A method for establishing a cancer diagnosis model based on optical signals and its application

Abstract

The present invention provides a method and application for establishing a cancer diagnosis model based on optical signals, comprising the following steps: using surface functionalized polystyrene microspheres to separate extracellular vesicles from healthy people and patient plasma respectively; using an optical microcavity detection device to detect extracellular vesicles from healthy people and patients respectively, obtaining spectral data, performing variable standardization algorithm normalization processing on the spectral data, obtaining healthy people and patient training sets; establishing a CNN model, training the CNN model with the healthy people and patient training sets, and establishing a diagnosis model. The method for establishing a cancer diagnosis model based on optical signals described in the present invention uses polystyrene microspheres to achieve rapid separation of extracellular vesicles in plasma, combines spectral signal changes to achieve feature capture of extracellular vesicles in plasma, and empowers the convolutional neural network algorithm to achieve disease diagnosis, with the advantages of low cost, high efficiency, and ultra-sensitive detection.

Description

Technical Field

[0001] The present invention belongs to the field of biological detection technology, and in particular relates to a method and application of establishing a cancer diagnosis model based on optical signals. Background Art

[0002] Liquid biopsy is a companion technology in the field of tumor detection. Extracellular vesicles are characterized by their ability to appear in the early stages of tumors and carry tumor-related markers. They are circulating markers with great clinical diagnostic value. Extracellular vesicles (EVs) are a general term for particles with a phospholipid bilayer structure and nano to micron sizes that are actively secreted by cells. EVs carry proteins, lipids, nucleic acids and other substances derived from mother cells to receptor cells and induce corresponding functional responses, realizing information interaction with the external microenvironment, directly affecting tumor cell invasion, distant metastasis and other processes.

[0003] Traditional EV isolation methods include density gradient centrifugation, ultracentrifugation, size exclusion chromatography, etc., and detection methods include Western blotting, nanoparticle tracking analysis (NTA), transmission electron microscopy and other technologies. However, due to the small diameter of EVs and the complex plasma system, separation and detection often rely on large instruments. The above methods are time-consuming and cannot be adapted to clinical application scenarios, thus limiting the actual application of EVs in clinical scenarios.

[0004] Optical microcavity is a highly sensitive, small-sized, fast-response microcavity sensing technology. Based on the absorption characteristics of the object to be tested for light of different wavelengths, the transmission spectrum is analyzed to determine the absorption of light of different wavelengths in the sample. Optical microcavities can achieve ultra-sensitive detection at the single-molecule level in solution, but there are currently no reports on the use of optical microcavities for plasma extracellular vesicle detection to achieve early gastric cancer screening. The utility model patent with publication number CN208171807U discloses a lymphocyte bio-optical sensor based on an optofluidic capillary microcavity. This optical sensor is aimed at using optical devices for biological testing. It is necessary to functionalize the surface of the microcavity and then detect the cells. However, since the cell solution will cause the refractive index in the microcavity to change, continuous multi-sample detection cannot be achieved.

[0005] Therefore, it is necessary to develop an ultrasensitive optical microcavity detection and analysis method for plasma-derived EVs to overcome the problems to be solved in the existing technology. Summary of the invention

[0006] In view of this, the present invention aims to propose a method and application for establishing a cancer diagnosis model based on optical signals, so as to achieve trace, ultrasensitive, rapid and clinically applicable detection of plasma extracellular vesicles.

[0007] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0008] In a first aspect, the present invention provides a method for establishing a cancer diagnosis model based on optical signals, comprising the following steps:

[0009] S1. Use surface functionalized polystyrene microspheres to separate extracellular vesicles from healthy subjects and patient plasma.

[0010] S2. Using an optical microcavity detection device to detect the extracellular vesicles of healthy people and patients, respectively, to obtain spectral data, and performing variable standardization algorithm normalization on the spectral data to obtain training sets of healthy people and patients;

[0011] S3. Establish a CNN model, train the CNN model through healthy people and patient training sets, and establish a diagnostic model.

[0012] Furthermore, the surface functionalized polystyrene microspheres are polystyrene microspheres connected with distearoylphosphatidylethanolamine;

[0013] Preferably, the method for preparing the surface functionalized polystyrene microspheres comprises the following steps:

[0014] (1) Prepare a polystyrene microsphere solution, add EDC and NHS, and mix by inversion;

[0015] (2) centrifuging and washing the mixed solution to obtain carboxyl-activated polystyrene microspheres;

[0016] (3) Preparing DSPE-PEG1000-NH2 solution;

[0017] (4) mixing the carboxyl-activated polystyrene microspheres and the DSPE-PEG1000-NH2 solution by inversion to fully react and obtain a reaction solution;

[0018] (5) Centrifuging and washing the reaction solution to obtain surface functionalized polystyrene microspheres.

