Blood parameter optical detection and calibration method based on RKHS-PLS

Abstract

The invention discloses a blood parameter optical detection and calibration method based on RKHS-PLS, and belongs to the field of blood physiological parameter detection.The blood parameter optical detection and calibration method comprises the following steps that S1, incident light intensity and transmitted light intensity of each wavelength are recorded; s2, calculating preliminary predicted values of hematocrit and oxyhemoglobin saturation of each wavelength; s3, training a partial least square regression method calibration model; s4, evaluating the trained partial least square regression method calibration model; and S5, inputting the newly collected incident light intensity and transmission light intensity under each wavelength into the partial least square regression method calibration model output in the step S4, and generating final prediction values of hematocrit and oxyhemoglobin saturation. By adopting the blood parameter optical detection and calibration method based on RKHS-PLS, through integration of physical modeling and data analysis, the theoretical deviation of the traditional Lambert-Beer law is made up, the physical interpretability is reserved, the detection precision is remarkably improved, and the calibration frequency is greatly reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of blood physiological parameter detection, and particularly relates to a blood parameter optical detection and calibration method based on RKHS-PLS. BACKGROUND

[0002] Extracorporeal membrane oxygenation (ECMO) as a core extracorporeal life support technology for patients with severe cardiopulmonary failure, its core function is to maintain the patient's vital signs through extracorporeal respiration and circulation support, and has been widely used in cardiogenic shock, severe acute respiratory failure, severe circulatory failure and other clinical scenarios. The therapeutic effect of ECMO system fundamentally depends on the accurate real-time monitoring of key physiological parameters in blood, i.e. hematocrit and blood oxygen saturation: among them, blood oxygen saturation as the core index of respiratory and circulatory function evaluation, directly quantifies the oxygen-carrying capacity of hemoglobin, and reflects the oxygen supply state of the patient's body; hematocrit represents the volume ratio of red blood cells in whole blood, and is the key basis for diagnosing anemia and warning thrombosis complications. Clinical practice shows that real-time and accurate monitoring of hematocrit and blood oxygen saturation not only can timely capture the changes in the patient's body state, but also can provide data support for early intervention of complications (such as hypoxemia and anemic shock), thus becoming an indispensable part of the ECMO treatment process.

[0003] However, the existing blood physiological parameter detection mostly adopts the "gold standard" equipment (such as SYSMEX XN-1000 blood cell analyzer, Siemens RAPIDLab1265 blood gas analyzer), which can provide high-precision detection results, but needs to collect patient blood samples through puncture for offline analysis - this "blood sampling detection" mode has two core problems: first, it cannot realize real-time monitoring, and there is a time difference between sample collection and detection result feedback, which is difficult to match the dynamic changes of the patient's physiological state in ECMO treatment; second, frequent blood sampling will increase the risk of infection and blood loss for patients, especially for severe patients.

[0004] To solve the limitations of traditional blood sampling detection, researchers further developed the use of near-infrared spectroscopy (NIRS) to realize the in vitro detection of hematocrit and blood oxygen saturation based on the optical absorption characteristics. At present, although the non-invasive monitoring method based on optical principles avoids the disadvantages of invasive detection, its measurement accuracy still cannot meet the requirements of accurate monitoring in actual application, especially in clinical scenarios such as extracorporeal circulation (e.g. ECMO pipeline). The technical bottleneck mainly comes from the following aspects: First, the classic optical model (such as Lambert-Beer's law) assumes that the medium is a uniform non-scattering system, but the presence of red blood cells in blood causes strong light scattering, resulting in a significant increase in the actual optical path length compared to the physical thickness, which causes the "optical path multiplication" effect. The theoretical model cannot accurately describe the real transmission behavior of light in blood, resulting in inherent bias in the calculation of hematocrit and blood oxygen saturation. Second, individual physiological differences, hemodynamic changes (such as turbulence) in extracorporeal circulation pipelines, and differences in pipeline wall thickness can further cause fluctuations, distortion, and baseline drift of the light intensity signal. The existing technology lacks an effective compensation mechanism for such dynamic interference, making it difficult to achieve stable detection of the signal. Finally, when simultaneously monitoring multiple parameters such as blood oxygen saturation and hematocrit, the existing schemes and devices have problems such as frequent calibration, cumbersome operation, and difficulty in achieving truly continuous and accurate monitoring, resulting in measurement errors generally exceeding the clinically acceptable 5% threshold, limiting their accuracy and stability in complex clinical environments. SUMMARY

[0005] The purpose of the present application is to provide a blood parameter optical detection and calibration method and system based on RKHS-PLS, which solves the above technical problems.

