Fast Reading and Analysis Method and Application of Fluorescent Immunochromatographic Test Strip Based on Deep Learning

Abstract

The present application provides a rapid reading and analysis method and application of a fluorescence immunochromatographic test strip based on deep learning. The method includes the following steps: collecting and obtaining the fluorescence spectral data of a sample to be tested on the fluorescence immunochromatographic test strip; constructing a signal preprocessing model; constructing a signal feature extraction model; constructing a time series prediction model; constructing a detection result output model; inputting the fluorescence intensity signal in the reaction progress stage collected into the signal preprocessing model for time domain discrimination and annotation, the signal feature extraction model for extracting signal feature vectors, and the time series prediction model for predicting the complete reaction feature vectors, and finally obtaining the detection result of the analyte in the sample to be tested through the detection result output model. Through the deep learning method, the present application can predict and detect the analyte to be tested according to the reaction progress signal without waiting for the reaction of the analyte on the fluorescence immunochromatographic test strip to be complete, meeting the requirement of rapid detection.

Description

Technical Field

[0001] This application relates to the technical field of fluorescence immunoassay chromatography, and specifically relates to a rapid reading and analysis method and application of a fluorescence immunoassay chromatography test strip based on deep learning. Background Art

[0002] Machine learning, especially deep learning, can handle complex non-linear relationships. When there are complex and variable dependencies between input data, it can capture high-dimensional features in the data through learning. Machine learning has been widely used in multiple medical fields, especially in image analysis, disease prediction, genomics, and biological signal processing. In the diagnosis of cardiovascular diseases, deep learning is used to analyze electrocardiogram (ECG), ultrasound images, and other relevant data to help doctors make faster and more accurate diagnoses. For example, deep learning-based models can automatically identify abnormal phenomena such as arrhythmia and myocardial infarction in ECG signals, helping doctors detect potential risks in advance and improve the clinical response speed. In the field of in vitro diagnosis, machine learning can also help analyze data and enhance accuracy. The present invention innovatively uses machine learning algorithms during the reading process of test strips to predict the growth trend of fluorescence intensity signals through time series data, shortening the diagnosis time of myocardial infarction. Summary of the Invention

[0003] This application provides a rapid reading and analysis method of a fluorescence immunoassay chromatography test strip based on deep learning, which can solve the technical problem in the prior art that the sample addition reaction time of the fluorescence immunoassay chromatography test strip is relatively long and is not conducive to rapid detection.

[0004] In a first aspect, this application provides a rapid reading and analysis method of a fluorescence immunoassay chromatography test strip based on deep learning, including the following steps:

[0005] Collect and obtain the fluorescence spectrum data of the sample to be tested on the fluorescence immunoassay chromatography test strip;

[0006] Construct a signal preprocessing model for distinguishing and labeling the reaction stage of the collected fluorescence spectrum data to obtain the labeled fluorescence spectrum data, and the reaction stage includes the reaction not started, the reaction in progress, and the reaction completed;

[0007] Construct a signal feature extraction model for extracting the fluorescence intensity signal feature vector from the labeled fluorescence spectrum data;

[0008] Construct a time series prediction model for performing time series prediction processing on the feature vector of the reaction in progress to obtain the feature vector of the reaction completed;

[0009] Construct a detection result output model for diagnosing according to the feature vector of the reaction completed to obtain the diagnosis result of the analyte in the sample to be tested;

[0010] The fluorescence intensity signal in the reaction progress stage collected is input into the signal preprocessing model for time-domain discrimination annotation, the signal feature extraction model for extracting signal feature vectors, and the time-series prediction model for predicting the complete reaction feature vectors. Finally, the detection result of the analyte in the sample to be tested is obtained through the detection result output model.

[0011] In an embodiment of the present invention, the acquisition of the fluorescence spectral data of the sample to be tested on the fluorescence immunochromatographic test strip specifically includes the following steps:

[0012] The sample to be tested is dropped onto the fluorescence immunochromatographic test strip, and the fluorescence intensity signal at each position on the fluorescence immunochromatographic test strip is read at a preset frequency until the reaction is complete, and the fluorescence spectral data is obtained.

[0013] In an embodiment of the present invention, the signal before the reaction starts refers to the fluorescence spectral signal read before the sample to be tested migrates to the detection line of the fluorescence immunochromatographic test strip; the signal when the reaction is complete refers to the signal read when the change rate of the fluorescence spectral signal is lower than the change rate threshold within a preset time period; the signal during the reaction progress refers to the fluorescence spectral signal read during the time between the start of the reaction and the completion of the reaction.

