System and method for detecting urinary albumin based on photonic crystal microarray chip

Abstract

The invention relates to the technical field of physical analysis for processing biological materials, in particular to a urinary albumin detection system and method based on a photonic crystal microarray chip, and the system comprises a spectrum fingerprint calibration module, a signal real-time deconstruction module, a dynamic feature extraction module and a protein concentration mapping module. According to the method, the steady-state reflection spectrum of the photonic crystal microarray is accurately calibrated and is endowed with a digital identity, fine-grained tracking of wavelength change is realized in detection, and then the maximum binding rate and the stable response amplitude are extracted in dynamic binding reaction, so that the deviation of single end-point detection is avoided through the acquisition of the dual characteristics, and the detection accuracy is improved. The result can reflect the binding rate and the concentration characteristic of the stable stage at the same time, on the basis, the accuracy and stability of urinary albumin concentration recognition are effectively improved in combination with a data-driven mapping analysis mode, the comprehensive analysis mode fuses dynamic process analysis and steady state characteristic extraction, and more reliable quantitative result output is achieved.

Description

Technical Field

[0001] This invention relates to the field of physical analysis technology for processing biological materials, and in particular to a urinary albumin detection system and method based on a photonic crystal microarray chip. Background Technology

[0002] The field of physical analysis technology for processing biomaterials involves the determination and analysis of the physicochemical properties of biological samples such as blood, urine, saliva, and tissue fluid. Core aspects include the detection of sample component concentrations, molecular structure analysis, and the identification of specific proteins or metabolites. This technology typically relies on optical analysis, electrochemical detection, and microfluidics to construct a complete detection system from sample pretreatment to quantitative analysis, providing fundamental support for clinical medicine, biological diagnostics, and environmental health monitoring. Traditional urine albumin detection systems refer to devices and methods for the quantitative or qualitative determination of albumin components in urine. These systems typically achieve detection through immunoturbidimetry, radioimmunoassay, or enzyme-linked immunosorbent assay (ELISA), utilizing specific antigen-antibody reactions and colorimetric determination methods to analyze albumin content. This patent, however, focuses on a urine albumin detection system based on a photonic crystal microarray chip.

[0003] Traditional urine albumin detection relies on immunoturbidimetry, radioimmunoassay, or enzyme-linked immunosorbent assay (ELISA). These methods typically rely on endpoint colorimetric signals, making it difficult to capture the dynamic characteristics of the antigen-antibody binding process. When there is complex matrix interference or signal fluctuations in the sample, the detection results are prone to insufficient sensitivity or decreased specificity. This is especially true under low concentration conditions, which can easily lead to missed detections or misjudgments. Furthermore, because they depend on chemical reactions and colorimetric changes, the detection time is often limited by the reaction rate, making it difficult to achieve rapid and high-precision results. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and to propose a urinary albumin detection system and method based on a photonic crystal microarray chip.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The urine albumin detection system based on a photonic crystal microarray chip includes: a spectral fingerprint calibration module, which calculates the center wavelength of the steady-state mixed reflection spectrum of the photonic crystal microarray, assigns a digital identifier to the center wavelength and constructs a reference wavelength identifier library, and transmits the reference wavelength identifier library to the real-time signal deconstruction module; The real-time signal deconstruction module calculates the wavelength shift of each digital identifier in the mixed reflectance spectrum sequence based on the reference wavelength identifier library, generates an instantaneous wavelength shift sequence, and transmits the instantaneous wavelength shift sequence to the dynamic feature extraction module. The kinetic feature extraction module calculates the instantaneous binding rate based on the instantaneous wavelength drift sequence to screen out the maximum binding rate, and calculates the average instantaneous wavelength drift during the stable period as the saturation response amplitude, constructs a kinetic feature set and passes it to the protein concentration mapping module. The protein concentration mapping module, based on the maximum binding rate and saturation response amplitude in the kinetic feature set, calls a support vector machine model to perform regression analysis and calculate the urinary albumin concentration value.

[0006] As a further embodiment of the present invention, the reference wavelength identification library includes a partition index, an initial wavelength calibration value, and an identity code; the instantaneous wavelength drift sequence specifically includes a timestamp sequence, a drift amount data stream, and a partition association identifier; the kinetic feature set includes the maximum binding rate, the saturation response amplitude, and the maximum dissociation rate; and the urinary albumin concentration value is a quantitative detection result, a confidence interval, and a sample number.

[0007] As a further embodiment of the present invention, the spectral fingerprint calibration module includes: a spectral preprocessing submodule, which acquires the steady-state mixed reflection spectrum of the photonic crystal microarray, performs curve fitting on the steady-state mixed reflection spectrum using a Gaussian fitting algorithm to obtain the peak position, and uses the peak position as the center wavelength to generate initial center wavelength data; The identifier library construction submodule assigns a unique digital identifier to each center wavelength based on the initial center wavelength data, establishes a mapping relationship between the center wavelength and the digital identifier, and stores the mapping relationship in memory to construct the reference wavelength identifier library.

