Noninvasive bladder cancer detection method, system and equipment

Abstract

The invention discloses a non-invasive bladder cancer detection method, system and equipment. The method comprises the following steps: acquiring the intensity, phase, bright field and dark field images of cells in a urine sample to be detected; and inputting the obtained intensity, phase, bright field and dark field images of the cells into the trained bladder cancer cell detection model, and outputting a bladder cancer cell detection result. By constructing, training and optimizing a bladder cancer cell detection network model based on machine learning and combining with a multi-parameter flow cytometry imaging device, high-precision, noninvasive and unmarked bladder cancer detection analysis is realized, and the problems that during existing bladder cancer detection, trauma is large, identification indexes are not diversified enough, sensitivity is not enough in the diagnosis process, and the diagnosis accuracy is poor are solved. And the cancer cells cannot be accurately identified.

Description

Technical Field

[0001] The present invention relates to the field of bladder cancer detection, and in particular to a bladder cancer cell detection method based on an optofluidic time-domain stretching microscopy imaging system, quantitative phase imaging, optical polarization imaging and a machine learning algorithm. Background Art

[0002] At present, the treatment of bladder cancer is relatively simple, mainly using surgical resection of the lesion plus chemotherapy. Because bladder cancer grows in an invasive manner, patients still have a high risk of recurrence during the postoperative recovery phase. According to statistics, the postoperative recurrence rate of bladder cancer is 50%~70%. If bladder cancer is not discovered in time, the abundant blood vessels and lymphatic vessels in the bladder will provide cancer cells with a very convenient channel for further spread, thereby aggravating the metastasis of cancer cells, which will pose a great threat to the patient's life. For the early metastasis of bladder cancer, this metastasis often occurs at the cellular level, so timely single-cell diagnosis of cancer is very important for the treatment of the disease.

[0003] At present, the commonly used detection method in clinical practice generally adopts the method of cystoscopy. This method obtains tissue images of the lesion location by puncture, and can obtain accurate information on cancer. This method also has many disadvantages. For example, the microscopic examination process will cause damage to the urethra and bladder mucosa, causing bleeding, and the risk of urinary tract infection after the examination is very high; cystoscopy will cause pain and psychological discomfort to the examinee, and the patient's compliance is not high enough; from the doctor's point of view, this method will also affect the accuracy of the results due to its own technical proficiency, etc. These shortcomings are not conducive to the detection of bladder cancer. Considering that the bladder is an organ for storing urine, the shed bladder epithelial cells will be excreted from the body with urine. Therefore, it is possible to achieve non-invasive detection of bladder cancer by detecting cells or metabolites in urine, avoiding secondary damage caused by puncture.

[0004] Urine is mainly composed of water, glucose, urea, uric acid, inorganic salts, proteins and DNA. The proportion of cells in urine is relatively small and there are many types, which increases the difficulty of urine detection. In order to clearly distinguish cancerous bladder cells from other substances, Fish fluorescence is used as a specific marker to mark individual cells in urine. However, this method is cumbersome, expensive and depends on the doctor's own proficiency, and cannot achieve high-speed, low-cost and accurate cancer cell detection. In addition to the above-mentioned methods of detecting cells and tissues, human metabolites can also be used as detection objects, such as miRNA, hyaluronic acid, etc.: Although miRNA can be used as a specific detection object, its detection cost is relatively high, and the miRNA fragments specifically expressed in bladder cancer cannot be determined, and further exploration is still needed, which makes the current accuracy of this detection method not high; and hyaluronic acid-like proteins, as metabolites of the human body, are expressed at high levels in many cases, so the use of this detection method will result in a high false positive rate. In recent years, people have tried to combine machine learning with conventional imaging methods to improve the accuracy of bladder cancer detection: for example, in 2018, I. Sokolov's research team at Tufts University in the United States extracted cell features through atomic force microscopy, and then used machine learning to distinguish bladder cancer cells from normal cells. This study cleverly combined atomic force microscopy with machine learning for bladder cancer detection in urine. However, this method still finds it difficult to achieve high-speed, high-throughput cell detection and is difficult to meet the needs of clinical applications. How to achieve non-invasive, efficient, fast, and accurate bladder cancer detection has become an urgent problem to be solved.