[0006] To achieve the above purpose, the present application provides a blood parameter optical detection and calibration method based on RKHS-PLS, comprising the following steps: S1, considering the optical absorption characteristics of hematocrit and blood oxygen saturation, emitting monochromatic light of multiple wavelengths to the blood extracorporeal circulation pipeline, and receiving the transmitted light through the blood in the blood extracorporeal circulation pipeline in real time, recording the incident light intensity and transmitted light intensity at each wavelength after amplification and filtering, obtaining the original light intensity signal, and using the regenerative Hilbert space anomaly signal identification and correction to preprocess the original light intensity signal, obtaining the corrected light intensity signal; S2, based on the incident light intensity and transmitted light intensity at each wavelength recorded in step S1, using the corrected Lambert-Beer's law with calibrated factors and optical path multiplication factors to calculate the preliminary prediction values of hematocrit and blood oxygen saturation at each wavelength; S3. Based on the corrected light intensity signal obtained in step S1, and combined with the preliminary predicted values ​​of absorbance, hematocrit and blood oxygen saturation at each wavelength calculated in step S2, the independent variable matrix is ​​obtained, and the actual values ​​of hematocrit and blood oxygen saturation are set as dependent variables. After standardization, the model is calibrated by partial least squares regression and trained. S4. Evaluate the trained partial least squares regression calibration model using multidimensional evaluation indicators. If the requirements are met, output the trained partial least squares regression calibration model; otherwise, return to step S3. S5. Process the newly acquired incident light intensity and transmitted light intensity at each wavelength according to steps S2 and S3, and then input them into the partial least squares regression calibration model output in step S4 to generate the final predicted values ​​of hematocrit and blood oxygen saturation.

[0007] Preferably, the multi-wavelength monochromatic light mentioned in step S1 includes monochromatic light of 660nm, 808nm, 940nm, and 1310nm. Specifically, it includes the following steps: S11. Construct a regenerated Hilbert spatial representation model for light intensity at each wavelength, wherein the kernel function expression of the regenerated Hilbert spatial representation model is as follows: (1); (2); In the formula, wavelength Next, the A light intensity sample and the A light intensity sample Kernel function value in regenerated Hilbert space; wavelength The corresponding kernel width parameter; This is the function for calculating standard deviation. wavelength The normal light intensity subset is selected from the following; S12. Based on the kernel function described in step S11, construct the Gram matrix for each wavelength of light intensity to quantify the global correlation of the samples. Specifically, for each wavelength... of Each light intensity sample has its Gram matrix. The element is defined as the light intensity sample in the regenerated Hilbert space. The inner product of the two is: (3); In the formula, The first in the Gram matrix line, number Column elements, is a positive definite matrix under Mercer condition of Gaussian kernel, and its spectral decomposition , is the eigenvector matrix obtained by spectral decomposition of Gram matrix, is the transpose operation, and is used to reflect the correlation of light intensity samples, are the first eigenvalues respectively after spectral decomposition of Gram matrix; is the mapping function of light intensity samples to ; S13, solving the projection operator of normal signal subspace by robust optimization in to separate normal light intensity signal and singular signal; S131, defining the projection coefficient vector , are the first components of the projection coefficient vector , and the optimization objective is to minimize the reconstruction error of normal light intensity signal, while suppressing the contribution of singular value by norm, and the optimization model is: (4); wherein, is a regularization parameter; is the projection coefficient related to the th light intensity sample and the th light intensity sample for wavelength ; S132, iteratively performing step S131 until convergence to obtain the optimal coefficient , and defining the projection residual of light intensity sample on as , and taking it as a quantitative index for singular value identification: (5); wherein, is the projection operator of normal signal subspace ; S133, using the reproducing property of reproducing Hilbert space to expand the projection residual into the following form: (6); wherein, is the projection coefficient related to the th light intensity sample for wavelength ;light intensity sample projection coefficient, the optimal projection coefficient value obtained after robust optimization iterative convergence; is the wavelength , the projection coefficient of the first light intensity sample projection coefficient, the optimal projection coefficient value obtained after robust optimization iterative convergence; is the wavelength , the first light intensity sample , the first light intensity sample kernel function value in the reproducing Hilbert space; is the wavelength , the total number of light intensity samples involved in modeling and optimization; S134, screening light intensity is normal light intensity, and the remaining light intensity is regarded as a singular value and is removed; S135, based on the screened , the control false positive probability is calculated: (7); wherein, (8); (9); wherein, is the wavelength , the threshold value set based on statistical significance; is the normal residual mean; is the normal residual standard deviation; 3.29 is the 99.9% quantile of the standard normal distribution; is the wavelength , the total number of normal light intensity samples obtained by screening; is the wavelength , the first light intensity sample determined to be normal; S136, when , the first singular value is determined, and the singular value is physically consistent, ensuring that the modified light intensity signal complies with the non-negativity and attenuation rules of light intensity, and the modification target is: (10); Substitute the projection operator expression and derive to obtain: (11); wherein, is the wavelength , the first A light intensity sample identified as an outlier (anomaly) With the A normal light intensity sample Kernel function value in regenerated Hilbert space; The light intensity variable to be corrected; It is the set of non-negative real numbers.

