] 43 Mie scattering and Rayleigh scattering

[0138] 44 Raman scattering (or fluorescence)

[0139] 45 spectrometer

[0140] 47 Optical Filter (Excitation Light Cut-off Filter)

Claims

1. A method for identifying cells by irradiating an electromagnetic wave to a cell-containing sample collected from a patient, obtaining an electromagnetic wave scattering spectrum of the sample, and identifying normal cells and abnormal cells using a learning model that has learned electromagnetic wave scattering spectrum characteristics of normal cells and abnormal cells.

2. A method for identifying cells by irradiating an electromagnetic wave containing all wavelengths in a specific wavelength range to a cell-containing sample collected from a patient, obtaining an electromagnetic wave scattering spectrum of Rayleigh scattering and Mie scattering of the sample, and identifying normal cells and abnormal cells using a learning model that has learned electromagnetic wave scattering spectrum of normal cells and abnormal cells.

3. The method for identifying cells according to claim 1 or 2, wherein the electromagnetic wave scattering spectrum is two-dimensional data showing a relationship between scattering electromagnetic wave intensity and electromagnetic wave wavelength, and the learning model is realized by learning the two-dimensional data, thereby classifying differences between normal cells and abnormal cells.

4. The method for identifying cells according to claim 3, wherein the learning model extracts a characteristic amount of the two-dimensional data using principal component analysis, extracts principal components by diagonalization processing of a variance-covariance matrix of scattering intensity at each wavelength, thereby classifying differences between characteristic amounts of normal cells and abnormal cells.

5. The method for identifying cells according to claim 4, wherein in the principal component analysis, the first principal component is a characteristic vector with the largest eigenvalue, and is a component that maximizes the variance of scattering intensity, the second principal component is a characteristic vector with the second largest eigenvalue, and is a component that is orthogonal to the first principal component and maximizes the variance of scattering intensity, and the nth principal component is a characteristic vector with the nth largest eigenvalue, and is a component that is orthogonal to the (n-1)th principal component and maximizes the variance of scattering intensity.

6. The method for identifying cells according to claim 1 or 2, wherein the sample is a sample that has been fixed and stained with ethanol or formalin.

7. The method for identifying cells according to claim 1 or 2, wherein the sample is a living cell.

8. The method for identifying cells according to claim 1 or 2, wherein the electromagnetic wave scattering spectrum is a forward scattering spectrum.

9. The method for identifying cells according to claim 1 or 2, wherein the wavelength range of the irradiated electromagnetic wave is 100 nm to 1 µm.

10. The method for identifying cells according to claim 1 or 2, wherein the irradiated electromagnetic wave is visible light.

11. The method for identifying cells according to claim 1, wherein the electromagnetic wave scattering spectrum is a combination of one or both of scattering spectra generated by Rayleigh scattering and Mie scattering.

12. The method for identifying cells according to claim 2 or 11, wherein the difference between the electromagnetic wave scattering spectrum between normal cells and abnormal cells is a scattering enhancement or weakening phenomenon due to a change in the size of the cell structure, and the size change is less than or equal to the length of a visible light wave.

13. The method for identifying cells according to any one of claims 1, 2, or 11, wherein the abnormal cells are tumor cells.

14. The method for identifying cells according to claim 13, wherein ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ The tumor cells are cancer cells.

15. The cell discrimination method according to claim 1 or 2, wherein, The electromagnetic wave for irradiation is ultraviolet or X-ray.

16. The cell discrimination method according to claim 1 or 2, wherein, The electromagnetic wave for irradiation is infrared.

17. A cell discrimination device comprising: a scattered light spectrum acquisition unit that acquires an electromagnetic wave scattered light spectrum by irradiating an electromagnetic wave to a cell-containing sample collected from a patient; a learning model obtained by learning electromagnetic wave scattered light spectrum data of normal cells and abnormal cells; and a discrimination unit that discriminates normal cells and abnormal cells using the learning model.

18. A cell discrimination device comprising: an irradiation unit that irradiates an electromagnetic wave containing all wavelengths within a specific wavelength range to a cell-containing sample collected from a patient; an electromagnetic wave scattered light spectrum acquisition unit that acquires Rayleigh scattering and Mie scattering; a learning model obtained by learning electromagnetic wave scattered light spectrum data of normal cells and abnormal cells; and a discrimination unit that discriminates normal cells and abnormal cells using the learning model.

Images