[0005] The disclosed invention patent (application number: CN202211169459.5) describes a bladder cancer marker detection kit comprising Fe3O4@Au@DNA1 and UCNPs@DNA2, which are coupled with the bladder cancer marker miRNA-96 in urine, and the content of the bladder cancer marker is determined based on the fluorescence after the reaction, thereby detecting the degree of bladder cancer. The method is non-invasive, low-cost, simple, rapid, highly sensitive, and highly specific, and can be used for bladder cancer screening. The disclosed invention patent (application number: Fe3O4@Au@DNA1) provides a bladder cancer detection kit, which includes a detection reagent for detecting circZFR. The invention determines the potential application value of circular RNAcZFR as a clinical diagnosis and treatment target for bladder cancer recurrence, and provides a new theoretical basis for preventing bladder cancer recurrence. The disclosed invention patent (application number: CN202210391403.8) discloses a DNA methylation multiplex PCR kit for bladder cancer detection and its application, the kit includes: primer pairs for DNA methylation markers for bladder cancer detection: primer pairs 1-7. It is mainly used for auxiliary diagnosis of bladder cancer patients, diagnosis and type identification before surgery, and provides an effective basis for subsequent treatment of patients.

[0006] The above patent uses staining to detect bladder cancer cells, but there are still problems such as insufficient diversity of identification indicators and insufficient sensitivity during the diagnosis process, which leads to the inability to accurately identify cancer cells. Therefore, it is urgent to establish a high-speed, high-throughput, and high-precision bladder cancer cell detection and intelligent analysis platform. Summary of the invention

[0007] In order to solve the problems of large trauma in existing bladder cancer detection, insufficient diversity of identification indicators and insufficient sensitivity during the diagnosis process, resulting in the inability to accurately identify cancer cells, the present invention provides a non-invasive bladder cancer detection method, system and equipment. By constructing, training and optimizing a bladder cancer cell detection network model based on machine learning, combined with a multi-parameter flow cytometer, high-precision, non-invasive and label-free bladder cancer detection and analysis can be achieved.

[0008] According to one aspect of the present invention, a non-invasive bladder cancer detection method is provided, comprising: Acquire intensity, phase, bright field and dark field images of cells in the urine sample to be tested; Input the intensity, phase, bright field and dark field images of the acquired cells into the trained bladder cancer cell detection model, and output the bladder cancer cell detection results; wherein the training of the bladder cancer cell detection model includes: Construct a bladder cancer cell training set; A machine learning-based bladder cancer cell detection model was constructed with MobileNeV3 as the backbone network, including: a feature extraction module for extracting multi-dimensional features of bladder cancer cell intensity, phase, bright field and dark field images; an attention module for adding spatial channel attention so that the model can quickly locate the bladder cancer cell area and encode and analyze the features of bladder cancer cells; a classification module for predicting the probability distribution of bladder cancer cell images; The constructed bladder cancer cell training set is used to train the bladder cancer cell detection model based on machine learning to obtain a trained bladder cancer cell detection model.

[0009] As a further technical solution, intensity, phase, bright field and dark field images of cells in the urine sample to be tested are obtained, including: Obtaining a urine sample to be tested and pre-treating it to prepare a cell sample suitable for detection by a multi-parameter flow cytometric imaging device; The intensity, phase, bright field and dark field images of cells were acquired using a multi-parameter flow cytometer.

[0010] As a further technical solution, a multi-parameter flow cytometric imaging device is used to obtain intensity, phase, bright field and dark field images of cells, including: A quantitative phase imaging component is integrated on the basis of the optofluidic time-domain stretch imaging component to obtain intensity and phase images of cells in urine; An optical polarization imaging component is integrated on the basis of the optofluidic time-domain stretching imaging component to obtain bright-field and dark-field images of cells in urine.

[0011] As a further technical solution, a bladder cancer cell training set was constructed, including: Collect urine samples and obtain cell images in urine after detection by a multi-parameter flow cytometer, including intensity, phase, bright field and dark field images of cells; Based on the acquired cell images, the image clustering method was used to perform preliminary bladder cancer cell classification, and the preliminary classification results were manually screened to establish a bladder cancer cell image database to form a bladder cancer cell training set.

[0012] As a further technical solution, building a bladder cancer cell detection model based on machine learning also includes: introducing an activation function module to optimize the nonlinearity of the model in the deep layer and reduce the number of parameters.