[0008] Preferably, step S2 specifically includes the following steps: S21, using calibrated factors and optical path multiplier The modified Lambert-Beer law is used to calculate the total absorbance of blood: (12); In the formula, Indicates wavelength Total absorbance of blood sample; This indicates a multiplier effect that increases the optical path length due to light scattering in the blood. Indicates blood thickness; among which, calibration factor Used to compensate for the loss of light signal caused by the thickness of the extracorporeal circulation tubing and scattering in the blood; Indicates wavelength Below, the molar extinction coefficient of light-absorbing substances in the blood; Indicates the molar concentration of light-absorbing substances in the blood; S22. Calculate the preliminary predicted values ​​of absorbance, hematocrit, and blood oxygen saturation at each wavelength: (13); (14); In the formula, and These represent preliminary predicted values ​​for hematocrit and blood oxygen saturation, respectively. Red blood cell concentration; Indicates plasma concentration; and They represent wavelengths respectively. The concentrations of reduced hemoglobin and oxyhemoglobin; S23. By combining formula (15) with the modified Lambert-Beer law, the absorbance ratios at each wavelength are obtained. relation: (15); In the formula, Indicates wavelength Total absorbance of blood sample; Furthermore, simplifying formulas (13) and (14) yields: (16); (17); wherein, , , , , , , , , , , and are constants; , , , respectively represent the absorbance ratio at 660nm, 808nm, 940nm, 1310nm wavelength.

[0009] Preferably, step S3 specifically comprises the following steps: S31, integrating the preliminary predicted values of the transmitted light intensity, the absorbance at each wavelength, the hematocrit and the blood oxygen saturation to obtain the independent variable , and using the true values of the hematocrit and the blood oxygen saturation measured by the blood cell counter and the blood gas analyzer to obtain the dependent variable ; S32, standardization processing: (18); wherein, and respectively represent the standardized independent variable and the dependent variable; and respectively represent the mean difference of the independent variable and the dependent variable; and respectively represent the standard deviation of the independent variable and the dependent variable; S33, using the following regression algorithm to establish the mapping relationship between the standardized independent variable and the dependent variable: (19); wherein, represents the regression coefficient matrix; represents the residual matrix.

[0010] Preferably, in step S4, the root mean square error of prediction , the mean absolute error , the mean absolute percentage error and the coefficient of determination are used as indexes to evaluate the calibrated partial least squares regression model after training.

[0011] Preferably, in step S5, the final prediction values of hematocrit and blood oxygen saturation are calculated by reverse normalization : (20).

[0012] The system for performing the RKHS-PLS-based blood parameter optical detection and calibration method comprises: a light source for emitting monochromatic light of multiple wavelengths; a blood extracorporeal circulation pipeline for simulating a human vascular environment; a photodetector for receiving transmitted light passing through blood in the blood extracorporeal circulation pipeline in real time; an extracorporeal circulation host for recording incident light intensity and transmitted light intensity at each wavelength after amplification filtering and outputting final prediction values of hematocrit and blood oxygen saturation; a circulation pump arranged on the blood extracorporeal circulation pipeline for simulating a heart.