[0019] Further preferably, the method for preparing the surface functionalized polystyrene microspheres comprises the following steps:

[0020] (1) Prepare a 5 mg / mL polystyrene microsphere solution, add 10 mg / mL EDC and NHS, and mix inverted in a vertical mixer at 25°C for 1 hour to activate the carboxyl groups;

[0021] (2) The mixed solution was centrifuged at 3000 rcf for 10 minutes at room temperature, and washed with PBS solution three times to obtain carboxyl-activated polystyrene microspheres;

[0022] (3) preparing 0.5 mg / mL DSPE-PEG1000-NH2 solution and performing ultrasound-assisted dissolution;

[0023] (4) The carboxyl-activated polystyrene microspheres at a concentration of 5 mg / mL and the DSPE-PEG1000-NH2 solution at a concentration of 0.5 mg / mL were mixed in a vertical mixer for 1 hour to obtain a reaction solution;

[0024] (5) The reaction solution was centrifuged at 3000 rcf for 10 minutes at room temperature and washed three times with PBS solution to obtain surface functionalized polystyrene microspheres.

[0025] Furthermore, the separation method in step S1 comprises the following steps:

[0026] The surface functionalized polystyrene microspheres and plasma were fully shaken and mixed, and then incubated at room temperature to obtain a mixed solution;

[0027] The mixed solution is centrifuged, washed, resuspended, and filtered to obtain extracellular vesicles;

[0028] Preferably, the separation method in step S1 comprises the following steps:

[0029] The polystyrene microspheres connected with distearoylphosphatidylethanolamine and the plasma filtered by the membrane were fully shaken and mixed, and then incubated at room temperature to obtain a mixed solution;

[0030] After the mixture is centrifuged, the supernatant is removed, and PBS solution is added for washing and centrifugation, which is repeated 2-5 times. PBS solution is added again for resuspending, and the mixture is filtered through a water filter to obtain extracellular vesicles.

[0031] Furthermore, the dosage ratio of polystyrene microspheres to plasma is 2-5 mg:1 mL, for example, 2 mg:1 mL, 3 mg:1 mL, 4 mg:1 mL, 5 mg:1 mL; preferably 3 mg:1 mL.

[0032] Furthermore, the centrifugation condition is 3000-5000rcf, for example, 3000rc, 3500rcf, 4000rcf, 4500rcf, 5000rcf, the centrifugation time is 5-15min, for example, 5min, 8min, 10min, 12min, 15min, and the centrifugation temperature is 3-10℃, for example, 3℃, 5℃, 7℃, 10℃.

[0033] Furthermore, the optical microcavity detection device includes an adjustable laser, an optical fiber polarization controller, a tapered optical fiber, an optical resonator, a light detector, an acquisition card and an oscilloscope, the tapered optical fiber is coupled to the optical resonator, the feed end of the optical resonator is connected to the sample container, and the discharge end is connected to the negative pressure vacuum pump;

[0034] Preferably, the optical resonator is formed by heating and taper-drawing a hollow capillary; preferably, the process comprises the following steps: heating and stretching the hollow silica capillary using a hydrogen flame to taper the capillary to a diameter of about 30 μm; and then further heating the capillary to gradually expand it into a bubble.

[0035] Furthermore, the laser wavelength used for detection in step S2 is 778-780 nm, preferably 780 nm.

[0036] Furthermore, the CNN model includes an input layer, two convolutional layers, a fully connected layer and an output layer.

[0037] Furthermore, the convolutional layer includes a ReLU activation layer and a maximum pooling layer.

[0038] In a second aspect, the present invention provides an application of an optical microcavity detection device in preparing a cancer diagnosis product, wherein the cancer diagnosis comprises the following steps:

[0039] S1. Isolation of extracellular vesicles from plasma using surface-functionalized polystyrene microspheres;

[0040] S2. Detecting extracellular vesicles using an optical microcavity detection device to obtain spectral data, and normalizing the spectral data using a variable normalization algorithm to obtain detection data;

[0041] S3. Input the detection data into the model established by any of the above methods to obtain a diagnosis result.

[0042] Furthermore, the cancer is early gastric cancer.