[0014] In an embodiment of the present invention, the construction of the signal preprocessing model specifically includes the following steps:

[0015] According to the reaction kinetic characteristics of the fluorescence immunochromatographic test strip, a reaction stage segmentation standard is established, and the reaction stage includes before the reaction starts, during the reaction progress, and after the reaction is complete;

[0016] According to the established reaction stage segmentation standard, the fluorescence spectral data collected is discriminated and annotated for the reaction stage.

[0017] In an embodiment of the present invention, the construction of the signal feature extraction model specifically includes the following steps:

[0018] Obtain the multi-dimensional eigenvalue of the fluorescence spectral data during the reaction progress;

[0019] Using the fluorescence spectral data during the reaction progress as the input and the multi-dimensional eigenvalue as the output, train to obtain the signal feature extraction model.

[0020] In an embodiment of the present invention, the multi-dimensional eigenvalue includes but is not limited to the peak area of the T line, the peak area of the C line, the fluorescence intensity of the T line, the fluorescence intensity of the C line, the peak area ratio of the T / C line, the fluorescence intensity ratio of the T / C line, and the background fluorescence intensity.

[0021] In an embodiment of the present invention, the construction of the time-series prediction model specifically includes the following steps:

[0022] Combine multiple fluorescence intensity signal features into a feature vector;

[0023] Set a time window and obtain a sequence of feature vectors of a fixed length;

[0024] Slide the time window to capture feature vectors covering the entire reaction stage and obtain training samples;

[0025] Input the training samples into the Transformer model to construct a time series prediction model.

[0026] In an embodiment of the present invention, the construction of the detection result output model specifically includes the following steps:

[0027] Set the concentration thresholds of the analyte corresponding to each diagnostic conclusion in the blood sample;

[0028] Obtain the fluorescence intensity signal feature vector of the analyte when the reaction is complete;

[0029] According to the set concentration diagnostic threshold of the analyte, detect and determine the complete reaction feature vector to obtain the detection result of the analyte in the test sample.

[0030] In an embodiment of the present invention, the complete reaction feature vector is the complete reaction feature vector extracted by the signal feature extraction model or the complete reaction feature vector obtained by time series prediction through the time series prediction model.

[0031] In a second aspect, the present application provides an application of a rapid reading and analysis method for a fluorescence immunochromatographic test strip based on deep learning, which is used to detect troponin I in blood.

[0032] In a third aspect, the present application provides a rapid reading and analysis system for a fluorescence immunochromatographic test strip based on deep learning, including:

[0033] A fluorescence spectrum data acquisition module, which is used to collect and obtain the fluorescence spectrum data of the test sample on the fluorescence immunochromatographic test strip;

[0034] A signal preprocessing model construction module, which is communicatively connected to the fluorescence spectrum data acquisition module and is used to construct a signal preprocessing model to distinguish and label the reaction stage of the collected fluorescence spectrum data to obtain the labeled fluorescence spectrum data, and the reaction stage includes the reaction not started, the reaction in progress, and the reaction complete;

[0035] A signal feature model construction module, which is communicatively connected to the signal preprocessing model construction module and is used to construct a signal feature extraction model to extract the fluorescence intensity signal feature vector from the labeled fluorescence spectrum data;

[0036] The time series prediction model construction module, which is communicatively connected to the signal feature model construction module, is used to construct a time series prediction model to perform time series prediction processing on the reaction feature vectors to obtain the complete reaction feature vectors;

[0037] The result output model construction module, which is communicatively connected to the time series prediction model construction module, is used to construct a detection result output model to perform diagnosis based on the complete reaction feature vectors to obtain the detection result of the analyte in the sample to be tested;

[0038] The detection result acquisition module, which is communicatively connected to the signal preprocessing model construction module, the signal feature model construction module, the time series prediction model construction module, and the result output model construction module, is used to input the fluorescence intensity signals at the reaction stage of the collected reaction into the time domain discrimination annotation of the signal preprocessing model, extract signal feature vectors by the signal feature extraction model, predict the complete reaction feature vectors by the time series prediction model, and finally obtain the detection result of the analyte in the sample to be tested through the detection result output model.

