Label-free single-cell classification system and method based on light scattering and phase imaging

Abstract

The application discloses a label-free single-cell classification system and method based on light scattering and phase imaging, relates to the technical field of biomedical optical detection, and comprises a laser emission unit, a sample conveying unit, a light scattering imaging acquisition unit, a quantitative phase imaging acquisition unit, a data processing and analysis unit, wherein the rear and side of a sample detection area are respectively arranged to form a quantitative phase imaging light path and a lateral light scattering imaging light path, and the two imaging light paths form a stereo orthogonal double light path with the sample detection area as an intersection point; homologous triggering of the two imaging channels is realized through an internal timer of the same software, the two channels are collected in parallel at the same frame frequency, two-dimensional lateral light scattering images and quantitative phase images of single cells are respectively acquired, time mismatch problems caused by time-sharing switching or sequential scanning are effectively avoided, label-free correlation acquisition of single cell morphology and physical information is realized, and high-precision identification and automatic classification of different cell subtypes are realized based on machine learning.

Description

Technical Field

[0001] This invention relates to the field of biomedical optical detection technology, and in particular to a label-free single-cell classification system and method based on light scattering and phase imaging. Background Technology

[0002] Cells are the basic structural and functional units of living organisms. Morphological and biophysical differences at the single-cell level serve as the core basis for distinguishing disease subtypes from healthy cells, assisting physicians in diagnosing and classifying diseases, and providing a reliable basis for clinical treatment planning and prognostic assessment. Taking leukemia as an example, traditional leukemia classification relies on techniques such as flow cytometry and fluorescence in situ hybridization to differentiate between four subtypes: acute lymphoblastic leukemia, chronic lymphoblastic leukemia, acute myeloid leukemia, and chronic myeloid leukemia. While this effectively improves the classification ability of leukemia subtypes, it is costly, time-consuming, and requires specialized fluorescent labeling or invasive procedures. These procedures alter the natural physiological state of cells, distort original biophysical characteristics, and introduce detection noise, affecting the accuracy of single-cell leukemia classification.

[0003] To avoid the damage to single cells caused by fluorescent labeling and chemical staining, the industry is gradually developing label-free single-cell optical imaging technologies that can acquire cellular information without labeling. Two-dimensional light scattering imaging technology acquires two-dimensional light scattering images through an imaging sensor within the lateral scattering angle range. These images can provide detailed information on organelles such as mitochondria, offering a comprehensive perspective for label-free single-cell analysis. Holographic quantitative phase imaging technology provides label-free phase quantitative analysis, enabling precise characterization of cell morphology and pathology. However, existing single-modal label-free optical technologies can only provide partial biophysical information. For example, two-dimensional light scattering imaging technology cannot directly invert quantitative parameters (such as phase distribution) from the scattering image, and the spatial resolution of holographic quantitative phase imaging technology may be limited, which will affect the accuracy of the final typing.

[0004] To improve typing accuracy, existing studies have attempted to integrate light scattering and digital holography into single-cell imaging for fusion analysis and accurate typing. However, this approach relies on a sequential acquisition architecture: a single cell must undergo forward small-angle light scattering and digital holography sequentially. Light scattering is limited to a small forward angle range (usually 2-30°). Combining digital holography to capture the macroscopic features of a single cell may miss some detailed information. More importantly, the two modalities of a single cell are located in different spatial positions, and the acquisition process is sequential. Data fusion analysis is a statistical cross-validation of the same cell population, which cannot simultaneously acquire the two modalities of the same single cell. That is, the image and the holographic phase image cannot be accurately matched to the same single cell. The dual-modal fusion analysis data is inherently distorted, affecting typing accuracy. Summary of the Invention

[0005] To address the problem of insufficient multimodal optical imaging of the same single cell in existing technologies, which affects the accuracy of single-cell classification, this invention provides a label-free single-cell classification system and method based on light scattering and phase imaging. By integrating and optimizing the optical path of light scattering and digital holography, two acquisition branches are designed to acquire two-dimensional light scattering images and interference images containing phase information of the same single cell in parallel within the same optical path. This avoids problems such as acquisition time mismatch and difficulty in matching dual-modal images. Subsequently, a quantitative phase image of the single cell is obtained through numerical reconstruction. Multidimensional features are then extracted from both modal images, and a machine learning classifier is used to achieve automatic and accurate identification and classification of single cells. This method can be used for label-free classification of normal cells and disease cells such as multi-subtype leukemia cells, providing a reliable basis for assisting physicians in disease diagnosis, disease classification, and clinical treatment plan formulation.

[0006] In a first aspect, the present invention provides a label-free single-cell classification system based on light scattering and phase imaging.

[0007] A label-free single-cell classification system based on light scattering and phase imaging, comprising: The laser emitting unit is used to generate a single coherent laser beam, which serves as the common excitation source for the system and is horizontally incident on the sample detection area. The sample delivery unit is used to deliver the cell suspension to the sample detection area at a constant low flow rate; The quantitative phase imaging acquisition unit is located behind the sample detection area along the horizontal main axis of the laser, forming a quantitative phase imaging optical path. It is used to record the interferogram and obtain a quantitative phase image of a single cell through numerical reconstruction. The light scattering imaging acquisition unit is located on the side of the sample detection area, forming a side scattering imaging optical path, which is used to collect two-dimensional scattered light from single cells in parallel within the side angle range and generate a scattering image. The data processing and analysis unit is used to extract multi-dimensional features from phase images and scattering images respectively. After feature fusion, the features are input into a machine learning classifier, and the output is the automatic single-cell typing result. Among them, the two imaging optical paths intersect at the sample detection area to form a three-dimensional orthogonal dual optical path.

[0008] In a further technical solution, the laser emitting unit includes a laser, a beam expander, a pinhole filter, and a first microscope objective connected in sequence via optical paths. The laser emits a single coherent laser beam, which is expanded by the beam expander and filtered by the pinhole filter in sequence. The first microscope objective then focuses the laser beam onto the sample detection area. After passing through the single cells of the cell suspension in the sample detection area, the laser beam enters the two branch optical paths corresponding to the quantitative phase imaging acquisition unit and the light scattering imaging acquisition unit, respectively.

[0009] A further technical solution is provided, wherein the sample delivery unit includes a square microglass capillary sample chamber and a high-precision injection pump; the cell suspension concentration is diluted to 1500 particles / mL, and the flow rate of the cell suspension is controlled at 10μL / h by the high-precision injection pump, so that the cell suspension is delivered to the sample detection area of ​​the sample chamber at a constant low flow rate; the sample chamber is also equipped with a three-axis displacement adjustment mechanism to adjust the single cell to always be in the focal plane during the image acquisition process.