[0043] In a third aspect, the present invention provides a cancer diagnosis system comprising a model established by any of the above methods.

[0044] Compared with the prior art, the method and application of establishing a cancer diagnosis model based on optical signals described in the present invention have the following advantages:

[0045] The method for establishing a cancer diagnosis model based on optical signals described in the present invention uses polystyrene microspheres to achieve rapid separation of extracellular vesicles in plasma, combines spectral signal changes to achieve feature capture of extracellular vesicles in plasma, and empowers through a convolutional neural network (CNN) algorithm. Standard diagnostic models are established based on spectral data in a training set, and a combined model is established after a comprehensive binary classification diagnostic model is used to establish a diagnostic process. After obtaining new spectral data in the training set, the source of the extracellular vesicles is determined by a standard combined diagnostic model to achieve disease diagnosis. The method has the advantages of low cost, high efficiency, and ultra-sensitive detection, overcomes the defects of traditional large-scale equipment dependence, complex experimental processes, and the need for a large number of samples. It is suitable for the diagnosis of early gastric cancer, and can also be used for scientific research purposes, population health record collection, drug analysis, etc. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0047] Figure 1 A schematic diagram of the structure of an optical resonator according to an embodiment of the present invention;

[0048] Figure 2 It is a schematic diagram of the structure of the optical microcavity detection device according to an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the results of SNV preprocessing of the full spectrum data of 10,000 pixels collected in an embodiment of the present invention;

[0050] Figure 4-5 The standard diagnostic model for early gastric cancer and healthy people established by the neural network model using the training set samples in the embodiment of the present invention, and the ROC curve and corresponding confusion matrix obtained by applying it to the validation set samples, and compared with the ROC curve of commonly used digestive system tumor markers;

[0051] Figure 6-7 The embodiment of the present invention uses the training set samples to train the neural network model to establish a standard diagnosis model for benign gastric diseases and healthy people, and applies the ROC curve and corresponding confusion matrix obtained by the validation set samples, and compares them with the ROC curves of commonly used digestive system tumor markers;

[0052] Figure 8-9 The standard diagnostic model for benign gastric diseases and early gastric cancer established by training a neural network model using training set samples in the embodiment of the present invention is applied to the ROC curve and corresponding confusion matrix obtained by applying it to the validation set samples, and compared with the ROC curve of commonly used digestive system tumor markers;

[0053] Figure 10-12This is the logical process of establishing a multi-classification combination model based on models 1, 2, and 3 in an embodiment of the present invention, and the diagnostic ROC curve and corresponding confusion matrix displayed when applied to all samples.

[0054] Description of reference numerals:

[0055] 1. Adjustable laser; 2. Fiber polarization controller; 3. Tapered fiber; 4. Optical resonator; 5. Photo detector; 6. Acquisition card; 7. Oscilloscope; 8. Sample container; 9. Negative pressure vacuum pump; 10. Waste liquid tank. DETAILED DESCRIPTION

[0056] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0057] The present invention will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0058] Example 1

[0059] The method of establishing a cancer diagnosis model based on optical signals in this embodiment includes the following steps:

[0060] (1) Construction of optical microcavity detection device

[0061] The optical resonator 4 was prepared by the “thermal expansion” method using a hollow silica capillary. First, the capillary was heated and stretched using a hydrogen flame, and the diameter of the capillary was tapered to about 30 μm. Then, the capillary was further heated to gradually expand into a bubble, and the optical resonator 4 was obtained. The high-quality Q factor of about 10 was obtained by Lorentz fitting. 7 ;

[0062] like Figure 2 As shown, the feed end of the optical resonator 4 is connected to the sample container, and the discharge end is connected to a negative pressure vacuum pump 9 (Fluigent, France, LU-FEZ-N800). A waste liquid pool 10 is also provided between the discharge end and the negative pressure vacuum pump 9 to collect the waste liquid after detection. Afterwards, the tunable laser 1, the optical fiber polarization controller 2, the tapered optical fiber 3, the optical resonator 4, the light detector 5, the acquisition card 6, and the oscilloscope 7 are connected in sequence, and the tapered optical fiber 3 is coupled to the optical resonator 4, as shown in FIG. Figure 1 As shown, the construction of the optical microcavity detection device is completed.