[0039] The beneficial effects brought by the technical solution provided by the embodiments of the present application at least include:

[0040] Through deep learning methods, it is possible to predict and detect the analyte according to the reaction signals without waiting for the reaction on the fluorescence immunochromatographic strip to be complete, meeting the requirements of rapid detection. Description of the Drawings

[0041] Figure 1 It is a method flow chart of the rapid reading and analysis method of the fluorescence immunochromatographic strip based on deep learning provided by the embodiments of the present application;

[0042] Figure 2 It is another method flow chart of the rapid reading and analysis method of the fluorescence immunochromatographic strip based on deep learning provided by the embodiments of the present application;

[0043] Figure 3 It is a typical fluorescence signal spectrogram of the fluorescence immunochromatographic method provided by the embodiments of the present application;

[0044] Figure 4 It is a schematic diagram of the network structure of the signal preprocessing model provided by the embodiments of the present application;

[0045] Figure 5 It is a schematic diagram of the network structure of the signal feature extraction model provided by the embodiments of the present application;

[0046] Figure 6 It is a schematic diagram of the network structure of the time series prediction model provided by the embodiments of the present application;

[0047] Figure 7 It is a schematic diagram of the network structure of the detection result output model provided by the embodiments of the present application. Detailed implementation mode

[0048] To enable those skilled in the art to better understand the solution of this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of this application.

[0049] The terms "including" and "having" in the specification, claims and above-mentioned accompanying drawings of this application, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally further includes steps or units not listed, or optionally further includes other steps or units inherent to these processes, methods, products or devices. The descriptions of terms such as "first", "second" and "third" are used to distinguish different objects, etc., which do not represent a sequence, nor do they limit that "first", "second" and "third" are of different types.

[0050] In the description of the embodiments of this application, "exemplary", "for example" or "for instance" etc. are used to indicate examples, illustrations or explanations. Any embodiment or design solution described as "exemplary", "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, the use of words such as "exemplary", "for example" or "for instance" is intended to present relevant concepts in a specific way.

[0051] In the description of the embodiments of this application, unless otherwise specified, " / " means "or". For example, A / B can mean A or B; "and / or" in the text is only a kind of association relationship describing associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "a plurality of" means two or more than two.

[0052] In some processes described in the embodiments of this application, there are multiple operations or steps that appear in a specific order. However, it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of this application or may be executed in parallel. The serial numbers of the operations are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following will further describe the embodiments of this application in detail with reference to the accompanying drawings.

[0054] In the first aspect, as Figure 1 - Figure 2 shown, the embodiments of this application provide a rapid reading and analysis method for a fluorescence immunochromatographic test strip based on deep learning, specifically including the following steps:

[0055] Step S1: Collect and obtain the fluorescence spectral data of the sample to be tested on the fluorescence immunochromatographic test strip. The sample to be tested is a liquid sample, such as blood.

[0056] Step S2: Construct a signal preprocessing model for distinguishing and labeling the reaction stage of the collected fluorescence spectral data to obtain the labeled fluorescence spectral data. The reaction stage includes the reaction not started, the reaction in progress, and the reaction completed.

[0057] Step S3: Construct a signal feature extraction model for extracting the fluorescence intensity signal feature vectors from the labeled fluorescence spectral data, including extracting the reaction in progress feature vectors from the fluorescence spectral data of the reaction not started and extracting the reaction completed feature vectors from the fluorescence spectral data of the reaction completed. In this application, it is mainly used to extract the reaction in progress feature vectors from the fluorescence spectral data of the reaction in progress.

[0058] Step S4: Construct a time series prediction model for performing time series prediction processing on the reaction in progress feature vectors to obtain the reaction completed feature vectors.

[0059] Step S5: Construct a detection result output model for diagnosing based on the reaction completed feature vectors to obtain the diagnosis result of the analyte in the sample to be tested.

[0060] Step S6: Input the collected fluorescence intensity signals in the reaction in progress stage into the signal preprocessing model for time domain discrimination and labeling, the signal feature extraction model for extracting signal feature vectors, and the time series prediction model for predicting the reaction completed feature vectors. Finally, obtain the detection result of the analyte in the sample to be tested through the detection result output model.

[0061] Through the deep learning method, based on the reaction in progress signal, predict and detect the analyte in the sample to be tested. It is possible to realize the prediction and detection of the analyte without waiting for the reaction of the analyte on the fluorescence immunochromatographic test strip to be completed, meeting the rapid detection requirements of the fluorescence immunochromatographic method for the analyte.