[0010] A further technical solution is that the quantitative phase imaging acquisition unit includes a second microscope objective, a beam splitter cube, a first reflecting mirror, a second reflecting mirror, a first lens barrel, and a first detector arranged sequentially along the main axis of laser horizontal transmission. The transmitted light passing through the single cell is magnified by the second microscope objective and enters a self-interference device based on a Michelson interferometer. The beam splitter cube in the device splits the transmitted light into two coherent beams from the same source. The two beams are emitted in the vertical and horizontal directions, respectively, and are reflected by the first reflecting mirror and the tilted second reflecting mirror, and then reflected back to the beam splitter cube to be re-recombined. After coherent superposition, an interference beam with a spatial carrier is formed. The recombined interference beam is converged by the first lens barrel and projected onto the first detector. The first detector records the holographic interferogram and transmits it to the computer for numerical reconstruction to obtain a quantitative phase image of the single cell. The second reflector is tilted at a small angle to generate a spatial carrier stripe pattern, causing the two beams of light to be spatially displaced on the camera plane, and the displacement is greater than the diameter of the cell sample.

[0011] Further technical solutions, including reconstructing quantitative phase maps from interferograms, include: The holographic interferogram is subjected to Fourier transform, and the required frequency components are retained by frequency domain filtering to obtain the preprocessed data; The complex field is reconstructed by inverse Fourier transform of the preprocessed data and the phase is calculated. Then, the 2π phase ambiguity is eliminated by a reliability-guided two-dimensional phase unwrapping algorithm for discontinuous paths to obtain the object light phase. Using the holographic interferogram acquired in the empty sample chamber as a reference, the reference phase is extracted, and the reference phase is subtracted from the object light phase to obtain the true phase difference distribution; Based on the true phase difference distribution, the average refractive index and cell dry mass of the cells were calculated respectively.

[0012] A further technical solution is that the light scattering imaging acquisition unit includes a third microscope objective, a second microscope lens, and a second detector arranged sequentially from top to bottom; single cells in the cell suspension in the sample detection area generate two-dimensional lateral scattered light from the laser beam, and the scattered light within a specific lateral angle range is collected by the third microscope objective on the lower side of the sample detection area. The amplified scattered beam continues to be transmitted downward to the second microscope lens for beam collimation and convergence to correct imaging distortion. The converged scattered light is finally input vertically downward into the second detector to complete the acquisition of the two-dimensional lateral scattering image; In this process, the exposure times for scattering imaging and holographic imaging are set independently. The dual-modal image acquisition is triggered by the same timer from the same source and acquired in parallel at the same frame rate. The trigger time deviation is much smaller than the residence time of the cell in the sample detection area, so as to ensure that the dual-modal image corresponds to the same single cell.

[0013] A further technical solution is that the data processing and analysis unit includes, in sequence: The feature extraction module is configured with two independent feature extraction logics to extract multi-dimensional features from the scattering image and the phase image, respectively. The extracted features of the scattering image include first-order statistical features and second-order texture features. The second-order texture features include gray-level co-occurrence matrix features, gray-level gradient co-occurrence matrix features, and gray-level run-length matrix features. The extracted features of the phase image include geometric morphological features and multi-dimensional features that are homologous to the scattering image. The feature fusion module is used to concatenate the multi-dimensional features extracted from the dual-modal images to form a multi-dimensional multimodal fusion feature vector; The classification module is used to input the fused multi-dimensional and multi-modal fused feature vector into the machine learning classifier, and uses ten-fold cross-validation to output the automatic classification result of a single cell; The machine learning classifier includes at least one of linear discriminant analysis, support vector machine, logistic regression, K-nearest neighbors, and decision tree, and the feature selection strategy used in the machine learning training process includes at least one of relevance feature selection, recursive feature elimination, and full feature retention.

[0014] Secondly, the present invention provides a label-free single-cell classification method based on light scattering and phase imaging.

[0015] A label-free single-cell classification method based on light scattering and phase imaging, implemented using the aforementioned label-free single-cell classification system based on light scattering and phase imaging, includes: The cell suspension to be tested is delivered to the sample detection area of ​​the sample chamber at a constant ultra-low flow rate; The laser is activated to acquire two-dimensional side-scattering images and holographic interferograms of the same single cell in parallel at a set frame rate; Numerical reconstruction of the holographic interferogram yields a quantitative phase image of the cell; Multi-dimensional features are extracted from the scattering image and the phase image respectively. After feature splicing and fusion, the results are input into a machine learning classifier and the output is the automatic single-cell typing result.

[0016] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing the above-described label-free single-cell classification method based on light scattering and phase imaging when executing the executable instructions stored in the memory.

[0017] Fourthly, the present invention also provides a computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the above-described label-free single-cell classification method based on light scattering and phase imaging.

[0018] The above one or more technical solutions have the following beneficial effects: This invention provides a label-free single-cell classification system and method based on light scattering and phase imaging. A single laser is used as a common excitation source to illuminate the sample detection area. Two parallel acquisition branches (i.e., two parallel imaging optical paths) are designed on the output side of this area. These two imaging optical paths intersect at the sample detection area, dividing into a horizontal transmission holographic optical path and a vertical lateral scattering optical path, forming a three-dimensional orthogonal co-source imaging structure. This allows for the parallel acquisition of a two-dimensional light scattering image and an interference image containing phase information of the single cell along the same optical path (the interference image, after numerical reconstruction, yields a quantitative phase image of the single cell). This provides a physical-level... This invention eliminates the effects of occlusion and stray light crosstalk between the two types of optical paths, and simultaneously employs ultra-low flow rate microfluidic delivery of cell suspension to the sample detection area. This ensures that the same cell maintains spatial position matching within the image acquisition windows of the two imaging channels, guaranteeing a precise one-to-one correspondence between single-cell dual-modal images. This ensures that the two modal data acquired in parallel originate from the same single cell, achieving accurate correspondence between the two modal data. It utilizes lateral scattering imaging to capture submicron particle details within the cell and quantitative phase imaging to accurately measure biophysical parameters such as cell refractive index and dry mass. The two modal features are complementary, and after feature fusion, machine learning is used to complete accurate cell typing. All operations in this invention require no fluorescent labeling or chemical staining, preserving the original state of the cell. The operation is simple and low-cost, applicable to the identification and classification of other biological cells, and has broad applicability. Compared to traditional single-modal optical imaging methods or existing dual-modal fusion analysis methods, the single-cell typing and identification accuracy and stability of this invention are significantly improved. It can be directly used for the clinical label-free rapid identification of hematological malignant tumor cells (such as leukemia cells), providing a reliable basis for assisting physicians in disease diagnosis, disease typing, and clinical treatment plan formulation.