[0063] (2) Preparation of surface functionalized polystyrene microspheres

[0064] Prepare a 5 mg / mL polystyrene microsphere solution, add 10 mg / mL EDC and NHS, and mix inverted in a vertical mixer at 25°C for 1 hour to activate the carboxyl group;

[0065] The mixed solution was centrifuged at 3000 rcf for 10 minutes at room temperature, and washed three times with PBS solution to obtain carboxyl-activated polystyrene microspheres;

[0066] Prepare 0.5 mg / mL DSPE-PEG1000-NH2 solution and dissolve it with ultrasound assistance;

[0067] The polystyrene microspheres activated with carboxyl groups at a concentration of 0.05 mg / mL and the DSPE-PEG1000-NH2 solution at a concentration of 0.2 mg / mL were mixed in a vertical mixer for 1 hour to fully react to obtain a reaction solution;

[0068] The reaction solution was centrifuged at 3000 rcf for 10 minutes at room temperature and washed three times with PBS solution to obtain surface functionalized polystyrene microspheres.

[0069] (3) Isolation and extraction of extracellular vesicles from plasma

[0070] Prepare the plasma samples of healthy controls, benign gastric diseases and early gastric cancer to be tested, and prepare the test samples according to the following method. Place the polystyrene microspheres connected with distearoylphosphatidylethanolamine (DSPE) in the plasma that has passed through the 0.45um filter membrane, take 10 microliters of microspheres with a concentration of 3 mg / mL and 5 microliters of plasma, shake and mix them thoroughly, and incubate at room temperature for 30 minutes to obtain a mixed solution after incubation of plasma and microspheres, and complete the capture of plasma exosomes.

[0071] Due to the complex composition and low transmittance of the plasma solution, the collection of the transmitted optical signal is affected when detecting in the microfluidic channel built into the optical microcavity. Therefore, when using the optical microcavity to detect EVs on the surface of microspheres, it is necessary to remove free substances in the plasma. The specific steps are as follows: centrifuge the mixed solution after incubation of plasma and microspheres at 3500rcf, 4°C for 10 minutes, remove the supernatant, add 200μL PBS solution to wash and centrifuge, and repeat three times. Add 1000μL PBS solution again to resuspend, filter through a 10μm pore size water filter to obtain the required test sample.

[0072] (4) Detection of extracellular vesicles from different plasma sources using an optical microcavity detection device

[0073] The test samples are injected into the sample container 8 respectively, and the transmission spectrum of the laser at a wavelength of 780nm is collected. The data acquisition card collects 500 pixels per minute, each pixel corresponds to 60,000 points, and every 10,000 points is a spectrum. The tdms file collected in the acquisition card is extracted into an array, and then the pixel corresponding to the required spectral data time point is confirmed according to the image collected by the acquisition card, and then a required spectral data (including 10,000 points) is selected from the pixel points; the spectral data is normalized by the variable standardization (SNV) algorithm to improve the comparability and reliability of the data, such as Figure 3 shown.

[0074] (5) Building a Convolutional Neural Network (CNN) Model

[0075] The CNN model established in this embodiment includes an input layer, two convolutional layers, a fully connected layer, and an output layer. The input data size is set to 9999*1, the output label size is set to 2, and each convolutional layer includes a ReLU activation layer and a maximum pooling layer. The default stride s=1, padding p=0, and learning rate=0.001.

[0076] The operating principle of the CNN model is as follows:

[0077] a. The input layer accepts spectral data of size 9999*1;

[0078] b. The first convolutional layer sets 64 filters of size 5 to learn the features of the input data;

[0079] c. A maximum pooling layer with a pooling window of 2 is set after the first convolutional layer. The maximum value of a feature extracted by each filter is retained in the maximum pooling layer;

[0080] d. The second convolutional layer also sets 64 filters of size 5 and is followed by a maximum pooling layer;

[0081] e. The max pooling layer is followed by a flattening layer that connects the feature vector to a fully connected layer containing 128 neurons to generate a weighted sum;

[0082] f. The output of the final fully connected layer passes through the softmax activation function layer to predict the classification label.

[0083] (6) Train and optimize the CNN model

[0084] Three binary classification diagnostic models were established between healthy people, benign gastric diseases and early gastric cancer (model 1-healthy people and early gastric cancer, model 2-healthy people and benign gastric diseases, model 3-benign gastric diseases and early gastric cancer). The normalized data (including healthy people samples, benign gastric disease samples and early gastric cancer samples) were divided into training set samples and validation set samples. The corresponding training set samples were input to train the binary classification diagnostic model to obtain the diagnostic model, and the validation set samples were input for validation after training. Finally, the ROC curves based on the results of the validation set sample application and the levels of the digestive system tumor screening markers (CEA, CA125, CA199, CA-50, CA725) screened in the clinical information of all samples and the confusion matrices of the training set sample application models are shown as follows Figure 4-9 shown.