[0062] In one embodiment, the step S1: Collect and obtain the fluorescence spectral data of the sample to be tested on the fluorescence immunochromatographic test strip, specifically including the following steps:

[0063] Add the sample to be tested to the fluorescence immunoassay chromatography test strip, and read the fluorescence intensity signals at each position on the test strip at a preset frequency. It can accurately locate the reaction line, dynamically quantify signal changes, extract multi-dimensional features, provide high-resolution spatio-temporal data for the machine learning model, and thus achieve rapid and high-precision early prediction detection until the reaction is complete and obtain the fluorescence spectrum data of the whole reaction process.

[0064] In a specific embodiment, step S1 is implemented as follows:

[0065] Each time, 75 μL of the blood sample of a suspected myocardial infarction patient is added dropwise to the fluorescence immunoassay chromatography test strip, and then the test strip is immediately placed in the reader, and the fluorescence intensity spectrum of the test strip is read every 30 s. Read continuously until the reaction is complete, that is, the fluorescence intensity of two adjacent time steps is basically unchanged.

[0066] In an embodiment, in step S2:

[0067] The signal before the reaction starts refers to the fluorescence spectrum signal read before the sample to be tested migrates to the detection line (T line) of the fluorescence immunoassay chromatography test strip;

[0068] The signal when the reaction is complete refers to the signal read when the change rate of the fluorescence spectrum signal is lower than the change rate threshold within a preset time period; specifically, the signal when the reaction is complete refers to the signal read when the change rate of the peak area ratio of the T / C line is lower than 10% within 30 s.

[0069] The signal during the reaction refers to the fluorescence spectrum signal read during the time period between the start of the reaction and the completion of the reaction.

[0070] In an embodiment, step S2: construct a signal preprocessing model, which specifically includes the following steps:

[0071] Step S21: According to the reaction kinetic characteristics of the fluorescence immunoassay chromatography test strip, establish a reaction stage segmentation standard, and the reaction stage includes before the reaction starts, during the reaction, and after the reaction is complete; among them, the reaction stage segmentation standard is:

[0072] The liquid sample flows from the sample application area towards the absorbent pad and has not reached the detection line (T line) and the control line (C line);

[0073] The target antigen (such as cTnI) in the liquid sample specifically binds to the fluorescently labeled antibody pre-embedded on the test strip to form an "antigen-antibody-fluorescent label" complex and deposits on the detection line (T line). At the same time, the excess fluorescently labeled antibody continues to flow and binds to the control line (C line), resulting in a gradual increase in the fluorescence spectrum signal intensity of the T line and the C line over time;

[0074] Reaction completion stage: The fluorescence spectral signal intensities of the T line and the C line reach a steady state; specifically, the change rate of the T / C line peak area ratio within 30 consecutive seconds is less than 10%.

[0075] Figure 3 In (a), it is a fluorescence intensity change diagram of the fluorescence immunochromatography method for strongly positive samples;

[0076] Figure 3 In (b), it is a fluorescence intensity change diagram of the fluorescence immunochromatography method for weakly positive samples;

[0077] Figure 3 In (c), it is a fluorescence intensity change diagram of the fluorescence immunochromatography method for negative samples.

[0078] As Figure 3 According to the fluorescence intensity spectral data in (b), within the time period of 0 min - 1 min, the blood sample has not migrated to the T line, which is the reaction not started stage; and the reaction reaches completion at 8 min. The signals read from 0 - 1.5 min are marked as reaction not started signals; the signals read from 1.5 - 7.5 min are marked as reaction in progress signals; the signals read at 8 min are marked as reaction completed signals;

[0079] Step S22: According to the established reaction stage segmentation criteria, construct a two - layer fully connected neural network, namely the signal pre - processing model for signal classification, to distinguish and label each signal in the collected fluorescence spectral data for the reaction stage. The network structure is as Figure 4 shown, including an input layer, a one - dimensional flattening layer, a hidden layer, a Softmax layer, and an output layer. The input layer is used to input the collected fluorescence spectral data for data processing. The fluorescence spectral data includes fluorescence intensity and position. The one - dimensional flattening layer is used to erase the position information in the data and arrange the fluorescence intensities in position order to obtain a one - dimensional vector of fluorescence intensity. The one - dimensional flattening layer is used to erase the position information in the data and arrange the fluorescence intensities in position order to obtain a one - dimensional vector of fluorescence intensity. The hidden layer is used to learn the mapping relationship between the one - dimensional vector and the reaction stage by adjusting weights and biases; the Softmax layer is used to convert the 128 - dimensional vector output by the hidden layer into a probability distribution; the output layer is used to distinguish and label each reaction stage according to the probability distribution output by the Softmax layer.