[0013] Therefore, the present application adopts the above-mentioned RKHS-PLS-based blood parameter optical detection and calibration method and system, which has the following beneficial effects: 1. Significantly improved detection accuracy: the detection error rate of hematocrit and blood oxygen saturation is reduced to below 5%, wherein the mean absolute error (MAE) of hematocrit is increased by 1.4808%, and the MAE of blood oxygen saturation is increased by 0.3885%; the root mean square error (RMSE) of hematocrit prediction is 1.4071%, the determination coefficient (R 2 ) is 0.9801, the RMSE of blood oxygen saturation prediction is 1.2299%, and the R 2 is 0.9348, which is more accurate than traditional linear regression algorithms and similar partial least squares regression (PLS) research results in recent years; 2. Significantly reduced calibration frequency: the calibration requirement of the detection instrument is optimized from "calibration before each use" to "calibration once every few months", reducing the clinical operation process and time cost, and improving the use convenience and efficiency of the detection equipment; 3. Retained physical interpretability: the partial least squares regression method calculates the initial estimate values of hematocrit and blood oxygen saturation based on the modified Lambert-Beer law as a theoretical baseline, avoiding the problem of lack of physical meaning in pure data-driven algorithms, and ensuring the reliability and traceability of the detection principle; 4. Enhanced adaptability to complex environments: by fusing physical model theoretical calculation values and multi-wavelength light intensity signal features, the partial least squares regression method algorithm overcomes the limitations of light scattering effects, blood flow dynamics interference, and other limitations in the ECMO complex environment, improving the stability and applicability of the detection system in actual clinical scenarios; 5. Support clinical diagnosis and treatment safety: can realize the accurate real-time monitoring of hematocrit and blood oxygen saturation during ECMO operation, can timely detect the changes of patient's body state, provide reliable data support for effective prevention of complications, and then help to improve the medical quality and patient life safety guarantee level.

[0014] The technical solutions of the present application will be further described in detail below by means of the drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 Flow chart of the blood parameter optical detection and calibration method based on RKHS-PLS of the present application; Figure 2 Calibration curve result graph described in the experimental example of the present application, wherein (a) is the hematocrit calibration curve graph of sample 1, (b) is the hematocrit calibration curve graph of sample 2, (c) is the hematocrit calibration curve graph of sample 3, (d) is the blood oxygen saturation calibration curve graph of sample 1, (e) is the blood oxygen saturation calibration curve graph of sample 2, and (f) is the blood oxygen saturation calibration curve graph of sample 3; Figure 3 Linear regression and partial least squares regression calibration hematocrit result comparison graph described in the experimental example of the present application, wherein (a) is a linear regression curve graph; (b) is a partial least squares regression calibration fitting curve graph; (c) is a linear regression and true value comparison result graph; (d) is a partial least squares regression method predicted value and true value comparison result graph; Figure 4 Linear regression and partial least squares regression calibration blood oxygen saturation result comparison graph described in the experimental example of the present application, wherein (a) is a linear regression curve graph; (b) is a partial least squares regression calibration fitting curve graph; (c) is a linear regression and true value comparison result graph; (d) is a partial least squares regression method predicted value and true value comparison result graph. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear and understandable, the embodiments of the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present application and do not limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application. The examples of the embodiments are shown in the drawings, wherein the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions throughout.

[0017] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.