[0008] As a further embodiment of the present invention, the real-time signal deconstruction module includes: a real-time spectral acquisition submodule, which monitors and acquires the mixed reflectance spectral sequence during the reaction process in real time, timestamps each frame of spectral data in the mixed reflectance spectral sequence, and generates a time-stamped spectral sequence. The wavelength drift calculation submodule calls the reference wavelength identifier library, finds the current center wavelength corresponding to each digital identifier for each frame of the time-stamped spectral sequence, calculates the difference between the current center wavelength and the initial wavelength calibration value in the reference wavelength identifier library, uses the difference as the wavelength drift amount, integrates the wavelength drift amounts at all time points, and generates the instantaneous wavelength drift sequence.

[0009] As a further embodiment of the present invention, the dynamic feature extraction module includes: a binding rate calculation submodule, which calculates the rate of change of the wavelength drift between adjacent timestamps based on the instantaneous wavelength drift sequence, uses the rate of change as the instantaneous binding rate, and iterates through and filters all the calculated instantaneous binding rates to obtain the maximum binding rate; The saturation response extraction submodule analyzes the overall trend of the instantaneous wavelength drift sequence, determines that when the rate of change of the instantaneous binding rate is less than a preset rate of change threshold and the duration exceeds a preset stable time threshold, it enters the stable period, extracts all instantaneous wavelength drift data within the stable period, calculates the arithmetic mean of the instantaneous wavelength drift data, uses this mean as the saturation response amplitude, and constructs the dynamic feature set. The preset rate of change threshold is an upper limit of rate fluctuation determined based on the baseline noise statistical distribution during the reaction process of multiple blank samples; The preset stabilization time threshold is determined based on the shortest duration required for the instantaneous binding rate of the standard sample to remain below the upper limit of the rate fluctuation when it reaches the saturated reaction state.

[0010] As a further embodiment of the present invention, the protein concentration mapping module is specifically used to obtain the kinetic feature set, take the maximum binding rate and the saturation response amplitude in the kinetic feature set as input feature vectors, and call the pre-trained support vector machine model to perform regression calculation; Among them, the urine albumin concentration value The calculation formula is: ; in, The final calculated urinary albumin concentration value, For summation, For the index of the support vectors, The number of support vectors, and The Lagrange multipliers obtained from model training, For the training sample, the first Feature vectors of support vectors The input feature vector of the sample to be tested. For kernel function, These are the bias terms of the support vector machine model; Based on the regression calculation results, the urinary albumin concentration value is output.

[0011] As a further aspect of the present invention, the instantaneous binding rate is calculated using the following formula: ; in, Indicates in The instantaneous binding rate at time t, For time, This is the index of the current sampling point. for The instantaneous wavelength shift at time t. For the previous moment The instantaneous wavelength shift, The time interval between two adjacent sampling points. To and The time-weighted factor relating the signal-to-noise ratio of the signal at any given moment; The time weighting factor The time weighting factor is negatively correlated with the signal-to-noise ratio (SNR). Specifically, in the initial stage of signal acquisition, when the SNR is low, a larger time weighting factor is used to enhance the smoothing effect. In the stable stage of signal acquisition, when the SNR is high, a smaller time weighting factor is used to preserve signal details.

[0012] As a further aspect of the present invention, the support vector machine model adopts a radial basis kernel function, and uses a method combining a grid search strategy based on Bayesian optimization and K-fold cross-validation to adaptively optimize the model hyperparameters; Specifically, within the preset penalty coefficient With kernel function parameters Within the combination space, all parameter combinations are systematically traversed, and K-fold cross-validation is used to evaluate the model performance under each combination. The penalty coefficient with the minimum average regression error is selected. With kernel function parameters As the final model parameters.

[0013] As a further aspect of the present invention, the dynamic feature extraction module performs filtering and peak detection operations on the instantaneous binding rate before screening the maximum binding rate; Specifically, the sequence of instantaneous binding rates is first smoothed using a Savitzky-Golay filter to generate a smoothed binding rate curve. An extreme value search algorithm is applied to the smooth binding rate curve to identify all local maxima points, and the rate values ​​corresponding to all the local maxima points are compared. The largest rate value is determined as the maximum binding rate.

[0014] The method for detecting urinary albumin based on photonic crystal microarray chip, the method being executed based on the aforementioned urinary albumin detection system based on photonic crystal microarray chip, includes the following steps: S1: spectral fingerprint calibration step, calculating the center wavelength of the steady-state mixed reflection spectrum of the photonic crystal microarray, assigning a digital identifier to the center wavelength and constructing a reference wavelength identifier library; S2: Real-time signal deconstruction step: Based on the reference wavelength identifier library, calculate the wavelength shift of each digital identifier in the mixed reflectance spectrum sequence and generate an instantaneous wavelength shift sequence; S3: Dynamic feature extraction step, based on the instantaneous wavelength drift sequence, calculate the instantaneous binding rate to screen out the maximum binding rate, and calculate the average instantaneous wavelength drift during the stable period as the saturation response amplitude to construct a dynamic feature set; S4: Protein concentration mapping step, based on the maximum binding rate and saturation response amplitude in the kinetic feature set, calls the support vector machine model to perform regression analysis and calculate the urinary albumin concentration value.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by precisely calibrating and assigning a digital identity to the steady-state reflectance spectrum of a photonic crystal microarray, fine-grained tracking of wavelength changes can be achieved during the detection process. This allows for the extraction of the maximum binding rate and the stable response amplitude in the dynamic binding reaction. This dual-feature acquisition not only avoids the bias caused by single-endpoint detection but also enables the detection results to simultaneously reflect the rate characteristics of the binding process and the concentration characteristics of the steady-state phase. Based on this, combined with a data-driven mapping analysis method, the accuracy and stability of urinary albumin concentration identification can be effectively improved. This transforms the detection from a traditional static measurement method that relies on colorimetric signal intensity to a comprehensive analysis mode that integrates dynamic process analysis and steady-state feature extraction, achieving more reliable quantitative results. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the overall workflow of the urine albumin detection system based on a photonic crystal microarray chip according to the present invention. Figure 2 This is a flowchart illustrating the workflow of the spectral fingerprint calibration module of the present invention. Figure 3 This is a flowchart illustrating the real-time signal deconstruction module of the present invention. Figure 4 This is a flowchart of the dynamic feature extraction module of the present invention; Figure 5 This is a flowchart of the protein concentration mapping module of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the software-based technical solution is described in detail below with reference to system architecture diagrams and embodiments. It should be understood that the specific embodiments described herein are only for explaining the technical solutions of this invention and do not constitute a limitation on the scope of protection.