[0013] According to one aspect of the present specification, a non-invasive bladder cancer detection system is provided, comprising an optofluidic time-domain stretch imaging part, a cell intensity and phase image acquisition part, a cell bright field and dark field image acquisition part and a processing part, wherein the optofluidic time-domain stretch imaging part is used to acquire the mapping of cell spatial information to the pulse time-domain waveform, so as to process and obtain a complete two-dimensional image of the flowing cells; the cell intensity and phase image acquisition part is used to construct a quantitative phase imaging system based on the interference principle, so as to collect time-domain interference signals to restore the intensity and phase images of the cells; the cell bright field and dark field image acquisition part is used to construct an optical polarization imaging system, so as to obtain the bright field and dark field images of the cells respectively; the processing part is used to construct and train a bladder cancer cell detection model, and use the trained bladder cancer cell detection model to process the acquired intensity, phase, bright field and dark field images of the cells, so as to obtain the bladder cancer cell detection results.

[0014] As a further technical solution, the optofluidic time-domain stretching imaging part includes: a laser for generating femtosecond pulses; a dispersive optical fiber for separating the different wavelength components of ultrashort pulses in the time domain to achieve frequency-time mapping; a diffraction grating for projecting the different wavelength components of ultrashort pulses to different positions in space to achieve frequency-space mapping; a microscope objective for focusing the spatially dispersed light pulses on the cells in the logistics channel; a microfluidic chip for making the cells flow through the channel stably and quickly; a photodetector for signal acquisition; and a digitizer for obtaining a complete two-dimensional image of the flowing cells based on the acquired signals.

[0015] As a further technical solution, the cell intensity and phase image acquisition part includes: a beam splitter, which is used to divide the light pulse into detection light and reference light to construct a quantitative phase imaging system based on the interference principle; a delay device, which is used to adjust the optical path of the reference light so that the detection light and the reference light reach the photodetector at the same time to achieve spatial interference.

[0016] As a further technical solution, the cell bright field and dark field image acquisition part includes a linear polarizer, a beam splitter, a 1 / 2 wave plate and a polarization beam splitter. The linear polarizer converts the light pulse passing through the cell into linear polarized light. The linear polarized light passing through the cell reaches different photodetectors after passing through the 1 / 2 wave plate and the polarization beam splitter to construct an optical polarization imaging system. By adjusting the 1 / 2 wave plate and the polarization beam splitter, the bright field and dark field images of the cell can be obtained respectively.

[0017] According to one aspect of the present invention, a non-invasive bladder cancer detection device is provided, comprising: An image acquisition module, used to acquire intensity, phase, bright field and dark field images of cells in the urine sample to be tested; The image detection module is used to input the intensity, phase, bright field and dark field images of the acquired cells into the trained bladder cancer cell detection model and output the bladder cancer cell detection results; wherein the training of the bladder cancer cell detection model includes: Construct a bladder cancer cell training set; A machine learning-based bladder cancer cell detection model was constructed with MobileNeV3 as the backbone network, including: a feature extraction module for extracting multi-dimensional features of bladder cancer cell intensity, phase, bright field and dark field images; an attention module for adding spatial channel attention so that the model can quickly locate the bladder cancer cell area and encode and analyze the features of bladder cancer cells; a classification module for predicting the probability distribution of bladder cancer cell images; The constructed bladder cancer cell training set is used to train the bladder cancer cell detection model based on machine learning to obtain a trained bladder cancer cell detection model.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention develops a high-throughput, ultrafast flow imaging system based on the optofluidic time-domain stretch imaging system, and provides an imaging technology that is suitable for cells in urine. First, the present invention introduces quantitative phase detection technology and optical polarization detection technology into the optofluidic time-domain stretch imaging system so that the imaging system can simultaneously measure the intensity, phase, bright field, and dark field information of the light pulse; the light pulse signal is used to invert information of different dimensions such as the size, structure, dry mass, and roughness of the cell to provide complete cell parameters. Then, the present invention extracts and analyzes the intensity, phase, bright field, and dark field images of the collected cells, establishes and trains a high-precision bladder cancer detection network model, and realizes the detection of bladder cancer cells.

[0019] 2. The present invention can obtain a large amount of bladder cancer cell image data through a multi-parameter flow cytometric imaging system, thereby achieving a comprehensive evaluation of bladder cancer cells in urine; in addition, due to the label-free detection method of the method of the present invention, the operation steps are simplified and the detection efficiency is improved.

[0020] 3. The present invention analyzes the bladder cancer cell image database through the established machine learning-based detection model, extracts and classifies the bladder cancer cell image features, and can achieve rapid and accurate bladder cancer cell identification and derive the proportion of bladder cancer cells. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, a brief introduction is given below to the drawings used in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 A schematic flow chart of a non-invasive bladder cancer detection method provided in an embodiment of the present invention.

[0023] Figure 2 A schematic diagram of a bladder cancer cell detection model provided in an embodiment of the present invention.