[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] like Figure 1 As shown, the optical detection and calibration method for blood parameters based on RKHS-PLS includes the following steps: S1. Considering the optical absorption characteristics of hematocrit and blood oxygen saturation, multi-wavelength monochromatic light is emitted into the extracorporeal circulation tubing, and the transmitted light of the blood passing through the extracorporeal circulation tubing is received in real time. After amplification and filtering, the incident light intensity and transmitted light intensity at each wavelength are recorded. The multi-wavelength monochromatic light mentioned in step S1 includes monochromatic light of 660nm, 808nm, 940nm, and 1310nm. Specifically, it includes the following steps: S11. Because the attenuation patterns of light intensity differ for different wavelengths, it is necessary to specify the attenuation pattern for each wavelength. Define a Gaussian kernel function to generate a regenerating Hilbert space that satisfies the regeneration condition. The kernel function indirectly represents the inner product relationship of light intensity samples in high-dimensional space, avoiding the computational complexity of explicit mapping. The kernel function expression is: (1); (2); In the formula, wavelength Next, the A light intensity sample and the A light intensity sample Kernel function value in regenerated Hilbert space; wavelength The corresponding kernel width parameter; This is the function for calculating standard deviation. wavelength The normal light intensity subset is selected from the following; S12. Based on the kernel function described in step S11, construct the Gram matrix for each wavelength of light intensity to quantify the global correlation of the samples. Specifically, for each wavelength... of Each light intensity sample has its Gram matrix. The element is defined as the light intensity sample in the regenerated Hilbert space. The inner product of the two is: (3); In the formula, The first in the Gram matrix line, number Column elements, Given a Gaussian kernel, the Mercer condition positive definite matrix, and its spectral decomposition... middle, The eigenvector matrix is ​​obtained by spectral decomposition of the Gram matrix. For the transpose operation, the eigenvalues Used to reflect the correlation of light intensity samples The spectral decomposition of the Gram matrix is ​​respectively the first... One eigenvalue; For light intensity samples to The mapping function; S13, in The normal signal subspace is solved through robust optimization. Projection operator To separate normal light intensity signals from exotic signals; S131. Define the projection coefficient vector. , These are the projection coefficient vectors. The Each component is optimized to minimize the reconstruction error of the normal light intensity signal, while simultaneously... Norm regularization suppresses the contribution of singular values, and the optimized model is: (4); In the formula, For regularization parameters; For wavelength , No. A light intensity sample With the A light intensity sample The relevant projection coefficients; in this embodiment, they are determined through 5-fold cross-validation for each wavelength to avoid optimization bias caused by differences in noise levels at different wavelengths; equality constraints. Ensure that the translation of the projection operator remains unchanged; S132, iterate through step S131 until convergence, and obtain the optimal coefficients. Define light intensity sample exist The projection residual on is And use it as a quantitative indicator for singular value identification: (5); In the formula, For normal signal subspace The projection operator; S133, Utilizing the regenerative nature of regenerated Hilbert spaces Projection residual Expanded into the following form: (6); In the formula, wavelength Below, with the A light intensity sample The optimal projection coefficient values ​​are obtained after robust optimization iterative convergence. wavelength Below, with the A light intensity sample The optimal projection coefficient values ​​are obtained after robust optimization iterative convergence. wavelength Next, the A light intensity sample With the A light intensity sample Kernel function value in regenerated Hilbert space; wavelength The total number of light intensity samples participating in modeling and optimization; S134, Filtering The light intensity is normal light intensity, and the remaining light intensity is considered an outlier and discarded. S135, Based on the filtered Calculate the probability of misjudgment: (7); in, (8); (9); In the formula, wavelength The threshold is set based on statistical significance. This represents the mean of the normal residuals. The value is the standard deviation of the normal residuals; 3.29 is the 99.9th quantile of the corresponding standard normal distribution, which can control the misjudgment rate of normal light intensity to within 0.1%. wavelength The total number of normal light intensity samples obtained through screening; wavelength Next, the One light intensity sample was determined to be normal; S136, when At that time, the judgment Singular values Furthermore, physical consistency corrections are performed on singular values ​​to ensure the corrected light intensity signal. Consistent with the non-negativity and attenuation laws of light intensity, the correction target is: (10); Substituting the projection operator expression and taking the derivative, we get: (11); In the formula, wavelength Next, the A light intensity sample identified as an outlier (anomaly) With the A normal light intensity sample Kernel function value in regenerated Hilbert space; The light intensity variable to be corrected; It is the set of non-negative real numbers.

[0020] S2. Based on the incident light intensity and transmitted light intensity at each wavelength recorded in step S1, the Lambert-Beer law, corrected by calibration factor and optical path multiplication factor, is used to calculate the preliminary predicted values ​​of hematocrit and blood oxygen saturation at each wavelength. S21, using calibrated factors and optical path multiplier The modified Lambert-Beer law is used to calculate the total absorbance of blood: (12); In the formula, Indicates wavelength Total absorbance of blood sample; This indicates a multiplier effect that increases the optical path length due to light scattering in the blood. Indicates blood thickness; among which, calibration factor Used to compensate for the loss of light signal caused by the thickness of the extracorporeal circulation tubing and scattering in the blood; Indicates wavelength Below, the molar extinction coefficient of light-absorbing substances in the blood; Indicates the molar concentration of light-absorbing substances in the blood; S22. Calculate the preliminary predicted values ​​of absorbance, hematocrit, and blood oxygen saturation at each wavelength: (13); (14); In the formula, and These represent preliminary predicted values ​​for hematocrit and blood oxygen saturation, respectively. Red blood cell concentration; represents the plasma concentration; and respectively represent the concentrations of reduced hemoglobin and oxygenated hemoglobin at the wavelength S23, in order to eliminate system errors and enhance the sensitivity to changes in blood components, combined with the selected four characteristic wavelengths (660 nm, 808 nm, 940 nm, 1310 nm), by formula (15) and the revised Lambert-Beer law, the absorbance ratio at each wavelength relationship: (15); wherein, represents the total absorbance of blood at the wavelength absorbance ratio not only eliminates the influence of incident light intensity and path loss, but also enhances the sensitivity of the model to changes in different blood components; further simplify formula (13) and formula (14) to obtain: (16); (17); wherein, , , , , , , , , , , and are constants; , , , respectively represent the absorbance ratios at wavelengths of 660 nm, 808 nm, 940 nm and 1310 nm.