[0080] In one embodiment, the step S3: Construct a signal feature extraction model, which specifically includes the following steps:

[0081] Step S31: Obtain the multi-dimensional eigenvalues of the fluorescence spectrum data during the reaction; wherein, the multi-dimensional eigenvalues include but are not limited to the peak area of the T line, the peak area of the C line, the fluorescence intensity of the T line, the fluorescence intensity of the C line, the peak area ratio of the T / C line, the fluorescence intensity ratio of the T / C line, and the background fluorescence intensity;

[0082] Step S32: Use the fluorescence spectrum data during the reaction as the input and the multi-dimensional eigenvalues as the output to train and obtain the signal feature extraction model as shown in Figure 5 which includes an input layer, a one-dimensional flattening layer, a parallel fully-connected feature extraction layer, and a feature output layer. The input layer is used to input the fluorescence spectrum data during the reaction, which is two-dimensional data including position and fluorescence intensity. The one-dimensional flattening layer is used to flatten the two-dimensional fluorescence spectrum data into a one-dimensional vector. The parallel fully-connected feature extraction layer includes three independent fully-connected layers. The output feature of the first branch fully-connected layer is the T line area, which quantifies the total fluorescence signal of the test line (T line) and directly reflects the concentration of the analyte. The output feature of the second branch fully-connected layer is the C line area, which is the signal intensity of the control line (C line) to verify the effectiveness of the test strip. The output feature of the third branch fully-connected layer is the T / C ratio and the background intensity, which are used to standardize the signal (eliminate environmental interference) and background noise. Dimension of each layer: input 128 dimensions → output 128 dimensions (labeled "128" in the figure); the feature output layer extracts four major features, including the T line area, the C line area, the T / C ratio, and the background intensity.

[0083] In one embodiment, step S4: Construct a time series prediction model, which specifically includes the following steps:

[0084] Step S41: Combine multiple fluorescence intensity signal features into a feature vector; first, the combination of multiple feature vectors (such as the peak area ratio of the T / C line, fluorescence intensity, etc.) comprehensively captures the reaction dynamics and avoids the deviation of a single feature;

[0085] Step S42: Set a time window to obtain a sequence of feature vectors with a fixed length;

[0086] Step S43: Slide the time window to capture the feature vectors covering the entire reaction stage and obtain training samples; the sliding time window mechanism converts the continuously sampled time series data into trainable sequence samples, which not only expands the effective data volume but also retains the time dependence of the reaction kinetics;

[0087] Step S44: Input the training samples into the Transformer model, adjust the hyperparameters of the Transformer model, and construct the time series prediction model as shown in Figure 6 through hyperparameter tuning to make the model converge quickly and achieve high prediction accuracy (the mean square error is 0.28 in the embodiment).

[0088] By means of multi-feature fusion, temporal modeling, and parameter optimization reactions, the signal enables the Transformer to accurately predict the endpoint signal based on the early reaction data.

[0089] In a specific embodiment, the step S4: constructing a temporal prediction model is specifically implemented as follows:

[0090] Taking the feature vectors of 13 reactions obtained on a single test strip as a group, there are several groups of feature vectors. The time window is set to 4, and the time window slides from the first time step (t1) to the 12th time step (t12) and stops, as Figure 6 shown. Performing this operation on each group of feature vectors results in a large amount of training data. Call the Transformer from the deep learning library and define the network structure: input dimension 4, hidden layer 128, number of attention heads 2, number of encoder layers 2. Use absolute position encoding to enable the Transformer to perceive the position of the time step in the input sequence. Extract the hidden state of the last position of the input sequence as the representation of the entire sequence, and generate the final prediction result through this state. Finally, using the time window as the input and the t13 feature vector as the output, train the Transformer iteratively 1000 times, and the mean square error is approximately 0.28.