[0024] Figure 3 A schematic diagram of the principle of non-invasive bladder cancer detection provided by an embodiment of the present invention.

[0025] Figure 4 A schematic diagram of the structure of a non-invasive bladder cancer detection system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In addition, the technical features in the various embodiments or single embodiments provided by the present invention are arbitrarily combined with each other to form a new technical solution. This combination is not restricted by the sequence of steps and / or the structural composition mode, but must be based on the ability of ordinary technicians in this field to achieve. When the combination of technical solutions is contradictory or cannot be achieved, it should be considered that this combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0027] An embodiment of the present invention provides a non-invasive bladder cancer detection method, which realizes bladder cancer detection based on a multi-parameter flow cytometer and machine learning. First, a bladder cancer cell detection model based on machine learning is constructed and trained, and then the trained model is used to detect images collected by the multi-parameter flow cytometer to predict the probability distribution of bladder cancer cell images, thereby realizing high-accuracy and high-efficiency bladder cancer cell detection.

[0028] See also Figure 1The non-invasive bladder cancer detection method provided by the embodiment of the present invention includes: obtaining the intensity, phase, bright field and dark field images of cells in the urine sample to be tested; inputting the obtained intensity, phase, bright field and dark field images of the cells into the trained bladder cancer cell detection model, and outputting the bladder cancer cell detection results.

[0029] The step of obtaining the intensity, phase, bright field and dark field images of cells in the urine sample to be tested is obtained by a multi-parameter flow cytometric imaging device, and specifically includes: collecting a urine sample from a patient, washing the urine sample with PBS, and preparing a cell sample suitable for detection by a multi-parameter flow cytometric imaging system; passing the washed urine and PBS solution into the inlet of a microfluidic chip, so that the cells can flow at a high speed and stably in the detection channel of the microfluidic chip; detecting bladder cancer cells in the urine sample by the multi-parameter flow cytometric imaging device to obtain the intensity, phase, bright field and dark field images of the bladder cancer cells, and transmitting the collected images to a computing device.

[0030] Furthermore, the computing device sends the obtained cell intensity, phase, bright field and dark field images to a trained bladder cancer cell detection model based on machine learning to detect bladder cancer cells and predict the probability distribution of bladder cancer cell images.

[0031] In an embodiment of the present invention, the implementation process of the multi-parameter flow cytometer is as follows: on the basis of the optofluidic time-domain stretching imaging system, quantitative phase imaging technology is integrated to obtain intensity and phase images of cells in urine; on this basis, optical polarization imaging technology is integrated to obtain bright field and dark field images of cells in urine.

[0032] Specifically, the implementation process of the optofluidic time-domain stretch imaging system is as follows: using femtosecond pulses as the probe light, first use a section of dispersive optical fiber to separate the different wavelength components of the ultrashort pulse in the time domain to achieve frequency-time mapping, then use a diffraction grating to project the different wavelength components of the ultrashort pulse spectrum to different positions in space to achieve frequency-space mapping, and then use a microscope objective to focus the spatially dispersed light pulses on the cells in the logistics channel, thereby achieving the mapping of cell spatial information to the pulse time domain waveform. Finally, a single-pixel photodetector is used to collect signals. As the cells flow in the microfluidic channel, each light pulse detects the information of a cross section of the cell, and the pulses are spliced ​​together at the back end through digital signal processing to obtain a complete two-dimensional image of the flowing cell.

[0033] The process of obtaining the intensity and phase images of cells in urine is as follows: a beam splitter is introduced after the dispersion optical fiber to divide the light pulse into a detection light and a reference light to construct a quantitative phase imaging system based on the interference principle, the optical path of the reference light is adjusted by a delay device so that the detection light and the reference light reach the photodetector at the same time to achieve spatial interference, and after digital signal processing, the intensity and phase images of the cells are restored from the collected time domain interference signal.

[0034] The process of obtaining bright field and dark field images of cells in urine is as follows: a linear polarizer is added in front of a diffraction grating so that the light pulse passing through the cells is linearly polarized light, and then the linear polarized light passing through the cells passes through a 1 / 2 wave plate and a polarization beam splitter and reaches different photodetectors respectively, so as to construct an optical polarization imaging system. By adjusting the 1 / 2 wave plate and the polarization beam splitter, the bright field and dark field images of the cells can be obtained respectively.