[0018] In the description of this invention, the system architecture relationships or data processing flows indicated by terms such as "layer," "module," "interface," "data flow," "client," and "server" are all defined based on the architecture diagram or flowchart corresponding to the embodiments. This way of describing is only used to clearly illustrate the logical relationships between the elements in the technical solution, and not to limit the physical deployment form. The term "multiple" includes two or more technical units, including but not limited to multiple data nodes, processing threads, service instances, or functional components and other scalable elements. The specific number is determined according to the actual business scenario and needs to be specifically specified.

[0019] Please see Figure 1 and Figure 2 This invention provides a technical solution: a urinary albumin detection system based on a photonic crystal microarray chip includes: The spectral fingerprinting module calculates the center wavelength of the steady-state mixed reflection spectrum of the photonic crystal microarray, assigns a digital identifier to the center wavelength, and constructs a reference wavelength identifier library, which is then transmitted to the real-time signal deconstruction module. The spectral fingerprinting module includes: a spectral preprocessing submodule, which acquires the steady-state mixed reflection spectrum of the photonic crystal microarray, uses a Gaussian fitting algorithm to perform curve fitting on the steady-state mixed reflection spectrum to obtain the peak position, and uses the peak position as the center wavelength to generate initial center wavelength data; The identifier library construction submodule assigns a unique digital identifier to each center wavelength based on the initial center wavelength data, establishes a mapping relationship between the center wavelength and the digital identifier, and stores the mapping relationship in memory to construct the reference wavelength identifier library; The reference wavelength identifier library includes a partition index, initial wavelength calibration values, and identification codes.

[0020] Before performing spectral fingerprinting, this invention first prepares a photonic crystal microarray chip. This chip is divided into an 8x12 matrix, comprising 96 independent sensing zones. Each zone is pre-fixed with a monoclonal antibody that specifically recognizes human urinary albumin via chemical bonding. The prepared chip is placed in the sample chamber of the detection system, and a phosphate-buffered saline (PBS) solution is introduced onto the chip surface via a microfluidic system at a flow rate of 20 μL / min for 5 minutes, until the chip surface and the buffer solution have fully interacted and reached wetting equilibrium. Afterward, the liquid flow is stopped, and the spectral acquisition unit, consisting of a broadband light source, a fiber optic probe, and a spectrometer, is activated. Light emitted from the light source is incident perpendicularly onto the surface of the specific sensing zone of the microarray chip via the fiber optic probe. The reflected light is then collected by the same probe and transmitted to the spectrometer. The integration time of the spectrometer is set to 100 ms, the average number of scans is 10, and the acquisition wavelength range is 450 nm to 750 nm. The system automatically scans all 96 sensing partitions sequentially, acquiring steady-state mixed reflectance spectral data for each partition in a PBS buffer environment. Taking partition A1 as an example, the raw spectral data it acquires consists of a series of light intensity versus wavelength values, presented as a peak-shaped curve centered on a specific wavelength.

[0021] For the steady-state mixed reflectance spectrum obtained from partition A1, the spectral preprocessing submodule is initiated. This submodule retrieves the spectral data and performs a Gaussian fitting algorithm on it. The algorithm's execution process is as follows: First, an initial peak position estimate is determined in the spectral data, for example, by finding the wavelength corresponding to the maximum light intensity, denoted as . Subsequently, with A fitting window with a width of 30 nanometers is defined centered on the target. Within this window, the three core parameters of the Gaussian function are iteratively optimized using a nonlinear least squares method: peak height... Central position and standard deviation The goal of the iterative process is to minimize the difference between the actual spectral data points and the Gaussian function. Calculate the sum of squared residuals between the values. When the change in the sum of squared residuals between two consecutive iterations is less than a preset convergence threshold... The iteration stops when the time is reached. The parameters obtained at this point are... This is determined to be the center wavelength of the sensor partition. For example, the original spectral data for partition A1 has the highest light intensity near a wavelength of 551 nm. After the Gaussian fitting calculation described above, its precise center wavelength is finally obtained as 551.23 nm. This process is applied to all 96 sensing partitions to generate an initial center wavelength dataset containing 96 center wavelength values, in the format [551.23, 552.01, 550.98, ...]. This dataset constitutes the initial center wavelength data.