[0019] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1 This is a schematic diagram of the structure of a label-free single-cell classification system based on light scattering and phase imaging in an embodiment of the present invention; Figure 2 This is a flowchart of a label-free single-cell classification method based on light scattering and phase imaging, as described in an embodiment of the present invention. Figure 3 This is a schematic diagram of the two-dimensional light scattering experimental results of three microspheres with different diameters in an embodiment of the present invention; wherein, (a) is a diameter of 3.87 μm, (b) is a diameter of 4.19 μm, and (c) is a diameter of 3.93 μm; Figure 4 The above are simulated scattering diagrams of three microspheres based on Mie theory in an embodiment of the present invention. Figure 5 This is a spectral matching diagram of two-dimensional light scattering experiments and simulations of three types of microspheres in an embodiment of the present invention; Figure 6 This is a schematic diagram of the interference fringes of three types of microspheres in an embodiment of the present invention; Figure 7 These are reconstructed phase diagrams of three types of microspheres in an embodiment of the present invention; Figure 8 This is a scatter plot of the refractive index of three microspheres relative to their diameter in an embodiment of the present invention; Figure 9 This is a schematic diagram of imaging of chronic lymphocytic leukemia cells in an embodiment of the present invention; wherein, (a) is an interferogram, (b) is a reconstructed quantitative phase map, (c) is a three-dimensional phase rendering map, and (d) is a two-dimensional light scattering map; Figure 10 This is a schematic diagram of imaging of chronic myeloid leukemia cells in an embodiment of the present invention; wherein, (a) is an interferogram, (b) is a reconstructed quantitative phase map, (c) is a three-dimensional phase rendering map, and (d) is a two-dimensional light scattering map; Figure 11 This is a schematic diagram of normal cell imaging in an embodiment of the present invention; wherein, (a) is an interferogram, (b) is a reconstructed quantitative phase map, (c) is a three-dimensional phase rendering map, and (d) is a two-dimensional light scattering map; Figure 12This is the optimal three-class confusion matrix based on multimodal features in this embodiment of the invention; where CLL represents chronic lymphocytic leukemia cells, CML represents chronic myeloid leukemia cells, and NCs represent normal cells; Figure 13 The first 25 features used in this embodiment of the invention to distinguish between chronic lymphocytic leukemia cells and normal cells are importance scores; where LSI represents the light scattering module and QPI represents the quantitative phase module. Figure 14 The importance scores of the first 25 features used to distinguish chronic myeloid leukemia cells from normal cells in this embodiment of the invention; Figure 15 This is a heatmap of the correlation matrix of 148 features extracted in this embodiment of the invention; Figure 16 The image shows the violin distribution of four representative biophysical characteristics among different cell types in this embodiment of the invention; where (a) is refractive index, (b) is dry mass, (c) is surface area, and (d) is smoothness.

[0022] Among them, 1. Laser; 2. Beam expander; 3. Pinhole filter; 4. First microscope objective; 5. Sample chamber; 6. Second microscope objective; 7. First reflecting mirror; 8. Beam splitter cube; 9. Second reflecting mirror; 10. First microscope tube lens; 11. First detector; 12. Third microscope objective; 13. Second microscope tube lens; 14. Second detector; 15. Injection pump. Detailed Implementation

[0023] It should be noted that the following detailed descriptions are exemplary and are intended only to describe specific embodiments and to provide further explanation of the invention, and are not intended to limit the scope of exemplary embodiments of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0024] As noted in the background section, existing single-modal label-free optical detection techniques have low typing accuracy due to limitations in extracting cell feature information. The proposed multimodal fusion analysis methods integrating light scattering and digital holography are also limited by sequential acquisition architecture, making it impossible to accurately match scattering images and holographic phase images to the same single cell, resulting in distortion in multimodal data fusion analysis.

[0025] To address the aforementioned issues, an approach was considered to integrate light scattering and holographic phase imaging into a single coaxial optical path for parallel dual-modal image acquisition. However, this simple integration introduces several interconnected and difficult-to-reconcile core problems: Firstly, the coaxial optical path design for forward small-angle light scattering imaging and holographic imaging suffers from several issues. On one hand, the simple stitching of dual light sources results in systemic deviations due to inconsistent light intensity and wavelength references. Secondly, the lack of physical isolation within the coaxial optical path allows strong diffuse reflection light generated by scattering to potentially intrude into the holographic interferometry module, affecting subsequent measurement results. Furthermore, the stringent alignment requirements of off-axis holography in a coaxial layout interfere with the scattering imaging optical path, making synchronous assembly and adjustment of the two optical paths extremely difficult. Furthermore, to ensure the accuracy of the acquired dual-modal image data, time-division switching is still required even in a coaxial design, resulting in unsynchronized acquisition timing.

[0026] To address this, this invention innovatively optimizes the single coaxial optical path design, integrating lateral scattering acquisition on the basis of holographic phase imaging, and using a single-beam coherent laser to construct the imaging system: a single coherent laser beam output from the laser emitting unit is horizontally incident on the sample detection area; the transmitted light passing through the cells is transmitted along the horizontal principal axis to form a quantitative phase imaging optical path; the lateral scattered light from the cells is received by the acquisition components vertically arranged on the side of the sample to form a lateral scattering imaging optical path; the horizontal transmission optical path and the vertical scattering optical path are separated from each other, together forming a three-dimensional orthogonal dual-optical-path structure; the two main imaging optical paths intersect only at the cell detection point, relying on spatial layering and the lateral scattering acquisition angle (…). The system achieves natural stray light isolation (79°~101°) to eliminate crosstalk interference between optical paths. Building upon this, ultra-low flow rate microfluidic control is used to deliver the cell suspension to the confluence point, ensuring spatial matching of the same single cell within the image acquisition windows of the two imaging channels. This guarantees a precise one-to-one correspondence between the dual-modal images of a single cell. Furthermore, scattering imaging and holographic imaging share the same software timer for co-source triggering and acquire data in parallel at the same frame rate. The triggering time deviation is much smaller than the cell's residence time within the detection area, further ensuring that the dual-modal images correspond to the same cell and effectively avoiding time mismatch issues caused by time-division switching or sequential scanning. Simultaneously, the lateral scattering branch specifically collects a wide range of lateral two-dimensional scattered light from 79°~101°, completely capturing intracellular microscopic particle textures and compensating for the lack of forward scattering information.

[0027] In summary, the system proposed in this invention uses only a single laser to unify the two imaging illumination references. The light source, holography, and scattering modules are independently pre-calibrated and fixed. Cell alignment and confocal adjustment of the two optical paths are completed synchronously using only a three-axis sample displacement stage, simplifying the complex assembly and adjustment process. At the same time, the holographic interference unit uses an integrated locking module to stably generate a spatial carrier, which can effectively reduce phase calculation errors. This system can achieve parallel acquisition of single-cell two-dimensional light scattering images and phase information interference images. Subsequently, single-cell phase images are obtained through numerical reconstruction. Then, 67-dimensional texture features of the scattering image and 81-dimensional features containing quantitative biophysical parameters of the phase image are extracted respectively. Feature fusion is completed based on the low correlation and complementary information characteristics of the two types of features. High-precision single-cell typing is achieved by combining multi-class machine learning classifiers and ten-fold cross-validation.