[0035] In an embodiment of the present invention, the training of the bladder cancer cell detection model includes: Construct a bladder cancer cell training set; A machine learning-based bladder cancer cell detection model was constructed with MobileNeV3 as the backbone network, including: a feature extraction module for extracting multi-dimensional features of bladder cancer cell intensity, phase, bright field and dark field images; an attention module for adding spatial channel attention so that the model can quickly locate the bladder cancer cell area and encode and analyze the features of bladder cancer cells; a classification module for predicting the probability distribution of bladder cancer cell images; The constructed bladder cancer cell training set is used to train the bladder cancer cell detection model based on machine learning to obtain a trained bladder cancer cell detection model.

[0036] Specifically, the steps of constructing a bladder cancer cell training set include: collecting urine samples, obtaining cell images in the urine after detection through a multi-parameter flow cytometer, including cell intensity, phase, bright field and dark field images; using an image clustering method to preliminarily classify the acquired bladder cancer cells, and manually screening to select and correctly classify bladder cancer cell images that are misclassified in image clustering to establish a bladder cancer cell image database; constructing a training set, a test set and a validation set based on the established bladder cancer cell image database, which are used for model training, testing and validation, respectively.

[0037] See also Figure 2In order to enable the machine learning module to fully grasp the characteristics of bladder cancer cell images and analyze bladder cancer cell images with high precision, MobileNeV3 (M3) is used as the backbone network for cell classification to design a lightweight and efficient cell classification network. While ensuring the accuracy of the model, the model parameters are compressed and the network inference speed is improved. M3 consists of four modules, and introduces deep separable convolution, attention module, and H-swish activation function to improve the model performance. For deep separable convolution, it is an important application to reduce the multiplication and addition operators by releasing parameters. It replaces the traditional convolution layer with deep convolution and point convolution. Deep convolution calculates convolution by convolving a single tensor channel to extract features. For the attention module, by introducing the weights of each channel parameter, the model can pay more attention to the key feature information of urine particle images of different forms and categories, be more sensitive to the specificity of particle images, enhance the interpretability of the model, and achieve faster and better bladder cancer cell image classification results. For the H-swish activation function, because its calculation is simpler than the swish and sigmoid functions, it is introduced into the lightweight neural network. At the same time, the nonlinearity of the model in the deep layer can be improved and the number of parameters can be reduced. By introducing the Softmax function as a classifier after the last fully connected layer, the probability distribution of bladder cancer cell images can be predicted.

[0038] See also Figure 3 , when training a machine learning-based bladder cancer cell detection model, including: S1, collecting urine samples from patients, and preparing cell samples suitable for detection by a multi-parameter flow cytometer after pretreatment, detecting bladder cancer cells in the urine samples by the multi-parameter flow cytometer to obtain intensity, phase, bright field and dark field images of bladder cancer cells; S2, based on the intensity, phase, bright field and dark field images of a large number of bladder cancer cells, feature clustering is performed using an image clustering method to establish a bladder cancer cell image database, and a training set, a test set and a validation set are constructed based on the established bladder cancer cell image database, which are used for model training, testing and validation respectively; S3, using the training set to train the constructed bladder cancer cell detection model, using the test set to test the trained bladder cancer cell detection model, and strengthening and optimizing the model based on the test results and accuracy requirements; S4, the bladder cancer cell detection model was validated and analyzed using the validation set.

[0039] The embodiment of the present invention obtains multi-dimensional bladder cancer morphology information through a multi-parameter flow cytometric imaging device, and uses this information as a training set for a machine learning algorithm to input into a high-accuracy, high-robustness, and lightweight classification model, thereby obtaining a detection model that can achieve high-accuracy detection. When actually detecting the morphology of bladder cancer cells in patients, the intensity, phase, bright field, and dark field morphology images of bladder cancer cells are collected without labeling at a flux of 1,000,000 cells per second, and the obtained multi-dimensional image data is input into the trained classification model to achieve rapid and high-accuracy classification of the bladder cancer morphology of patients and the corresponding proportions. Finally, the results of the system are optimized through the detection results of the clinical smear gold standard, thereby obtaining a high-accuracy, high-speed bladder cancer cell detection device.

[0040] An embodiment of the present invention also provides a non-invasive bladder cancer detection system, including an optofluidic time-domain stretch imaging part, a cell intensity and phase image acquisition part, a cell bright field and dark field image acquisition part and a processing part, wherein the optofluidic time-domain stretch imaging part is used to acquire a mapping of cell spatial information to a pulse time-domain waveform, so as to process and obtain a complete two-dimensional image of the flowing cells; the cell intensity and phase image acquisition part is used to construct a quantitative phase imaging system based on the interference principle, so as to collect time-domain interference signals to restore the intensity and phase images of the cells; the cell bright field and dark field image acquisition part is used to construct an optical polarization imaging system, so as to obtain the bright field and dark field images of the cells respectively; the processing part is used to construct and train a bladder cancer cell detection model, and use the trained bladder cancer cell detection model to process the acquired intensity, phase, bright field and dark field images of the cells to obtain bladder cancer cell detection results.