[0022] Next, the identifier library construction submodule is activated, which builds a reference wavelength identifier library based on the initial center wavelength data generated in the previous step. The process is as follows: the system sequentially assigns unique digital identifiers to the 96 sensing partitions, with the encoding rule starting from 1001 and incrementing. Partition A1 is assigned identifier 1001, partition A2 is assigned 1002, and so on, until the last partition H12 is assigned identifier 1096. Subsequently, a mapping relationship is established between the center wavelength, the partition physical index, and the digital identifier. For example, the physical index of partition A1, its center wavelength of 551.23 nm, and its digital identifier 1001 are associated together. These 96 sets of mapping relationships are stored in the system's non-volatile memory in the form of a data table, completing the construction of the reference wavelength identifier library. The structured data of this library is stored in a database table.

[0023] Table 1. Example of a reference wavelength identifier library Partitioned Index Initial wavelength calibration value (nm) Identity Code A1 551.23 1001 A2 552.01 1002 A3 550.98 1003 ... ... ... H12 551.55 1096 As shown in Table 1, the reference wavelength identifier library contains the physical location of each sensing partition (partition index), the precise center wavelength of the reflectance spectrum in a pure buffer environment (initial wavelength calibration value), and a digital tag (identification code) used to uniquely identify the partition in subsequent data processing. This reference wavelength identifier library is then passed to the real-time signal deconstruction module.

[0024] Please see Figure 1 and Figure 3 The real-time signal deconstruction module, based on the reference wavelength identifier library, calculates the wavelength shift corresponding to each digital identifier in the mixed reflectance spectrum sequence, generates an instantaneous wavelength drift sequence, and transmits the instantaneous wavelength drift sequence to the dynamic feature extraction module. The real-time signal deconstruction module includes: a real-time spectral acquisition submodule, which monitors and acquires the mixed reflectance spectral sequence during the reaction process in real time, timestamps each frame of spectral data in the mixed reflectance spectral sequence, and generates a time-stamped spectral sequence; The wavelength drift calculation submodule calls the reference wavelength identifier library, finds the current center wavelength corresponding to each digital identifier for each frame of the time-stamped spectral sequence, calculates the difference between the current center wavelength and the initial wavelength calibration value in the reference wavelength identifier library, uses this difference as the wavelength drift amount, integrates the wavelength drift amounts of all time points, and generates an instantaneous wavelength drift sequence. The instantaneous wavelength drift sequence specifically includes a timestamp sequence, a drift amount data stream, and a partition association identifier.

[0025] The module first operates through a real-time spectral acquisition submodule. After constructing the reference wavelength identifier library, the urine sample to be tested is injected into the photonic crystal microarray chip via a microfluidic system. The urine sample flows continuously across the chip surface at a rate of 15 μL / min to simulate the dynamic process of biomolecule binding. During this period, the real-time spectral acquisition submodule continuously triggers the spectral acquisition unit at fixed time intervals. The time interval is set to 0.5 seconds, i.e., the acquisition frequency is 2 Hz. With each trigger, the spectral acquisition unit quickly scans all 96 sensing partitions, acquiring the mixed reflectance spectrum of each partition at the current moment. Simultaneously, the system binds the current timestamp (accurate to milliseconds) provided by a high-precision time server with all 96 spectral data points acquired in that frame. For example, at 10.000 seconds after the reaction begins, the system completes a full array scan, generating 96 spectral data points, and this entire set of data is marked with a timestamp of "10.000s". This process is repeated at subsequent time points such as "10.500s", "11.000s", etc. This generates a time-stamped spectral sequence, whose data structure is a time series, with each time point associated with a set containing 96 spectral data arrays.

[0026] Subsequently, the wavelength drift calculation submodule is activated to process the time-stamped spectral sequence. This submodule retrieves the previously constructed reference wavelength identifier library. For each frame of data in the time-stamped spectral sequence, such as the frame "10.000s", the submodule iterates through its 96 spectra. Taking the spectrum of partition A1 (identity code 1001) as an example, the submodule uses the same spectral preprocessing procedure as the spectral fingerprint calibration module to calculate the center wavelength of partition A1 at that moment. Assuming that at "10.000s", the calculated center wavelength of partition A1 is 551.48 nm. Next, the submodule searches for the entry with identity code 1001 in the reference wavelength identifier library and reads its corresponding initial wavelength calibration value, which is 551.23 nm. Then, it calculates the difference between these two wavelength values ​​to obtain the wavelength drift amount: This calculation process is performed one by one on all 96 partitions within the frame of "10.000s". At the next time point "10.500s", the center wavelength of partition A1 may change to 551.52 nanometers, and its wavelength shift is recalculated as follows: By integrating the wavelength drift values ​​of all partitions at all time points, a transient wavelength drift sequence is generated. The data structure of this sequence is a series of records, each containing three fields: timestamp sequence, drift data stream, and partition association identifier. Specifically, an example of a record is: {timestamp: 10.500s, drift data stream: [0.29, 0.31, 0.28, ...], partition association identifier: [1001, 1002, 1003, ...]}. This transient wavelength drift sequence is then passed to the dynamic feature extraction module.