[0085] Example 2

[0086] like Fig.10 As shown in FIG. 1 , a combination model of the discrimination logic flow is set, and the three binary classification diagnosis models in Example 1 are combined into a multi-classification diagnosis model. The plasma to be tested is tested according to steps (1) to (4) of Example 1 to obtain spectral data. After the spectral data is input into the combination model, the combination model determines whether the source of the substance is a healthy population, a benign gastric disease, or an early gastric cancer. The ROC curve of the combination model is shown in FIG. Fig.11 The accuracy of the combined model is compared with that of a single model. Fig.12 ,From the results, it can be seen that the diagnostic accuracy of ,the combined model is significantly improved and can be used for ,auxiliary diagnosis of early gastric cancer.

[0087] The method of the present invention builds an optical microcavity detection device, prepares an optical resonator by using a thermal expansion method, and connects various components to complete the device construction; extracellular vesicles in plasma are separated and extracted by polystyrene microspheres, free substances are removed by centrifugation and PBS washing operations, and extracellular vesicles from different sources are detected by using an optical microcavity detection device, spectral signals collected by 10,000 pixels at a wavelength of 780 nm are collected and classified and diagnosed by applying a convolutional neural network model, and multiple binary classification diagnosis models and combined models are established, which can realize the diagnosis and discrimination of healthy people, benign gastric diseases and early gastric cancer, and the method shows good accuracy and reliability in experiments.

Claims

1. A method for establishing a cancer diagnosis model based on optical signals, characterized in that: The following steps are involved: S1. Using surface functionalized polystyrene microspheres to separate extracellular vesicles from healthy people and patient plasma, respectively, wherein the surface functionalized polystyrene microspheres are polystyrene microspheres connected with distearoylphosphatidylethanolamine; S2. Using an optical microcavity detection device to detect the extracellular vesicles of healthy people and patients, respectively, to obtain spectral data, and performing variable standardization algorithm normalization on the spectral data to obtain training sets of healthy people and patients; S3. Establish a CNN model, train the CNN model through healthy people and patient training sets, and establish a diagnostic model.

2. The method according to claim 1, characterized in that: The preparation method of the surface functionalized polystyrene microspheres comprises the following steps: (1) Prepare a polystyrene microsphere solution, add EDC and NHS, and mix by inversion; (2) centrifuging and washing the mixed solution to obtain carboxyl-activated polystyrene microspheres; (3) Preparing DSPE-PEG1000-NH2 solution; (4) mixing the carboxyl-activated polystyrene microspheres and the DSPE-PEG1000-NH2 solution by inversion to fully react and obtain a reaction solution; (5) Centrifuging and washing the reaction solution to obtain surface functionalized polystyrene microspheres.

3. The method according to claim 1, characterized in that: The separation method in step S1 comprises the following steps: The surface functionalized polystyrene microspheres and plasma were fully shaken and mixed, and then incubated at room temperature to obtain a mixed solution; The mixed solution is centrifuged, washed, resuspended and filtered to obtain extracellular vesicles.

4. The method according to claim 3, characterized in that: The dosage ratio of surface functionalized polystyrene microspheres to plasma is 2-5 mg:1 mL.

5. The method according to claim 3, characterized in that: The centrifugal conditions are 3000-5000 rcf, the centrifugal time is 5-15 min, and the centrifugal temperature is 3-10°C.

6. The method according to claim 1, characterized in that: The optical microcavity detection device comprises an adjustable laser, an optical fiber polarization controller, a tapered optical fiber, an optical resonator, a light detector, an acquisition card and an oscilloscope. The tapered optical fiber is coupled to the optical resonator, the feed end of the optical resonator is connected to a sample container, and the discharge end is connected to a negative pressure vacuum pump.

7. The method according to claim 6, characterized in that: The optical resonator is formed by heating and taper of a hollow capillary.

8. The method according to claim 1, characterized in that: The laser wavelength used for detection in step S2 is 778-780 nm.

9. The method according to claim 1, characterized in that: The CNN model includes an input layer, two convolutional layers, a fully connected layer and an output layer.

10. The method according to claim 9, characterized in that: The convolutional layer includes a ReLU activation layer and a maximum pooling layer.

11. A cancer diagnosis system comprising a model established by any of the methods described in claims 1-10.

Images