[0041] See also Figure 4 The photofluidic time-domain stretch imaging part includes a broadband femtosecond laser, a single-mode optical fiber, a diffraction grating, a microscope objective, a microfluidic chip, a photodetector, a digitizer (oscilloscope 119) and a computer 120; the cell intensity and phase image acquisition part includes a beam splitter and a delay device 113; the cell bright field and dark field image acquisition part includes a linear polarizer, a beam splitter, a 1 / 2 wave plate and a polarization beam splitter.

[0042] The broadband femtosecond laser 101 is a Vitara-P laser from Coherent, USA, with a central wavelength / repetition rate / spectral width / pulse width of 800nm / 80MHz / 40nm / 20fs. The single-mode optical fiber 102 with a length of 1.5 kilometers is a single-mode optical fiber of model YOFC-780-1.5 from Changfei Company. The beam splitter 103 with a beam splitting ratio of 90:10 is a beam splitter BS041 from Thorlabs. The reflective diffraction gratings 105 and 109 with a line density of 1200 lines / mm are GR26-0608 from Thorlabs. The microscope objective lenses 106 and 108 with a magnification of 50X and a numerical aperture of 0.65 are LCPLN-IR 50X from Olympus. The beam splitter 110 with a beam splitting ratio of 50:50 is BS005 from Thorlabs. The photodetectors 116, 117 and 118 are 1544-B-50 from Newport, with a bandwidth of 12.5 GHz. The high-speed oscilloscope 119 with a sampling rate of 40 GSa / s is DSA91304A from Keysight Technologies, USA. The parameters of the microfluidic chip 107 are a channel width of 80 µm and a channel height of 40 µm. The polarization beam splitter 115 is PBS-612 from Lubang. The linear polarizer 104 is FLP25-NIR-M from Lubang. The 1 / 2 wave plate 111 is AHWP20-SNIR from Lubang. The beam splitter 114 has a beam splitting ratio of 25:75.

[0043] The multi-parameter flow cytometric imaging system is used to obtain the intensity, phase, bright field and dark field images of bladder cancer cells at high speed and without labeling; the microfluidic chip is used to make the cells flow through the channel stably and quickly, so that the multi-parameter flow cytometric imaging device can stably image; the computer uses machine learning methods to adaptively analyze the massive cell data and multi-dimensional information obtained based on high-throughput, multi-parameter flow cytometric detection, and realizes the detection of bladder cancer cells by analyzing the intensity, phase, bright field and dark field images.

[0044] The optical fluidic time-domain stretch imaging part uses femtosecond pulses as the detection light. First, a section of dispersive optical fiber is used to separate the different wavelength components of the ultrashort pulse in the time domain to achieve frequency-time mapping. Then, a diffraction grating is used to project the different wavelength components of the ultrashort pulse spectrum to different positions in space to achieve frequency-space mapping. Subsequently, a microscope objective is used to focus the spatially dispersed light pulses on the cells in the logistics channel, thereby achieving the mapping of the cell spatial information to the pulse time domain waveform. Finally, a single-pixel photodetector is used to collect signals. As the cells flow in the microfluidic channel, each light pulse detects the information of a cross section of the cell, and the pulses are spliced ​​together at the back end through digital signal processing to obtain a complete two-dimensional image of the flowing cell, which can achieve a detection flux of more than 1,000,000 cells / second and a spatial resolution better than 780nm.

[0045] The cell intensity and phase image acquisition part introduces a beam splitter after the dispersion optical fiber to divide the light pulse into detection light and reference light to construct a quantitative phase imaging system based on the interference principle. The optical path of the reference light is adjusted by a delay device so that the detection light and the reference light reach the photodetector at the same time to achieve spatial interference. After digital signal processing, the intensity and phase images of the cells are restored from the collected time domain interference signals.

[0046] The cell bright field and dark field image acquisition part (i.e., the polarization imaging part) is configured to construct an optical polarization imaging system by adding a linear polarizer in front of the diffraction grating so that the light pulse passing through the cell is linearly polarized light. The linear polarized light passing through the cell then passes through a 1 / 2 wave plate and a polarization beam splitter and reaches different photodetectors respectively. The bright field and dark field images of the cell can be obtained respectively by adjusting the 1 / 2 wave plate and the polarization beam splitter.