[0027] Please see Figure 1 and Figure 4 The kinetic feature extraction module calculates the instantaneous binding rate based on the instantaneous wavelength drift sequence to screen out the maximum binding rate, and calculates the average instantaneous wavelength drift during the stable period as the saturation response amplitude, constructs the kinetic feature set and passes it to the protein concentration mapping module. Before filtering for the maximum binding rate, the dynamic feature extraction module also performs filtering and peak detection operations on the instantaneous binding rate; Specifically, the Savitzky-Golay filter is first used to smooth the sequence of instantaneous binding rates to generate a smooth binding rate curve. An extreme value search algorithm is applied to the smooth binding rate curve to identify all local maxima points, and the rate values ​​corresponding to all local maxima points are compared. The maximum rate value is determined as the maximum binding rate. The dynamic feature extraction module includes: a binding rate calculation submodule, which calculates the rate of change of wavelength drift between adjacent time stamps based on the instantaneous wavelength drift sequence, uses this rate of change as the instantaneous binding rate, and iterates through all the calculated instantaneous binding rates to obtain the maximum binding rate; The instantaneous binding rate is calculated using the following formula: ; in, Indicates in The instantaneous binding rate at any given moment For time, This is the index of the current sampling point. for The instantaneous wavelength shift at time t. For the previous moment The instantaneous wavelength shift, The time interval between two adjacent sampling points. To and The time-weighted factor relating the signal-to-noise ratio of the signal at any given moment; Time weighting factor It is negatively correlated with the signal-to-noise ratio. Specifically, in the initial stage of signal acquisition, the signal-to-noise ratio is low, so a larger time weighting factor is used to enhance the smoothing effect. In the stable stage of signal acquisition, the signal-to-noise ratio is high, so a smaller time weighting factor is used to preserve signal details. The saturation response extraction submodule analyzes the overall trend of the instantaneous wavelength drift sequence, determines when the rate of change of the instantaneous binding rate is less than a preset rate of change threshold and the duration exceeds a preset stable time threshold, enters a stable period, extracts all instantaneous wavelength drift data within the stable period, calculates the arithmetic mean of the instantaneous wavelength drift data, uses this mean as the saturation response amplitude, and constructs a dynamic feature set. The preset rate of change threshold is the upper limit of rate fluctuation determined based on the baseline noise statistical distribution during the reaction process of multiple blank samples; The preset stabilization time threshold is determined based on the shortest duration required for the instantaneous binding rate of the standard to remain below the upper limit of rate fluctuation when it reaches the saturated reaction state. The kinetic feature set includes the maximum binding rate and the saturation response amplitude.

[0028] This module first processes the received instantaneous wavelength drift sequence using the binding rate calculation submodule. The submodule calculates the instantaneous binding rate using the formula: Calculate the instantaneous binding rate at each time point. In the formula, Representative at The instantaneous binding rate at any given moment; It is a time variable; It is the index of the current sampling point; It was obtained from the instantaneous wavelength drift sequence. The instantaneous wavelength shift at a given moment; It was the previous moment The instantaneous wavelength shift; It is the time interval between two adjacent sampling points, which is determined by the sampling frequency of the real-time spectrum acquisition submodule; Is with The time-weighted factor is used to correlate the signal-to-noise ratio of the signal at any given time. The logic behind this formula is to quantify the rate of biomolecule binding by calculating the change in wavelength shift per unit time, and then introduce a time-weighted factor. The rate calculation results under different signal-to-noise ratios are dynamically adjusted.

[0029] The specific process of obtaining and calculating the parameters is as follows: Taking the data of the aforementioned partition A1 (identity code 1001) as an example, The time is 10.500s. If the time is 10.000s, then the time interval is Seconds. Found from the instantaneous wavelength drift sequence. nm, nm. Time weighting factor The determination of SNR depends on the signal-to-noise ratio (SNR). SNR is calculated by taking the peak intensity of the Gaussian fitted peak in the current spectral data. Divide by the standard deviation of light intensity in the spectral baseline region (e.g., the 700-750 nm range). Assuming in Time, measured (Any unit) ,but . The relationship with SNR was calibrated experimentally. In the initial stage of signal acquisition (e.g., the first 5 seconds), the reactant concentration is low, and the SNR is below 20; this range is defined as the low signal-to-noise ratio (SNR) stage. During the stable reaction stage, the SNR is greater than or equal to 20; this range is defined as the high SNR stage. The weighting factor is set according to the rule: when SNR < 20, When SNR At 8 PM, According to this rule, in At that time, the SNR was 16, therefore Substitute all parameter values ​​into the formula: nm / s. This calculation is repeated at all time points to generate an instantaneous binding rate sequence.

[0030] After obtaining the instantaneous binding rate sequence, filtering and peak detection are performed. First, the sequence is smoothed using a Savitzky-Golay filter. This process is not directly named but described as follows: For each point in the rate sequence, a calculation window is formed by selecting itself and two points before and after it, for a total of five data points. A least-squares fit is performed on these five points using a quadratic polynomial, and the calculated value of the fitted polynomial at the center point is used as the new, smoothed rate value. This window slides along the entire rate sequence, generating the smoothed binding rate curve. Next, an extremum search is applied to the smoothed curve. This process involves iterating through each point on the smoothed curve; if the rate value at a point is strictly greater than the rate values ​​of its left and right adjacent points, then that point is marked as a local maximum. After the iteration, the rate values ​​corresponding to all marked local maxima are compared, and the largest value is determined as the maximum binding rate. For example, the maximum binding rate of partition A1 is calculated to be 0.15 nm / s.