[0028] Example 1 This embodiment provides a label-free single-cell classification system based on light scattering and phase imaging. By integrating lateral scattering acquisition on the basis of holographic phase imaging and building an imaging system using a single-beam homogeneous laser, the two imaging channels are triggered from the same source using an internal timer in the same software, enabling the two channels to acquire data in parallel at the same frame rate. This obtains two-dimensional lateral light scattering images and phase information interference images of single cells, effectively avoiding time mismatch problems caused by time-division switching or sequential scanning. It enables label-free correlation acquisition of morphological and physical information such as single cell size, internal particle size distribution, and precise refractive index and dry mass, providing more universal multi-dimensional data support for high-precision identification and automated classification of different cell subtypes.

[0029] like Figure 1 As shown, the label-free single-cell classification system based on light scattering and phase imaging proposed in this embodiment mainly includes a laser emission unit, a sample delivery unit, a light scattering imaging acquisition unit, a quantitative phase imaging acquisition unit, and a data processing and analysis unit. The laser is transmitted along the horizontal principal axis to the quantitative phase imaging acquisition unit, thus forming the quantitative phase imaging optical path. Light scattering imaging acquisition units are arranged on the side of the sample detection area, forming a lateral light scattering imaging optical path. That is, the two imaging optical paths intersect at the sample detection area, forming a three-dimensional orthogonal dual optical path. The above units are configured as follows: 1) The laser emitting unit is used to generate a single coherent laser beam as the common excitation source of the system, which is horizontally incident on the sample detection area; 2) Sample delivery unit, used to deliver cell suspension to the sample detection area at a constant low flow rate; 3) A quantitative phase imaging acquisition unit is arranged behind the sample detection area along the horizontal principal axis of the laser. It is used to record the interferogram and obtain the quantitative phase image of a single cell through numerical reconstruction. 4) A light scattering imaging acquisition unit is located on the side of the sample detection area to collect two-dimensional scattered light from single cells in parallel within the lateral angle range and generate a scattering image; 5) Data processing and analysis unit, used to extract multi-dimensional features from phase image and scattering image respectively, and input them into machine learning classifier after feature fusion, and output single cell automatic typing results.

[0030] Specifically, the laser emitting unit includes a laser 1, a beam expander 2, a pinhole filter 3, and a first microscope objective 4 connected in sequence via optical paths. The laser 1 emits a single coherent laser beam, which is expanded by the beam expander 2 and filtered by the pinhole filter 3. The first microscope objective 4 then focuses the laser beam onto the cell sample in the sample detection area. After passing through the cell sample, the laser beam enters the two branch optical paths corresponding to the quantitative phase imaging acquisition unit and the light scattering imaging acquisition unit, respectively.

[0031] In this embodiment, the laser 1 is a Crystal Laser CL532-075 from the United States; the beam expander 2 is a Zolix LBE532-10 from China; the pinhole filter 3 is a Thorlabs SM1D12 from the United States; and the first microscope objective 4 is an Olympus PLN4X from Japan.

[0032] The light scattering imaging acquisition unit includes a third microscope objective 12, a second microscope lens 13, and a second detector 14 arranged sequentially from top to bottom. In the sample detection area, cell samples are subjected to two-dimensional lateral scattering light generated by a laser beam. The third microscope objective 12, located on the side of the sample detection area, collects the scattered light within a specific lateral angle range. The magnified scattered beam is then transmitted downwards to the second microscope lens 13 for beam collimation and focusing to correct imaging distortion. The focused scattered light is finally input vertically downwards into the second detector 14, completing the acquisition of the two-dimensional lateral scattering image.

[0033] Furthermore, the distribution of scattered light varies with the polar angle. θ The azimuth angle φ (assuming the laser incident along the Z-axis, the Y-axis perpendicular to the detector acquisition plane, the azimuth angle is the angle between the scattered light projection onto the XY plane and the X-axis, and the polar angle is the spatial angle between the scattered light vector and the laser incident principal axis) is determined by the numerical aperture of the objective lens. Under the experimental conditions of this embodiment, the polar angle range is 79°~101°, the defocus distance is 200μm, and the scattering imaging exposure time is 14ms.

[0034] In this embodiment, the third microscope objective 12 is of the model number: Olympus, Japan, PLN10X; the second microscope tube lens 13 is of the model number: Thorlabs, USA, TTL180-A; and the second detector 14 is of the model number: MQ013CG-E2, Germany.

[0035] The quantitative phase imaging acquisition unit includes a second microscope objective 6, a beam splitter cube 8, a first reflecting mirror 7, a second reflecting mirror 9, a first lens barrel 10, and a first detector 11, arranged sequentially along the main axis of laser horizontal transmission. The transmitted light passing through the single cell is magnified by the second microscope objective 6 and enters a self-interference device based on a Michelson interferometer. The beam splitter cube 8 splits the transmitted light into two coherent beams from the same source. These two beams are emitted vertically and horizontally, respectively, and are reflected by the first reflecting mirror 7 and the tilted second reflecting mirror 9, before returning to the beam splitter cube to recombine. Through coherent superposition, they form interference light with a spatial carrier wave. The recombinated interference light is converged by the first lens barrel 10 and projected onto the first detector 11. The first detector 11 records a holographic interferogram (containing phase information) and transmits the interferogram to a computer (e.g., to a data processing and analysis unit) via a USB 3.0 interface for phase reconstruction, obtaining a quantitative phase image of the single cell. Preferably, the holographic exposure time is 60 μs. It should be noted that the second reflector 9 is set at a small angle, that is, the second reflector 9 is slightly tilted to generate a spatial carrier stripe pattern. This tilt causes the two beams of light to produce spatial displacement on the camera plane, and the displacement is greater than the cell diameter.

[0036] As one implementation method, reconstructing a quantitative phase map based on an interferogram includes: performing a Fourier transform on the aforementioned holographic interferogram (i.e., the phase information interferogram), and retaining the required frequency components through frequency domain filtering to obtain preprocessed data; performing an inverse Fourier transform on the preprocessed data to reconstruct the complex field and calculate the phase, and then using a reliability-guided discontinuous path two-dimensional phase unwrapping algorithm to eliminate 2π phase ambiguity to obtain the object light phase; using a holographic interferogram acquired in an empty sample chamber as a reference, extracting the reference phase, and subtracting the reference phase from the object light phase to obtain the true phase difference distribution; and calculating the values ​​of biophysical parameters such as the average refractive index and cell dry mass based on the true phase difference distribution.

[0037] In this embodiment, the exposure times for scattering imaging and holographic imaging are set independently, and their acquisition is triggered by the same software's SWtrigger from the same source. They are acquired in parallel at the same frame rate, and the trigger time deviation is much smaller than the cell's residence time in the detection area, ensuring that the dual-modal images correspond to the same cell.