[0047] The specific steps of using the non-invasive bladder cancer detection system of the embodiment of the present invention to detect bladder cancer cells include: S10: collecting urine samples from patients, washing the urine samples with PBS, and preparing cell samples suitable for detection by a multi-parameter flow cytometric imaging system; S20: passing the cleaned urine and PBS solution into the inlet of the microfluidic chip, so that the cells can flow in the detection channel of the microfluidic chip at a high speed and stably; S30: using a multi-parameter flow cytometric imaging system including an optofluidic time-domain stretch imaging part, a cell intensity and phase image acquisition part, and a cell bright field and dark field image acquisition part to collect cells flowing through the detection channel, and to obtain intensity, phase, bright field, and dark field images, and to transmit the collected images to a computer; S40: The computer sends the obtained cell intensity, phase, bright field and dark field images to the trained bladder cancer cell detection network model based on machine learning to detect bladder cancer cells and predict the probability distribution of bladder cancer cell images.

[0048] Based on the same inventive concept as the aforementioned embodiment, the embodiment of the present invention further provides a non-invasive bladder cancer detection device, comprising: an image acquisition module, used to acquire the intensity, phase, bright field and dark field images of cells in a urine sample to be tested; an image detection module, used to input the acquired intensity, phase, bright field and dark field images of cells into a trained bladder cancer cell detection model, and output bladder cancer cell detection results; wherein the training of the bladder cancer cell detection model comprises: constructing a bladder cancer cell training set; constructing a bladder cancer cell detection model based on machine learning with MobileNeV3 as the backbone network, comprising: a feature extraction module, used to extract multi-dimensional features of bladder cancer cell intensity, phase, bright field and dark field images; an attention module, used to add spatial channel attention, so that the model can quickly locate the bladder cancer cell area and encode and feature analyze the bladder cancer cell features; a classification module, used to predict the probability distribution of bladder cancer cell images; and using the constructed bladder cancer cell training set to train the bladder cancer cell detection model based on machine learning to obtain a trained bladder cancer cell detection model.

[0049] The non-invasive bladder cancer detection device provided by the embodiment of the present invention adopts the above-mentioned several modules, based on the multi-dimensional bladder cancer morphology information obtained by the multi-parameter flow cytometer, and utilizes the bladder cancer cell detection model based on machine learning to achieve high-accuracy detection of bladder cancer cells.

[0050] In summary, the present invention discloses a non-invasive bladder cancer detection method, which is based on a multi-parameter flow cytometric imaging system to obtain images of cells in urine, establish a bladder cancer cell image database, and construct, train and optimize a bladder cancer cell detection network model based on machine learning. On this basis, the trained bladder cancer cell detection network model is used to analyze the cell images obtained by the multi-parameter flow cytometric imaging system to detect urine samples, analyze the bladder cancer cells in the urine samples, achieve high-precision, non-invasive, and label-free bladder cancer detection and analysis, and dig out the morphological changes of bladder epithelial cells in urine after canceration.

[0051] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.

Claims

1. A non-invasive bladder cancer detection method, characterized in that: include: Acquire intensity, phase, bright field and dark field images of cells in urine samples to be tested; Input the intensity, phase, bright field and dark field images of the acquired cells into the trained bladder cancer cell detection model, and output the bladder cancer cell detection results; wherein the training of the bladder cancer cell detection model includes: Construct a bladder cancer cell training set; A machine learning-based bladder cancer cell detection model was constructed with MobileNeV3 as the backbone network, including: a feature extraction module for extracting multi-dimensional features of bladder cancer cell intensity, phase, bright field and dark field images; an attention module for adding spatial channel attention so that the model can quickly locate the bladder cancer cell area and encode and analyze the features of bladder cancer cells; a classification module for predicting the probability distribution of bladder cancer cell images; The constructed bladder cancer cell training set is used to train the bladder cancer cell detection model based on machine learning to obtain a trained bladder cancer cell detection model.

2. A non-invasive bladder cancer detection method according to claim 1, characterized in that: Acquire intensity, phase, brightfield, and darkfield images of cells in urine samples under test, including: Obtaining a urine sample to be tested and pre-treating it to prepare a cell sample suitable for detection by a multi-parameter flow cytometric imaging device; The intensity, phase, bright field and dark field images of cells were acquired using a multi-parameter flow cytometric imaging device.