[0021] After the derivation of the Lambert-Beer law and its modified model, the physical mapping relationship from the multi-wavelength transmitted light intensity to the key parameters of blood (hematocrit and blood oxygen saturation) is established. This model not only reveals the basic law of light propagation in blood, but also provides a theoretical basis and initial estimate for the subsequent calibration algorithm.

[0022] ​​S3, based on the corrected light intensity signal obtained in step S1, combined with the preliminary predicted values of the hemoglobin and oxygen saturation of each wavelength calculated in step S2, obtain the independent variable matrix, and set the true values of the hemoglobin and oxygen saturation as the dependent variable, after standardization, input the partial least squares regression calibration model for training; Step S3 specifically includes the following steps: S31, integrate the transmitted light intensity, the preliminary predicted values of the hemoglobin and oxygen saturation of each wavelength to obtain the independent variable , and use the true values of the hemoglobin and oxygen saturation measured by the blood cell counter and the blood gas analyzer to obtain the dependent variable ; S32, standardization processing: (18); , and , and respectively represent the standardized independent variable and dependent variable; , and respectively represent the standard deviation of the independent variable and the dependent variable; S33, use the following regression algorithm to establish the mapping relationship between the standardized independent variable and the dependent variable: (19); , and represent the regression coefficient matrix; represent the residual matrix.

[0023] In step S4, the root mean square error of prediction , the mean absolute error , the mean absolute percentage error , and the determination coefficient are used as indicators to evaluate the trained partial least squares regression calibration model.

[0024] S4, use multi-dimensional evaluation indicators to evaluate the trained partial least squares regression calibration model, if the requirements are met, output the trained partial least squares regression calibration model, otherwise, return to step S3; S5, process the incident light intensity and transmitted light intensity at each wavelength of the newly collected data according to steps S2 and S3, and then input them into the partial least squares regression calibration model output in step S4 to generate the final predicted values of the hemoglobin and oxygen saturation.

[0025] In step S5, the final predicted values of the hemoglobin and oxygen saturation are calculated by reverse normalization : (20).

[0026] A system for performing optical detection and calibration methods for blood parameters based on RKHS-PLS includes: A light source is used to emit monochromatic light of multiple wavelengths; in this embodiment, the light source is a laser diode with a center wavelength of 660nm, 808nm, 940nm and 1310nm. Extracorporeal blood circulation tubing is used to simulate the human vascular environment; A photodetector is used to receive transmitted light from blood passing through extracorporeal circulation tubing in real time. The extracorporeal circulation unit is used to record the incident light intensity and transmitted light intensity at each wavelength after amplification and filtering, and output the final predicted values ​​of hematocrit and blood oxygen saturation. A circulation pump is installed in the extracorporeal blood circulation tubing to simulate the heart.

[0027] Experimental Example Firstly, to meet the detection requirements of both hematocrit and blood oxygen saturation, different sample preparation protocols were designed to ensure coverage of multiple concentration / saturation levels: (1) Steps for preparing hematocrit samples: First, prepare 6 clean, dry test tubes, labeled 1-6. Then, add 9 mL, 8 mL, 7 mL, 6 mL, and 5 mL of fresh animal blood to test tubes 1-5 respectively. Next, add physiological saline to test tubes 1-5 to a total volume of 10 mL. Homogenize the blood by gently shaking or constant stirring to obtain samples with different hematocrit concentrations. Meanwhile, use test tube 6 as a control group, adding 10 mL of undiluted blood without physiological saline.

[0028] (2) Preparation of blood oxygen saturation samples (gradient saturation) Record blood samples from any one of the test tubes (1-6) at six different contact time points to obtain six samples with different blood oxygen saturation levels.

[0029] Secondly, using the hematocrit and blood oxygen saturation detected by the system described in this invention (without calibration using the modified Lambert-Beer law method described in this invention) as the x-axis, and the actual values ​​of hematocrit and blood oxygen saturation detected by the blood routine analyzer and blood gas analyzer as the y-axis, three samples were plotted as follows. Figure 2 The calibration curves are shown. It can be seen that the correlation coefficients of the three calibration curves for hematocrit are R1 and R2, respectively. 2 =0.9663,R2 2 =0.9893, R3 2 =0.9605, the correlation coefficients of the three calibration curves for blood oxygen saturation are R1 and R2 respectively. 2 =0.9367,R2 2 =0.9863, R32 =0.9847. That is, the system detects a good linear relationship with the results of the blood cell analyzer and the blood gas analyzer, but the error between the two is larger, and further calibration is needed to improve accuracy.