[0038] In this embodiment, the second microscope objective 6 is of the model number: Olympus, Japan, PLN40X; the beam splitter cube 8 is of the model number: Thorlabs, USA, BS013; the first reflecting mirror 7 is of the model number: Thorlabs, USA, BB1-E02; the second reflecting mirror 9 is of the model number: Thorlabs, USA, BB1-E02; the first tube lens 10 is of the model number: Thorlabs, USA, TTL180-A; and the first detector 11 is of the model number: MQ013CG-E2, Germany.

[0039] The sample delivery unit includes a square microglass capillary sample chamber 5 and a high-precision syringe pump 15. The cell suspension is diluted to 1500 particles / mL, and the flow rate of the cell suspension is controlled at 10 μL / h by the high-precision syringe pump, delivering the cell suspension to the sample detection area of ​​the sample chamber at a constant flow rate. The sample chamber is also equipped with a three-axis displacement adjustment mechanism to ensure that the cells remain on the focal plane during image acquisition.

[0040] Specifically, the sample delivery unit uses a square microglass capillary chamber 5 with an inner diameter of 1000 μm and a wall thickness of 200 μm, driven by a syringe pump 15. The image acquisition rate is 60 frames / second, and the sample flow rate is strictly controlled at 10 μL / h to ensure that motion blur does not affect optical resolution or interferogram quality. Simultaneously, the sample is diluted to a concentration of 1500 particles / mL. Under these extremely low flow rate conditions, the cells flow slowly in the detection area, allowing the light scattering imaging unit and the quantitative phase imaging unit to capture both the scattering map and hologram of the same cell, thus achieving a one-to-one correspondence between the two modalities of the same cell. Preferably, to keep the cell always at the focal plane during image acquisition, compensation is achieved by manually adjusting the position of the sample chamber, effectively correcting potential vertical drift and minor changes in sample height.

[0041] In this embodiment, the sample chamber 5 is a Japanese VitroCom 8100-100; the syringe pump 15 is a British LongerPump LSP01-1A.

[0042] The data processing and analysis unit includes the following components arranged in sequence: 5.1) Feature extraction module, configured with two independent feature extraction logics, used to extract multi-dimensional features of scattering image and phase image respectively; wherein, the extracted features of scattering image include first-order statistical features and second-order texture features, the second-order texture features include gray-level co-occurrence matrix features, gray-level gradient co-occurrence matrix features, and gray-level run-length matrix features; the extracted features of phase image include geometric morphological features and multi-dimensional features of the same origin as scattering image; 5.2) Feature fusion module, used to concatenate the multi-dimensional features extracted from the dual-modal images to form a multi-dimensional multimodal fusion feature vector; 5.3) The classification module is used to input the fused multi-dimensional and multi-modal fused feature vector into the machine learning classifier, and use ten-fold cross-validation to output the automatic classification result of a single cell; wherein, the machine learning classifier includes at least one of linear discriminant analysis, support vector machine, logistic regression, K-nearest neighbors, and decision tree, and the feature selection strategy used in the machine learning training process includes at least one of relevance feature selection, recursive feature elimination, and full feature retention.

[0043] Example 2 This embodiment proposes a label-free single-cell classification method based on light scattering and phase imaging, implemented using the aforementioned label-free single-cell classification system based on light scattering and phase imaging. Figure 2 As shown, it includes the following steps: Step S1: The cell suspension to be tested is delivered to the sample detection area of ​​the sample chamber at a constant ultra-low flow rate. Specifically, the cell suspension to be tested is injected into the sample chamber and stabilized at a fixed flow rate, for example, at a flow rate of 10 μL / h, with a sample concentration of 1500 particles / mL.

[0044] Step S2: Activate the laser to acquire two-dimensional side scattering images and holographic interferograms (i.e., phase information interferograms) of the same single cell in parallel at a set frame rate. Specifically, turn on the laser and acquire two-dimensional light scattering images and interferograms of the single cell at a frame rate of 60 frames per second. At the same time, by controlling the extremely slow flow rate of the sample, the same cell is recorded by two imaging units of different modalities as it flows through the detection area, thereby obtaining image data of the same cell in two modalities.

[0045] Step S3: Perform numerical reconstruction on the holographic interferogram to obtain a quantitative phase image of the cell.

[0046] Specifically, the numerical reconstruction process of the holographic interferogram is as follows: First, a Fourier transform is performed on the holographic interferogram, and frequency domain filtering is applied to retain the required frequency components while filtering out the zero-order and conjugate images; then, after frequency domain filtering, an inverse Fourier transform is performed to reconstruct the complex field F, through... The phase is calculated; then, a reliability-guided discontinuous path phase unwrapping algorithm is used to solve the 2π ambiguity problem; finally, a hologram of the empty sample chamber is recorded as a reference before each experiment to extract the reference phase. From the unwrapped object phase Subtract the reference phase of unpacking To obtain the true continuous phase difference distribution, it can be expressed as: ; Where Re represents the real part of the complex number, and Im represents the imaginary part of the complex number.

[0047] Furthermore, a spherical model is used to approximate the physical thickness of the cell. Specifically, an edge fitting algorithm is used to locate the cell center: based on a preset initial radius... An initial point is found in the phase map, then the abrupt change position of the phase edge of the corresponding spherical model is searched. The measured edge contour is matched with the theoretical spherical phase model through iterative fitting, and finally the fitted cell center coordinates are output simultaneously. Based on this, cell thickness is calculated using the following formula. : ; Average refractive index of the sample Calculated using the following formula: ; Where λ is the wavelength of the illumination light (532nm), The average phase difference, The average thickness of the sample. The refractive index of the surrounding medium (PBS solution) is 1.334.

[0048] dry mass of cells Calculated using the following formula: ; in, The refractive index increment is 0.18 mL / g.

[0049] The numerical reconstruction of the quantitative phase images was performed using Python (version 3.10.9). The average computation time for reconstructing a single hologram into a quantitative phase image was approximately 0.15 seconds, and the imaging throughput was approximately 6 to 7 cells per second.

[0050] Through the above numerical reconstruction, a quantitative phase image of the cell is obtained, and the values ​​of biophysical parameters such as cell refractive index and dry mass are calculated. These parameters show significant differences in cell classification and can be used as geometric features for subsequent cell classification.

[0051] Step S4: Extract multi-dimensional features from the scattering image and phase image respectively, and input them into the machine learning classifier after feature splicing and fusion to output the automatic single-cell typing result.

[0052] Specifically, 67 features are extracted from the light scattering image, including 17 first-order statistical features and 50 second-order texture features. The second-order texture features include 20 gray-level co-occurrence matrix features, 15 gray-level gradient co-occurrence matrix features, and 15 gray-level run-length matrix features. The specific extraction steps are as follows: Step S4.1: Preprocess the light scattering image, including: cropping, converting to grayscale, normalizing to the 0 to 1 pixel intensity range, uniformly scaling to 256×256 pixels, and applying a Gaussian filter ( =2, kernel size 15×15) were smoothed.