3. A non-invasive bladder cancer detection method according to claim 2, characterized in that: Acquire intensity, phase, brightfield, and darkfield images of cells using a multi-parameter flow cytometer, including: A quantitative phase imaging component is integrated on the basis of the optofluidic time-domain stretch imaging component to obtain intensity and phase images of cells in urine; An optical polarization imaging component is integrated on the basis of the optofluidic time-domain stretching imaging component to obtain bright-field and dark-field images of cells in urine.

4. A non-invasive bladder cancer detection method according to claim 1, characterized in that: Construct a bladder cancer cell training set, including: Collect urine samples and obtain cell images in urine after detection by a multi-parameter flow cytometer, including intensity, phase, bright field and dark field images of cells; Based on the acquired cell images, the image clustering method was used to perform preliminary bladder cancer cell classification, and the preliminary classification results were manually screened to establish a bladder cancer cell image database to form a bladder cancer cell training set.

5. A non-invasive bladder cancer detection method according to claim 1, characterized in that: Building a bladder cancer cell detection model based on machine learning also includes: introducing an activation function module to optimize the nonlinearity of the model in the deep layer and reduce the number of parameters.

6. A non-invasive bladder cancer detection system, characterized in that: The invention comprises an optofluidic time-domain stretch imaging part, a cell intensity and phase image acquisition part, a cell bright field and dark field image acquisition part and a processing part. The optofluidic time-domain stretch imaging part is used to acquire the mapping of cell spatial information to the pulse time-domain waveform, so as to process and obtain a complete two-dimensional image of the flowing cell; the cell intensity and phase image acquisition part is used to construct a quantitative phase imaging system based on the interference principle, so as to collect the time-domain interference signal and restore the intensity and phase image of the cell; the cell bright field and dark field image acquisition part is used to construct an optical polarization imaging system, so as to obtain the bright field and dark field images of the cell respectively; the processing part is used to construct and train a bladder cancer cell detection model, and use the trained bladder cancer cell detection model to process the acquired intensity, phase, bright field and dark field images of the cell, so as to obtain the bladder cancer cell detection result.

7. A non-invasive bladder cancer detection system according to claim 6, characterized in that: The photofluidic time-domain stretch imaging part includes: a laser for generating femtosecond pulses; a dispersive optical fiber for separating different wavelength components of ultrashort pulses in the time domain to achieve frequency-time mapping; a diffraction grating for projecting different wavelength components of ultrashort pulses to different positions in space to achieve frequency-space mapping; a microscope objective for focusing spatially dispersed light pulses on cells in a logistics channel; a microfluidic chip for making cells flow through the channel stably and quickly; a photodetector for signal collection; and a digitizer for obtaining a complete two-dimensional image of the flowing cells based on the collected signals.

8. A non-invasive bladder cancer detection system according to claim 6, characterized in that: The cell intensity and phase image acquisition part includes: a beam splitter, which is used to split the light pulse into detection light and reference light to construct a quantitative phase imaging system based on the interference principle; a delay device, which is used to adjust the optical path of the reference light so that the detection light and the reference light reach the photodetector at the same time to achieve spatial interference.

9. A non-invasive bladder cancer detection system according to claim 6, characterized in that: The cell bright field and dark field image acquisition part includes a linear polarizer, a beam splitter, a 1 / 2 wave plate and a polarization beam splitter. The linear polarizer converts the light pulse passing through the cell into linear polarized light. The linear polarized light passing through the cell reaches different photodetectors after passing through the 1 / 2 wave plate and the polarization beam splitter to construct an optical polarization imaging system. The bright field and dark field images of the cell can be obtained respectively by adjusting the 1 / 2 wave plate and the polarization beam splitter.

10. A non-invasive bladder cancer detection device, characterized in that: include: An image acquisition module, used to acquire intensity, phase, bright field and dark field images of cells in the urine sample to be tested; The image detection module is used to input the intensity, phase, bright field and dark field images of the acquired cells into the trained bladder cancer cell detection model and output the bladder cancer cell detection results; wherein the training of the bladder cancer cell detection model includes: Construct a bladder cancer cell training set; A machine learning-based bladder cancer cell detection model was constructed with MobileNeV3 as the backbone network, including: a feature extraction module for extracting multi-dimensional features of bladder cancer cell intensity, phase, bright field and dark field images; an attention module for adding spatial channel attention so that the model can quickly locate the bladder cancer cell area and encode and analyze the features of bladder cancer cells; a classification module for predicting the probability distribution of bladder cancer cell images; The constructed bladder cancer cell training set is used to train the bladder cancer cell detection model based on machine learning to obtain a trained bladder cancer cell detection model.

Images