[0030] Under the premise of the system, the method is used to correct hematocrit and blood oxygen saturation, and results are shown in Figure 3 and Figure 4 After calibration, the RSME of the predicted hematocrit is 1.4071%, the MAE is 1.0492%, the MAPE is 4.1177%, and the R 2 for 0.9801. After calibration, the RSME of the predicted blood oxygen saturation is 1.2299%, the MAE is 1.0615%, the MAPE is 1.1313%, and the R 2 for 0.9348, which is higher than the accuracy of the traditional method, thereby proving the effectiveness of the method.

[0031] Finally, the NIRS combined with the partial least squares regression method modeling method is used as a control group, and the results are shown in Table 1. It can be seen that the accuracy of the method is higher than that of the traditional method, thereby proving the superiority of the method compared with the traditional method.

[0032] Table 1 Evaluation results

[0033] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application but not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that: it can still modify or equivalently replace the technical solutions of the present application, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present application.

Claims

1. A method for blood parameter optical detection and calibration based on RKHS-PLS, characterized in that: The method comprises the following steps: S1, considering the optical absorption characteristics of hematocrit and blood oxygen saturation, emitting multi-wavelength monochromatic light to the blood extracorporeal circulation pipeline, and receiving the transmitted light through the blood in the blood extracorporeal circulation pipeline in real time, recording the incident light intensity and the transmitted light intensity at each wavelength after amplification filtering, obtaining the original light intensity signal, and using the regenerative Hilbert space anomaly signal identification and correction preprocessing original light intensity signal to obtain the corrected light intensity signal; S2, based on the incident light intensity and the transmitted light intensity at each wavelength recorded in step S1, using the corrected Lambert-Beer law with the calibration factor and the optical path multiplication factor, calculating the preliminary prediction value of hematocrit and blood oxygen saturation at each wavelength; S3, based on the corrected light intensity signal obtained in step S1, combining the preliminary prediction value of absorbance hematocrit and blood oxygen saturation at each wavelength calculated in step S2, obtaining the independent variable matrix, and setting the true value of hematocrit and blood oxygen saturation as the dependent variable, after standardization processing, inputting the partial least squares regression calibration model for training; S4, using multi-dimensional evaluation index to evaluate the trained partial least squares regression calibration model, if it meets the requirements, output the trained partial least squares regression calibration model, otherwise, return to step S3; S5, processing the newly collected incident light intensity and transmitted light intensity at each wavelength by steps S2 and S3, and then inputting the partial least squares regression calibration model output by step S4 to generate the final prediction value of hematocrit and blood oxygen saturation.

2. The RKHS-PLS based optical detection and calibration method of blood parameters according to claim 1, characterized in that: The multi-wavelength monochromatic light of step S1 comprises 660nm, 808nm, 940nm and 1310nm monochromatic light; The method comprises the following steps: S11, constructing a regenerative Hilbert space representation model of light intensity at each wavelength, wherein the kernel function expression of the regenerative Hilbert space representation model is as follows: （1）； （2）； wherein is the wavelength The following, the first Light intensity sample And the first Light intensity sample The value of the kernel function in the reproducing Hilbert space; is the wavelength The corresponding kernel width parameter; is the standard deviation calculation function, is the wavelength Normal light intensity subset screened under S12, based on the kernel function described in step S11, construct the Gram matrix of each wavelength light intensity to quantify the global correlation of the sample, wherein, for the wavelength of the light intensity samples, the element of the Gram matrix is defined as the inner product of the light intensity samples in the reproducing Hilbert space , and the following is obtained: （3）； In the formula, The first in the Gram matrix line, number Column elements, Given a Gaussian kernel, the Mercer condition positive definite matrix, and its spectral decomposition... middle, The eigenvector matrix is ​​obtained by spectral decomposition of the Gram matrix. For the transpose operation, the eigenvalues Used to reflect the correlation of light intensity samples The spectral decomposition of the Gram matrix is ​​respectively the first... One eigenvalue; For light intensity samples to The mapping function; S13、in robustly optimize the projection operator of the normal signal subspace to separate normal light intensity signals from singular signals;​ S131、define the projection coefficient vector , are the first components of the projection coefficient vector , the optimization objective is to minimize the reconstruction error of the normal light intensity signal, while suppressing the contribution of singular values by the norm, and the optimization model is: （4）； wherein is a regularization parameter; is a wavelength-dependent parameter; , the projection coefficient associated with the th light intensity sample , the projection coefficient associated with the th light intensity sample . S132, loop iteration step S131 until convergence, get the optimal coefficient , define the light intensity sample The projection residual on is , and it is taken as a quantitative index for singular value identification: （5）； wherein is the projection operator onto the normal signal subspace of the normal signal subspace S133, exploiting the regenerativity of the reproducing Hilbert space The projection residual is unfolded into the form （6）； In the formula, wavelength Below, with the A light intensity sample The optimal projection coefficient values ​​are obtained after robust optimization iterative convergence. wavelength Below, with the A light intensity sample The optimal projection coefficient values ​​are obtained after robust optimization iterative convergence. wavelength Next, the A light intensity sample With the A light intensity sample Kernel function value in regenerated Hilbert space; wavelength The total number of light intensity samples participating in modeling and optimization; S134, screening The light intensity is normal light intensity, and the remaining light intensity is regarded as a singular value and removed. S135、based on the screening , calculate the control error probability: （7）； Wherein, （8）； （9）； In the formula, is the wavelength Below, the threshold value is set based on statistical significance; is the normal residual mean value; is the normal residual standard deviation; 3.29 is the 99.9% quantile of the standard normal distribution; is the wavelength Below, the total number of normal light intensity samples obtained by screening; is the wavelength Below, the first normal light intensity sample determined; S136、When the singular value is determined , the singular value is physically consistent, and the modified light intensity signal meets the non-negativity and attenuation rules of light intensity, and the modification target is: （10）； Substitute the projection operator expression and derive to obtain: （11）； In the formula, is the wavelength Next, the light intensity sample identified as a singular value (anomaly) and the normal light intensity sample kernel function value in the reproducing Hilbert space; is the light intensity variable to be corrected; is a set of non-negative real numbers.