[0053] Step S4.2: Extract corresponding features from each preprocessed scattering image, including: The 17 first-order statistical features are: mean, median, range, minimum, maximum, variance, skewness, kurtosis, entropy, uniformity, 10th percentile, 90th percentile, interquartile range, mean absolute deviation, robust mean absolute deviation, root mean square, and energy. The 20 gray-level co-occurrence matrix features are: autocorrelation, contrast, correlation, cluster prominence, cluster shading, dissimilarity, energy, entropy, homogeneity, maximum probability, sum of squares, sum of averages, sum of variances, sum of entropy, difference variances, difference entropy, information measure correlation 1, information measure correlation 2, inverse normalization, and inverse moment normalization. The 15 features of the gray-level gradient co-occurrence matrix are: small gradient dominance, large gradient dominance, gray-level asymmetry, gradient asymmetry, energy, gray-level mean, gradient mean, gray-level variance, gradient variance, correlation, gray-level entropy, gradient entropy, mixed entropy, inertia, and homogeneity. The 15 grayscale run matrix features are: short run advantage, long run advantage, grayscale non-uniformity, grayscale non-uniformity normalization, run percentage, run length non-uniformity, run length non-uniformity normalization, run variance, grayscale variance, low grayscale run advantage, high grayscale run advantage, short run low grayscale advantage, short run high grayscale advantage, long run low grayscale advantage, and long run high grayscale advantage.

[0054] Similarly, 81 features were extracted from the quantitative phase image, including 14 geometric morphological features, 17 first-order statistical features, 20 gray-level co-occurrence matrix features, 15 gray-level gradient co-occurrence matrix features, and 15 gray-level run-length matrix features. The feature extraction process was as follows: after reconstructing and correcting the background of the phase image, the cell center was automatically detected using an edge fitting algorithm. Cell segmentation was achieved through adaptive thresholding and morphological operations to generate a binary mask. 14 geometric morphological features were calculated from the mask and the phase image: area, perimeter, major axis length, minor axis length, elongation, circularity, eccentricity, smoothness, volume, surface area, refractive index, dry mass, face-to-volume ratio, and sphericity. At the same time, 17 first-order statistical features and 50 second-order texture features (i.e., 20 gray-level co-occurrence matrix features, 15 gray-level gradient co-occurrence matrix features, and 15 gray-level run-length matrix features) were extracted. The types are the same as those mentioned above and will not be repeated here.

[0055] Finally, the 67 features from the light scattering image and the 81 features from the quantitative phase image are concatenated into an array to form a 148-dimensional multimodal feature vector. This fused feature vector is then input into a machine learning classifier, and 10-fold cross-validation is used to achieve automatic cell classification. Preferably, the machine learning classifier may include one or more of linear discriminant analysis, support vector machine, logistic regression, K-nearest neighbors, and decision trees; the feature selection strategy may include one or more of correlation-based feature selection, recursive feature elimination, and retaining all features.

[0056] Furthermore, this embodiment uses standard microspheres to verify the performance of the above system and method. Specifically, three types of standard microspheres were used for verification: polystyrene microspheres (Bangs Laboratories, USA) with average diameters of 3.87 μm (standard deviation 0.25 μm) and 4.19 μm (standard deviation 0.27 μm), and silica microspheres (Bangs Laboratories, USA) with an average diameter of 3.93 μm (size variation coefficient 5%~10%). The polystyrene microspheres have a refractive index of 1.591 at an excitation wavelength of 589 nm, and the silica microspheres have a refractive index of 1.44. The standard microspheres were suspended in pure water and diluted to a concentration of approximately 1500 microspheres / mL.

[0057] The light scattering image is captured by the second detector 14 after passing through the third microscope objective 12 and the second microscope tube lens 13. Experimental results are as follows: Figure 3 As shown, microspheres of different diameters exhibit significantly different scattering fringe distributions, indicating that the two-dimensional light scattering pattern is sensitive to size variations. The experimentally obtained scattering patterns are then compared with simulation results based on Mie theory. Figure 4 For comparison, in the simulation, the refractive index of polystyrene microspheres was set to 1.591, the refractive index of silica microspheres was set to 1.44, the refractive index of the surrounding medium was set to 1.334, the excitation wavelength was 532 nm, and the polar angle and azimuth angle ranged from 79° to 101°. Figure 5 As shown, the simulated pattern is in high agreement with the experimental results in terms of stripe distribution. The size difference between the two types of polystyrene microspheres, 3.87 μm and 4.19 μm, is 320 nm, proving that this method has submicron-level particle size resolution.

[0058] Meanwhile, a quantitative phase module based on a Michelson interferometer was used to obtain the phase image and average refractive index value of the microsphere. Figure 6 The original interference fringe patterns of the three microspheres are shown. Figure 7 The corresponding reconstructed phase diagrams yield the following refractive index values: 1.590±0.029 for 3.87μm microspheres, 1.593±0.035 for 4.19μm microspheres, and 1.440±0.015 for 3.93μm microspheres, which are highly consistent with the supplier's nominal values.

[0059] like Figure 8 As shown, the particle size distribution widths obtained by a single off-axis digital holography method (3.87μm±0.33μm, 4.19μm±0.38μm) are greater than the supplier's nominal values ​​(±0.25μm, ±0.27μm), indicating that single quantitative phase imaging has limitations in particle size resolution.

[0060] Systematic validation using standard microspheres demonstrates that light scattering imaging can resolve particle size differences of up to 320 nm, and the refractive index values ​​obtained by quantitative phase imaging are in high agreement with the nominal values. The dual-modal analysis strategy can compensate for the above deficiencies through light scattering imaging, while simultaneously achieving submicron-level particle size resolution and accurate refractive index measurement.

[0061] Furthermore, in this embodiment, the automatic classification capability of the above system and method is verified using primary leukemia cells derived from patients. The specific steps are as follows: First, cell samples were prepared. It should be noted that the clinical bone marrow samples were obtained from patients currently treated in the hospital's hematology department. The research protocol was reviewed and approved by the hospital's medical ethics committee (approval numbers KYLL-202206-004 and KYLL-202311-033). All participating patients signed informed consent forms before sample collection, and there were no privacy or confidentiality issues involved.

[0062] Granulocytes were isolated from bone marrow samples from patients with chronic myeloid leukemia (CML). Normal granulocytes were isolated from peripheral blood samples from 10 independent healthy donors, separated by density gradient centrifugation, with the diluted blood suspension layered on top of lymphocyte separation medium (TBD; Lts1077-1). Normal granulocytes and CML granulocytes were isolated from neutrophils using a human bone marrow neutrophil isolation kit (Solarbio, P2600). The neutrophils were fixed in 4% paraformaldehyde for 30 minutes, washed, and resuspended in PBS for later use.