[0091] In an embodiment, the step S5: constructing a detection result output model specifically includes the following steps:

[0092] Step S51: Set the concentration thresholds of the analytes corresponding to each diagnostic conclusion in the blood sample; specifically, calculate the concentration by inversely calculating the T / C peak area ratio of the complete reaction, and define a blood sample with <0.1 ng / mL as negative, a blood sample with 0.1 - 0.3 ng / mL as weakly positive, and a blood sample with >0.3 ng / mL as positive;

[0093] Step S52: Obtain the complete reaction fluorescence intensity signal feature vector of the analyte; wherein, the complete reaction feature vector is the complete reaction feature vector extracted by the signal feature extraction model or the complete reaction feature vector obtained by temporal prediction through the temporal prediction model;

[0094] Step S53: Use the diagnostic result as the label, and the t13 feature vector output by the temporal prediction model as the input for training the detection result output model, and obtain as Figure 7The test result output model shown includes an input layer, a hidden layer and an output layer, wherein the input layer is used to input a complete reaction feature vector; the first layer of the hidden layer maps the 4-dimensional original features to a 128-dimensional space to extract nonlinear combination features; the second layer is used to deeply abstract features from the extracted nonlinear combination features to capture the interactive relationship between the T / C ratio and the background intensity; the output layer is used to output the test result of the test object according to the interactive relationship between the T / C ratio and the background intensity, when the concentration of the test object is lower than the detection threshold, it is negative; when the concentration of the test object is between the detection threshold and the critical value range, it is weakly positive; when the concentration of the test object is higher than the diagnostic threshold, it is positive.

[0095] In one embodiment, step S6: inputting the collected fluorescence intensity signal of the reaction process stage to the signal preprocessing model for time domain differentiation and annotation, the signal feature extraction model for signal feature vector extraction, and the time series prediction model for prediction of the complete reaction feature vector, and finally obtaining the test result of the analyte of the sample to be tested through the test result output model, is specifically implemented as follows:

[0096] Finally, integrate the trained model, such as Figure 2 As shown, the time-continuous spectral data are sequentially input into the signal preprocessing model, and the signals that cannot or do not need to be predicted, such as the reaction not started signal, are eliminated. The reaction progress signal is converted into a feature vector composed of 4 features through the signal feature extraction model. The feature vectors of four consecutive time steps are input into the timing prediction model to obtain the t13 feature vector. The t13 feature vector is passed into the test result output model to obtain the test result of the object to be tested. The traditional 15-minute detection process is compressed to 3.5 minutes, while maintaining a diagnostic accuracy rate of more than 95%, to gain golden treatment time for critical diseases such as acute myocardial infarction.

[0097] On the second aspect, this application provides an application of a deep learning-based fluorescent immunochromatographic test strip rapid reading method for detecting troponin I in the blood, which is conducive to shortening the in vitro diagnosis time of troponin I, so as to automatically identify the elevated concentration of troponin I in the test strip fluorescence intensity signal, help doctors to discover potential risks in advance, and improve clinical response speed. The final test accuracy was above 95%, and the diagnosis time was shortened to a minimum of 3.5 min.

[0098] In a third aspect, the present application provides a deep learning-based rapid reading system for fluorescent immunochromatographic test strips, comprising:

[0099] A fluorescence spectrum data acquisition module is used to collect and acquire the fluorescence spectrum data of the sample to be tested on the fluorescence immunochromatography test strip;

[0100] The signal preprocessing model construction module, which is communicatively connected to the fluorescence spectrum data acquisition module, is used to construct a signal preprocessing model, distinguish and label the acquired fluorescence spectrum data in the reaction stage, and obtain the labeled fluorescence spectrum data. The reaction stage includes the reaction not started, the reaction in progress, and the reaction completed.

[0101] The signal feature model construction module, which is communicatively connected to the signal preprocessing model construction module, is used to construct a signal feature extraction model and extract the fluorescence intensity signal feature vector from the labeled fluorescence spectrum data.

[0102] The time series prediction model construction module, which is communicatively connected to the signal feature model construction module, is used to construct a time series prediction model and perform time series prediction processing on the reaction in progress feature vector to obtain the reaction completed feature vector.

[0103] The result output model construction module, which is communicatively connected to the time series prediction model construction module, is used to construct a detection result output model, diagnose according to the reaction completed feature vector, and obtain the detection result of the analyte in the sample to be tested.