[0031] Subsequently, the saturation response extraction submodule is initiated. This submodule first determines two thresholds. The process for determining the preset rate of change threshold is as follows: 10 groups of blank samples without urinary albumin are prepared, reacted under conditions exactly the same as the actual test, and their respective instantaneous binding rate sequences are calculated. The fluctuation of all data points in these 10 sequences is statistically analyzed, and their standard deviation is calculated to obtain a value. nm / s. The preset rate of change threshold is set to 3 times this standard deviation, i.e. nm / The process for determining the preset stabilization time threshold: Five standard solutions containing different concentrations (10, 20, 50, 100, 150 mg / L) of urinary albumin were prepared, and the instantaneous binding rate was calculated after the reaction. The rate curve was observed to reach saturation (i.e., the rate value tended to stabilize), and the rate of change of the instantaneous binding rate was consistently lower than [a certain value]. nm / The shortest time. Measurements showed the shortest duration was 20 seconds, therefore the preset stabilization time threshold was set to 20 seconds. When processing the instantaneous wavelength drift sequence of the sample under test, the system calculates the rate of change of the instantaneous binding rate in real time. When this rate of change is less than... nm / If this state persists for more than 20 seconds, the system determines that the reaction has entered a steady-state period. The system extracts all instantaneous wavelength drift data from this point in time to the end of the reaction and calculates the arithmetic mean of these data. For example, if 100 instantaneous wavelength drift data points are collected during the steady-state period, and their sum is 85 nm, then the saturated response amplitude is... nm. Furthermore, after the reaction, the rate of negative wavelength shift was calculated using the same method by injecting elution buffer, yielding the maximum dissociation rate, for example, -0.05 nm / s. Finally, the constructed kinetic feature set is: {maximum binding rate: 0.15 nm / s, saturation response amplitude: 0.85 nm, maximum dissociation rate: -0.05 nm / s}. This feature set is then passed to the protein concentration mapping module.

[0032] Please see Figure 1 and Figure 5 The protein concentration mapping module, based on the maximum binding rate and saturation response amplitude in the kinetic feature set, calls the support vector machine model to perform regression analysis and calculate the urinary albumin concentration value. The protein concentration mapping module is specifically used to obtain a kinetic feature set, take the maximum binding rate and saturation response amplitude in the kinetic feature set as input feature vectors, and call a pre-trained support vector machine model to perform regression calculations. Among them, the urinary albumin concentration value The calculation formula is ; in, This is the final calculated urinary albumin concentration value. For summation, For the index of the support vectors, The number of support vectors, and The Lagrange multipliers obtained from model training, For the training sample, the first Feature vectors of support vectors The input feature vector of the sample to be tested. For kernel function, These are the bias terms of the support vector machine model; Based on the regression calculation results, the urinary albumin concentration value is output; The support vector machine model uses a radial basis kernel function and adopts a method that combines a grid search strategy based on Bayesian optimization with K-fold cross-validation to adaptively optimize the model's hyperparameters. Specifically, within the preset penalty coefficient With kernel function parameters Within the combination space, all parameter combinations are systematically traversed, and K-fold cross-validation is used to evaluate the model performance under each combination. The penalty coefficient with the minimum average regression error is selected. With kernel function parameters As the final model parameters; The urine albumin concentration value is a quantitative test result, confidence interval, and sample number.

[0033] After receiving the dynamic feature set, this module extracts the maximum binding rate and saturation response amplitude, and combines them into an input feature vector. Taking the data generated in the aforementioned embodiment as an example, this input feature vector... The value is [0.15, 0.85]. Subsequently, the module calls a pre-trained Support Vector Machine (SVM) model to perform regression calculations on this feature vector. (Urinary albumin concentration value) The calculation formula is: This formula is a decision function based on support vector regression. Wherein, This is the predicted value of the urinary albumin concentration to be calculated; Indicates all Summation operation is performed on each support vector; and These are the Lagrange multipliers that the model learns from the training data, and they reflect the importance of each support vector in constructing the decision function; It is the first in the training samples Each support vector has a data structure similar to the input feature vector. same; It is the bias term of the model, used to adjust the overall bias of the decision function; This is a kernel function used to map input data to a high-dimensional feature space. Here, the radial basis function (RBF) is used, and its specific form is: ,in It is an adjustable parameter of the kernel function. Representing vectors and The square of the Euclidean distance between them.

[0034] Hyperparameters (penalty coefficients) of support vector machine models With kernel function parameters During the model training phase, adaptive optimization was achieved using a combination of a Bayesian optimization-based grid search strategy and 5-fold cross-validation. The optimization process involved: first, setting a penalty coefficient... The search scope is Kernel function parameters The search scope is Experimental data containing 200 urinary albumin samples of known concentrations were used as the training set. In 5-fold cross-validation, the training set was randomly divided into 5 subsets. In each iteration, 4 subsets were selected for training the model, and the remaining subset was used for validation, with the root mean square error (RMSE) calculated. This process was repeated 5 times, ensuring that each subset was used as the validation set once. The final performance evaluation metric was the average of the RMSE from the 5 validation iterations. The Bayesian optimization algorithm was then used based on the evaluated data. To assess the performance of parameter pairs, a probabilistic surrogate model is built to predict the potential performance of other parameter pairs. Based on this prediction, the next optimal point is selected for evaluation, rather than blindly iterating. After optimization, the set of parameters with the minimum average regression error is chosen, for example... and As the final model parameters.