3. The RKHS-PLS based optical detection and calibration method of blood parameters according to claim 1, characterized in that: Step S2 specifically comprises the following steps: S21, using the calibration factor and the optical path multiplication factor Calculate the total blood absorbance using the modified Lambert-Beer law: （12）； wherein represents the wavelength represents the total blood absorbance represents the multiplication factor of the optical path length due to light scattering in the blood represents the blood thickness; wherein the calibration factor is used to compensate for the loss of optical signal due to the thickness of the extracorporeal blood circuit and scattering in the blood represents the wavelength represents the molar extinction coefficient of the light-absorbing substance in the blood represents the molar concentration of the light-absorbing substance in the blood S22, calculating the preliminary prediction value of absorbance hematocrit and blood oxygen saturation at each wavelength: （13）； （14）； wherein and respectively represent the preliminary predicted values of hematocrit and blood oxygen saturation; is the red blood cell concentration; represents the plasma concentration; and respectively represent the concentrations of reduced hemoglobin and oxygenated hemoglobin at the wavelength below; S23, the absorbance ratio at each wavelength is obtained by simultaneously solving formula (15) and the modified Lambert-Beer law Relationship: （15）； wherein indicates the wavelength total blood absorbance; Further simplify formula (13) and formula (14) to obtain: （16）； （17）； wherein , , , , , , , , , , and are constants; , , , respectively represent the absorbance ratio at 660 nm, 808 nm, 940 nm, 1310 nm wavelengths.

4. The RKHS-PLS based optical detection and calibration method of blood parameters according to claim 3, characterized in that: Step S3 specifically comprises the following steps: S31, the preliminary predicted values of the integrated transmitted light intensity, the absorbance of each wavelength, hematocrit and blood oxygen saturation are obtained as independent variables , the true values of hematocrit and blood oxygen saturation measured by the blood cell counter and the blood gas analyzer are obtained as dependent variables ; S32, standardization processing: （18）； wherein, and denote the standardized independent variable and dependent variable, respectively; and denote the mean deviation of the independent variable and dependent variable, respectively; and denote the standard deviation of the independent variable and dependent variable, respectively; S33, using the following regression algorithm to establish the mapping relationship between the standardized independent variable and the dependent variable: （19）； In the formula, denotes a regression coefficient matrix; denotes a residual matrix.

5. The RKHS-PLS based optical detection and calibration method of blood parameters according to claim 4, characterized in that: In step S4, the trained partial least squares regression calibration model is evaluated with the indicators of the predicted root mean square error , the mean absolute error , the mean absolute percentage error and the coefficient of determination .

6. The RKHS-PLS based optical detection and calibration method of blood parameters according to claim 4, characterized in that: In step S5, final prediction values for hematocrit and blood oxygen saturation are calculated using reverse normalization : （20）。

Images