[0063] Monocytes were isolated from bone marrow samples from patients with chronic lymphocytic leukemia using a human mononuclear cell separation kit (TBD; Lts1077-1). The isolated cells were resuspended in PBS, fixed with 4% paraformaldehyde for 30 minutes, and centrifuged to remove the fixative.

[0064] The final dataset consists of 100 chronic myeloid leukemia cells, 180 chronic lymphocytic leukemia cells, and 100 normal cells.

[0065] Next, images were acquired. The cell suspension to be tested was injected into the sample chamber and kept flowing steadily at a rate of 10 μL / h. The laser was turned on, and two-dimensional light scattering images and interferograms of individual cells were acquired at a frame rate of 60 frames / second. The images were preprocessed and uniformly scaled to 256×256 pixels. Figure 9 , Figure 10 , Figure 11The images are, respectively, interference maps of chronic lymphocytic leukemia cells, chronic myeloid leukemia cells, and normal cells; reconstructed quantitative phase maps; three-dimensional phase rendering maps; and two-dimensional light scattering maps.

[0066] Then, a classification experiment was set up. Features were extracted from two modalities: 81 features were extracted from the quantitative phase image: 14 geometric morphological features, 17 first-order statistical features, and 50 second-order texture features; 67 features were extracted from the light scattering image: 17 first-order statistical features and 50 second-order texture features. All features were concatenated into a 148-dimensional multimodal feature vector per cell.

[0067] The classification tasks include binary classification (chronic lymphocytic leukemia (CLL) cells versus chronic myeloid leukemia (CML) cells versus normal cells versus CML cells versus normal cells) and tri-class classification (simultaneous classification of the three cell types). Based on this, three feature sets were designed: quantitative phase imaging features (81 dimensions), light scattering imaging features (67 dimensions), and multimodal features (148 dimensions); three feature selection strategies were employed: correlation-based feature selection, recursive feature elimination, and retention of all features; five classifiers were used: linear discriminant analysis, support vector machine, logistic regression, K-nearest neighbors, and decision tree; and evaluation metrics included accuracy, precision, sensitivity, specificity, F1-score, and AUC.

[0068] Ten-fold cross-validation is employed with a fixed random seed (e.g., 14). The data is randomly divided into 10 subsets, with 9 subsets used as the training set and 1 subset as the test set in each iteration. All data transformation parameters and feature selection processes are computed only within the training subset of each iteration, ensuring that the test data is not visible during feature selection and model training, thus providing a reliable and unbiased evaluation.

[0069] Finally, the results of the three-class classification experiment were obtained. The optimal configuration for the three-class classification task is a combination of multimodal features, retaining all features, and a linear discriminant analysis classifier (Δ=1×10). -4 (Γ=0.5), the overall accuracy rate was 92.4%. Figure 12 The corresponding confusion matrix shows that the sensitivity is 93.33% for chronic lymphocytic leukemia cells, 90% for chronic myeloid leukemia cells, and 93% for normal cells, indicating minimal inter-class confusion. Detailed classification performance for each class is shown in Table 1.

[0070] Table 1 Comparison of Classification Performance by Category

[0071] Among them, multimodal features, all features retained, and a linear discriminant analysis classifier (Δ=1×10) are used. -4 Using the method of Γ=0.5, the overall accuracy rate was 92.4%.

[0072] Finally, the results of the binary classification experiment were obtained. The results show that the multimodal feature set achieved the best performance in all three paired classification tasks. The specific results are shown in Table 2.

[0073] Table 2 Performance Comparison of Different Classification Tasks

[0074] Based on the above results, modal comparison and feature analysis are performed: Different cell types exhibit varying sensitivities across different modalities. Light scattering imaging demonstrates superior performance in distinguishing between chronic lymphocytic leukemia (CLL) cells and normal cells, achieving an optimal classification accuracy approximately 2.1% higher than quantitative phase imaging. However, in distinguishing between chronic myeloid leukemia (CML) cells and normal cells, quantitative phase imaging significantly outperforms light scattering imaging, achieving an optimal classification accuracy approximately 12% higher. This complementary sensitivity stems from the different ultrastructural morphologies of the two leukemia cell types and the intrinsic contrast mechanism between the two optical modalities.

[0075] To elucidate the physical mechanism driving this complementary sensitivity, this embodiment employs the out-of-band permutation importance assessment of the random forest algorithm to evaluate the contribution of each feature to the classification decision, such as... Figure 13 , Figure 14 As shown, in distinguishing chronic lymphocytic leukemia from normal cells, the most significant feature is light scattering imaging feature number 4 (minimum). In distinguishing chronic myeloid leukemia from normal cells, quantitative phase imaging features rank highly, especially first-order statistical features such as feature number 17 (range) and feature number 19 (maximum).

[0076] like Figure 15 The feature correlation analysis results shown are illustrated, with red indicating positive correlation and blue indicating negative correlation. The quantitative phase imaging feature subset is marked with a yellow box, and the light scattering imaging feature subset is marked with a blue dashed box. Highly correlated features mainly cluster within the same modality, while cross-modal correlation is significantly lower, demonstrating that the two types of features are complementary rather than redundant. Figure 16 The violin plots show the distribution of four representative biophysical characteristics (refractive index, dry mass, surface area, and smoothness) among different cell types. The two ends of the violin represent the quartiles, and the horizontal line in the middle represents the median, showing that there are significant differences among the three cell types (*** indicates p<0.001).

[0077] Through the above verification, this embodiment verifies the synergistic effect of light scattering imaging and quantitative phase imaging as complementary modalities in label-free single-cell classification, reveals the differential sensitivity of different modalities under different cell types, and demonstrates the universality and robustness of multimodal fusion in the analysis of unknown clinical samples. For example, the fusion analysis of subwavelength morphological features and quantitative biophysical features shows the great potential of multimodal imaging technology in the differential diagnosis of hematological malignancies in cell pathology.

[0078] Example 3 This embodiment provides an electronic device, including: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory to implement the method provided in this embodiment.

[0079] Example 4 This embodiment also provides a computer-readable storage medium storing executable instructions, which, when executed by a processor, will cause the processor to execute the method described above in this embodiment.

[0080] The steps involved in Embodiments 2 to 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0081] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0082] The above description is only a preferred embodiment of the present invention. Although the specific implementation of the present invention has been described in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solution of the present invention, various modifications or variations that can be made by those skilled in the art without creative effort are still within the scope of protection of the present invention.