[0035] Table 2. Example of data for model hyperparameter optimization process. Penalty coefficient C Kernel function parameter γ Mean regression error (mg / L) 10 0.1 5.32 50 0.1 4.88 50 0.8 2.15 100 0.8 2.21 50 5.0 3.76 As shown in Table 2, the parameter combinations are as follows: It was selected because it achieved the minimum mean regression error.

[0036] When calculating the actual sample concentration, assume the pre-trained model has the following parameters: number of support vectors. ; bias term The three support vectors and their corresponding Lagrange multiplier differences are as follows: ; ; ; ; , ; Input feature vector of the sample to be tested First, calculate separately. With each support vector Kernel function value: ; ; ; ; ; Then, substitute these values ​​into the SVM formula: ; The calculation result mg / L is the quantitative detection result of urinary albumin concentration by the system. Based on the historical performance data of the model, a 95% confidence interval is calculated, for example, [48.2, 52.5] mg / L. The final output result is: {Sample ID: UrineSample_789, Quantitative Detection Result: 50.4 mg / L, Confidence Interval: [48.2, 52.5] mg / L}.

[0037] The method for detecting urinary albumin based on a photonic crystal microarray chip is performed using the aforementioned urinary albumin detection system based on a photonic crystal microarray chip, and includes the following steps: S1: Spectral fingerprinting step, calculate the center wavelength of the steady-state mixed reflection spectrum of the photonic crystal microarray, assign a digital identifier to the center wavelength and build a reference wavelength identifier library; S2: Real-time signal deconstruction step, based on the reference wavelength identifier library, calculates the wavelength drift of each digital identifier in the mixed reflectance spectrum sequence, and generates an instantaneous wavelength drift sequence; S3: Dynamic feature extraction step. Based on the instantaneous wavelength drift sequence, the instantaneous binding rate is calculated to screen out the maximum binding rate, and the average instantaneous wavelength drift during the stable period is calculated as the saturation response amplitude to construct a dynamic feature set. S4: Protein concentration mapping step, based on the maximum binding rate and saturation response amplitude in the kinetic feature set, calls the support vector machine model to perform regression analysis and calculate the urinary albumin concentration value.

[0038] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the scope of protection defined by the claims of the present invention.

Claims

1. A urinary albumin detection system based on a photonic crystal microarray chip, characterized in that, The system includes: a spectral fingerprint calibration module, which calculates the center wavelength of the steady-state hybrid reflection spectrum of the photonic crystal microarray, assigns a digital identifier to the center wavelength and constructs a reference wavelength identifier library, and transmits the reference wavelength identifier library to the real-time signal deconstruction module; The real-time signal deconstruction module calculates the wavelength shift of each digital identifier in the mixed reflectance spectrum sequence based on the reference wavelength identifier library, generates an instantaneous wavelength shift sequence, and transmits the instantaneous wavelength shift sequence to the dynamic feature extraction module. The kinetic feature extraction module calculates the instantaneous binding rate based on the instantaneous wavelength drift sequence to screen out the maximum binding rate, and calculates the average instantaneous wavelength drift during the stable period as the saturation response amplitude, constructs a kinetic feature set and passes it to the protein concentration mapping module. The protein concentration mapping module, based on the maximum binding rate and saturation response amplitude in the kinetic feature set, calls a support vector machine model to perform regression analysis and calculate the urinary albumin concentration value.

2. The urinary albumin detection system based on a photonic crystal microarray chip according to claim 1, characterized in that, The reference wavelength identifier library includes a partition index, an initial wavelength calibration value, and an identity code. The instantaneous wavelength drift sequence specifically includes a timestamp sequence, a drift amount data stream, and a partition association identifier. The kinetic feature set includes the maximum binding rate, the saturation response amplitude, and the maximum dissociation rate. The urinary albumin concentration value is a quantitative detection result, a confidence interval, and a sample number.

3. The urinary albumin detection system based on a photonic crystal microarray chip according to claim 1, characterized in that, The spectral fingerprinting module includes: a spectral preprocessing submodule, which acquires the steady-state mixed reflection spectrum of the photonic crystal microarray, performs curve fitting on the steady-state mixed reflection spectrum using a Gaussian fitting algorithm to obtain the peak position, and uses the peak position as the center wavelength to generate initial center wavelength data; The identifier library construction submodule assigns a unique digital identifier to each center wavelength based on the initial center wavelength data, establishes a mapping relationship between the center wavelength and the digital identifier, and stores the mapping relationship in memory to construct the reference wavelength identifier library.

4. The urinary albumin detection system based on a photonic crystal microarray chip according to claim 1, characterized in that, The real-time signal deconstruction module includes: a real-time spectral acquisition submodule, which monitors and acquires the mixed reflectance spectral sequence during the reaction process in real time, timestamps each frame of spectral data in the mixed reflectance spectral sequence, and generates a time-stamped spectral sequence; The wavelength drift calculation submodule calls the reference wavelength identifier library, finds the current center wavelength corresponding to each digital identifier for each frame of the time-stamped spectral sequence, calculates the difference between the current center wavelength and the initial wavelength calibration value in the reference wavelength identifier library, uses the difference as the wavelength drift amount, integrates the wavelength drift amounts at all time points, and generates the instantaneous wavelength drift sequence.

5. The urinary albumin detection system based on a photonic crystal microarray chip according to claim 1, characterized in that, The dynamic feature extraction module includes: a binding rate calculation submodule, which calculates the rate of change of the wavelength drift between adjacent timestamps based on the instantaneous wavelength drift sequence, uses the rate of change as the instantaneous binding rate, and iterates through and filters all the calculated instantaneous binding rates to obtain the maximum binding rate; The saturation response extraction submodule analyzes the overall trend of the instantaneous wavelength drift sequence, determines that when the rate of change of the instantaneous binding rate is less than a preset rate of change threshold and the duration exceeds a preset stable time threshold, it enters the stable period, extracts all instantaneous wavelength drift data within the stable period, calculates the arithmetic mean of the instantaneous wavelength drift data, uses this mean as the saturation response amplitude, and constructs the dynamic feature set. The preset rate of change threshold is an upper limit of rate fluctuation determined based on the baseline noise statistical distribution during the reaction process of multiple blank samples; The preset stabilization time threshold is determined based on the shortest duration required for the instantaneous binding rate of the standard sample to remain below the upper limit of the rate fluctuation when it reaches the saturated reaction state.

6. The urinary albumin detection system based on a photonic crystal microarray chip according to claim 1, characterized in that, The protein concentration mapping module is specifically used to obtain the kinetic feature set, take the maximum binding rate and the saturation response amplitude in the kinetic feature set as input feature vectors, and call the pre-trained support vector machine model to perform regression calculations. Among them, the urine albumin concentration value The calculation formula is: ; in, The final calculated urinary albumin concentration value, For summation, For the index of the support vectors, The number of support vectors, and The Lagrange multipliers obtained from model training, For the training sample, the first Feature vectors of support vectors The input feature vector of the sample to be tested. For kernel function, These are the bias terms of the support vector machine model; Based on the regression calculation results, the urinary albumin concentration value is output.

7. The urinary albumin detection system based on a photonic crystal microarray chip according to claim 5, characterized in that, The instantaneous binding rate is calculated using the following formula: ; in, Indicates in The instantaneous binding rate at time t, For time, This is the index of the current sampling point. for The instantaneous wavelength shift at time t. For the previous moment The instantaneous wavelength shift, The time interval between two adjacent sampling points. To and The time-weighted factor relating the signal-to-noise ratio of the signal at any given moment; The time weighting factor The time weighting factor is negatively correlated with the signal-to-noise ratio (SNR). Specifically, in the initial stage of signal acquisition, when the SNR is low, a larger time weighting factor is used to enhance the smoothing effect. In the stable stage of signal acquisition, when the SNR is high, a smaller time weighting factor is used to preserve signal details.

8. The urinary albumin detection system based on a photonic crystal microarray chip according to claim 6, characterized in that, The support vector machine model uses a radial basis kernel function, and adopts a method that combines a grid search strategy based on Bayesian optimization with K-fold cross-validation to adaptively optimize the model hyperparameters; Specifically, within the preset penalty coefficient With kernel function parameters Within the combination space, all parameter combinations are systematically traversed, and K-fold cross-validation is used to evaluate the model performance under each combination. The penalty coefficient with the minimum average regression error is selected. With kernel function parameters As the final model parameters.

9. The urinary albumin detection system based on a photonic crystal microarray chip according to claim 5, characterized in that, Before filtering the maximum binding rate, the dynamic feature extraction module also performs filtering and peak detection operations on the instantaneous binding rate. Specifically, the sequence of instantaneous binding rates is first smoothed using a Savitzky-Golay filter to generate a smoothed binding rate curve. An extreme value search algorithm is applied to the smooth binding rate curve to identify all local maxima points, and the rate values ​​corresponding to all the local maxima points are compared. The largest rate value is determined as the maximum binding rate.

10. A method for detecting urinary albumin based on a photonic crystal microarray chip, characterized in that, The method is used to implement the urinary albumin detection system based on a photonic crystal microarray chip as described in any one of claims 1-9. Includes the following steps: S1: Spectral fingerprinting step, calculate the center wavelength of the steady-state mixed reflection spectrum of the photonic crystal microarray, assign a digital identity identifier to the center wavelength and construct a reference wavelength identifier library; S2: Real-time signal deconstruction step: Based on the reference wavelength identifier library, calculate the wavelength shift of each digital identifier in the mixed reflectance spectrum sequence and generate an instantaneous wavelength shift sequence; S3: Dynamic feature extraction step, based on the instantaneous wavelength drift sequence, calculate the instantaneous binding rate to screen out the maximum binding rate, and calculate the average instantaneous wavelength drift during the stable period as the saturation response amplitude to construct a dynamic feature set; S4: Protein concentration mapping step, based on the maximum binding rate and saturation response amplitude in the kinetic feature set, calls the support vector machine model to perform regression analysis and calculate the urinary albumin concentration value.

Images