Claims

1. A label-free single-cell classification system based on light scattering and phase imaging, characterized in that, include: The laser emitting unit is used to generate a single coherent laser beam, which serves as the common excitation source for the system and is horizontally incident on the sample detection area. The sample delivery unit is used to deliver the cell suspension to the sample detection area at a constant low flow rate. The quantitative phase imaging acquisition unit is located behind the sample detection area along the horizontal main axis of the laser, forming a quantitative phase imaging optical path. It is used to record the interferogram and obtain a quantitative phase image of a single cell through numerical reconstruction. The light scattering imaging acquisition unit is located on the side of the sample detection area, forming a side scattering imaging optical path, which is used to collect two-dimensional scattered light from single cells in parallel within the side angle range and generate a scattering image. The data processing and analysis unit is used to extract multi-dimensional features from phase images and scattering images respectively. After feature fusion, the features are input into a machine learning classifier, and the output is the automatic single-cell typing result. Among them, the two imaging optical paths intersect at the sample detection area to form a three-dimensional orthogonal dual optical path.

2. The label-free single-cell classification system based on light scattering and phase imaging as described in claim 1, characterized in that, The laser emitting unit includes a laser, a beam expander, a pinhole filter, and a first microscope objective connected in sequence. The laser emits a single coherent laser beam, which is expanded by the beam expander and filtered by the pinhole filter in sequence. The first microscope objective focuses the laser beam onto the sample detection area. After passing through the single cells of the cell suspension in the sample detection area, the laser beam enters the two branch optical paths corresponding to the quantitative phase imaging acquisition unit and the light scattering imaging acquisition unit, respectively.

3. The label-free single-cell classification system based on light scattering and phase imaging as described in claim 1, characterized in that, The sample delivery unit includes a square microglass capillary sample chamber and a high-precision injection pump. The cell suspension concentration is diluted to 1500 particles / mL, and the flow rate of the cell suspension is controlled at 10 μL / h by the high-precision injection pump. The cell suspension is delivered to the sample detection area of ​​the sample chamber at a constant low flow rate. The sample chamber is also equipped with a three-axis displacement adjustment mechanism to ensure that the single cell is always in the focal plane during image acquisition.

4. The label-free single-cell classification system based on light scattering and phase imaging as described in claim 1, characterized in that, The quantitative phase imaging acquisition unit includes a second microscope objective, a beam splitter cube, a first reflecting mirror, a second reflecting mirror, a first lens barrel, and a first detector arranged sequentially along the main axis of laser horizontal transmission. The transmitted light passing through the single cell is magnified by the second microscope objective and enters a self-interference device based on a Michelson interferometer. The beam splitter cube in the device splits the transmitted light into two coherent beams from the same source. The two beams are emitted in the vertical and horizontal directions, respectively, and are reflected by the first reflecting mirror and the tilted second reflecting mirror, and then reflected back to the beam splitter cube to be re-recombined. After coherent superposition, they form interference light with a spatial carrier. The recombined interference light is converged by the first lens barrel and projected onto the first detector. The first detector records the holographic interferogram and transmits it to the computer for numerical reconstruction to obtain a quantitative phase image of the single cell. The second reflector is tilted at a small angle to generate a spatial carrier stripe pattern, causing the two beams of light to be spatially displaced on the camera plane, and the displacement is greater than the diameter of the cell sample.

5. The label-free single-cell classification system based on light scattering and phase imaging as described in claim 4, characterized in that, Reconstructing a quantitative phase map from an interferogram includes: The holographic interferogram is subjected to Fourier transform, and the required frequency components are retained by frequency domain filtering to obtain the preprocessed data; The complex field is reconstructed by inverse Fourier transform of the preprocessed data and the phase is calculated. Then, the 2π phase ambiguity is eliminated by a reliability-guided two-dimensional phase unwrapping algorithm for discontinuous paths to obtain the object light phase. Using the holographic interferogram acquired in the empty sample chamber as a reference, the reference phase is extracted, and the reference phase is subtracted from the object light phase to obtain the true phase difference distribution; Based on the true phase difference distribution, the average refractive index and cell dry mass of the cells were calculated respectively.

6. The label-free single-cell classification system based on light scattering and phase imaging as described in claim 1, characterized in that, The light scattering imaging acquisition unit includes a third microscope objective, a second microscope lens, and a second detector arranged sequentially from top to bottom. In the sample detection area, single cells of the cell suspension generate two-dimensional lateral scattered light from the laser beam. The third microscope objective on the side of the sample detection area collects the scattered light within a specific lateral angle range. The amplified scattered beam continues to be transmitted downward to the second microscope lens for beam collimation and convergence to correct imaging distortion. The converged scattered light is finally input vertically downward into the second detector to complete the acquisition of the two-dimensional lateral scattering image. In this process, the exposure times for scattering imaging and holographic imaging are set independently. The dual-modal image acquisition is triggered by the same timer from the same source and acquired in parallel at the same frame rate. The trigger time deviation is much smaller than the residence time of the cell in the sample detection area, so as to ensure that the dual-modal image corresponds to the same single cell.

7. The label-free single-cell classification system based on light scattering and phase imaging as described in claim 1, characterized in that, The data processing and analysis unit includes the following components arranged in sequence: The feature extraction module is configured with two independent feature extraction logics to extract multi-dimensional features from the scattering image and the phase image, respectively. The extracted features from the scattering image include first-order statistical features and second-order texture features. The second-order texture features include gray-level co-occurrence matrix features, gray-level gradient co-occurrence matrix features, and gray-level run-length matrix features. The extracted features from the phase image include geometric morphological features and multi-dimensional features that are homologous to the scattering image. The feature fusion module is used to concatenate the multi-dimensional features extracted from the dual-modal images to form a multi-dimensional multimodal fusion feature vector; The classification module is used to input the fused multi-dimensional and multi-modal fused feature vector into the machine learning classifier, and uses ten-fold cross-validation to output the automatic classification result of a single cell; The machine learning classifier includes at least one of linear discriminant analysis, support vector machine, logistic regression, K-nearest neighbors, and decision tree, and the feature selection strategy used in the machine learning training process includes at least one of relevance feature selection, recursive feature elimination, and full feature retention.

8. A label-free single-cell classification method based on light scattering and phase imaging, implemented using the label-free single-cell classification system based on light scattering and phase imaging as described in any one of claims 1-7, characterized in that, include: The cell suspension to be tested is delivered to the sample detection area of ​​the sample chamber at a constant ultra-low flow rate; The laser is activated to acquire two-dimensional side-scattering images and holographic interferograms of the same single cell in parallel at a set frame rate; Numerical reconstruction of the holographic interferogram yields a quantitative phase image of the cell; Multi-dimensional features are extracted from the scattering image and the phase image respectively. After feature splicing and fusion, the results are input into a machine learning classifier and the output is the automatic single-cell typing result.

9. An electronic device, characterized in that, include: Memory, used to store executable instructions; The processor, when executing executable instructions stored in the memory, implements the label-free single-cell classification method based on light scattering and phase imaging as described in claim 8.

10. A computer-readable storage medium, characterized in that, It stores executable instructions for causing the processor to execute the executable instructions to implement the label-free single-cell classification method based on light scattering and phase imaging as described in claim 8.

Images