[0104] The detection result acquisition module, which is communicatively connected to the signal preprocessing model construction module, the signal feature model construction module, the time series prediction model construction module, and the result output model construction module, is used to input the fluorescence intensity signal in the reaction in progress stage collected into the signal preprocessing model for time domain discrimination and labeling, the signal feature extraction model for signal feature vector extraction, and the time series prediction model for prediction of the reaction completed feature vector, and finally obtain the detection result of the analyte in the sample to be tested through the detection result output model.

[0105] In a fourth aspect, an embodiment of the present application provides a rapid reading and analysis device for a fluorescence immunochromatographic test strip for in vitro rapid diagnosis of troponin I based on deep learning. The rapid reading and analysis device for a fluorescence immunochromatographic test strip for in vitro rapid diagnosis of troponin I based on deep learning can be a device with data processing functions such as a personal computer (PC), a laptop computer, a server, an embedded device, etc.

[0106] The communication interface includes input / output (I / O) interfaces, physical interfaces, and logical interfaces, etc., which are used to implement the interconnection of components inside the rapid reading and analysis device for a fluorescence immunochromatographic test strip for in vitro rapid diagnosis of troponin I based on deep learning, and interfaces for implementing the interconnection of the rapid reading and analysis device for a fluorescence immunochromatographic test strip for in vitro rapid diagnosis of troponin I based on deep learning with other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, an optical fiber interface, an ATM interface, etc.; the user device can be a display (Display), a keyboard (Keyboard), etc.

[0107] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical memory, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0108] The processor can be a general-purpose processor, which can call the deep learning-based rapid reading and analysis program for in vitro rapid diagnosis of troponin I on a fluorescence immunochromatographic test strip stored in the memory and execute the deep learning-based rapid reading and analysis method for a fluorescence immunochromatographic test strip provided in the embodiments of the present application. For example, the general-purpose processor can be a central processing unit (CPU). Among them, the method executed when the deep learning-based rapid reading and analysis program for in vitro rapid diagnosis of troponin I on a fluorescence immunochromatographic test strip is called can refer to the various embodiments of the deep learning-based rapid reading and analysis method for a fluorescence immunochromatographic test strip of the present application, which will not be elaborated here.

[0109] In a fifth aspect, the embodiments of the present application further provide a readable storage medium.

[0110] The readable storage medium of the present application stores a deep learning-based rapid reading and analysis program for in vitro rapid diagnosis of troponin I on a fluorescence immunochromatographic test strip. When the deep learning-based rapid reading and analysis program for in vitro rapid diagnosis of troponin I on a fluorescence immunochromatographic test strip is executed by a processor, the steps of the deep learning-based rapid reading and analysis method for a fluorescence immunochromatographic test strip as described above are implemented.

[0111] Among them, the method implemented when the deep learning-based rapid reading and analysis program for in vitro rapid diagnosis of troponin I on a fluorescence immunochromatographic test strip is executed can refer to the various embodiments of the deep learning-based rapid reading and analysis method for a fluorescence immunochromatographic test strip of the present application, which will not be elaborated here.

[0112] It should be noted that the serial numbers of the embodiments of the present application above are only for description and do not represent the superiority or inferiority of the embodiments.

[0113] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-described embodiment methods can be implemented by means of software plus a necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases, the former is a better implementation. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above and includes several instructions for causing a terminal device to execute the methods described in various embodiments of the present application.

[0114] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.

Claims

1. A rapid reading and analysis method for fluorescence immunochromatographic test strips based on deep learning, characterized in that, It includes the following steps: Collect the fluorescence spectral data of the sample to be tested on the fluorescence immunochromatographic test strip; Construct a signal preprocessing model for distinguishing and labeling the reaction stages of the collected fluorescence spectral data to obtain the labeled fluorescence spectral data, where the reaction stages include the reaction not started, the reaction in progress, and the reaction completed; Construct a signal feature extraction model for extracting the fluorescence intensity signal feature vector from the labeled fluorescence spectral data; Construct a time series prediction model for performing time series prediction processing on the feature vector of the reaction in progress to obtain the feature vector of the reaction completed; The construction of the time series prediction model specifically includes the following steps: Combine multiple fluorescence intensity signals into a feature vector; Set a time window to obtain a sequence of feature vectors of a fixed length; Slide the time window to capture the feature vectors covering the entire reaction stage to obtain training samples; Input the training samples into the Transformer model, adjust the hyperparameters of the Transformer model, and construct the time series prediction model; Construct a detection result output model for diagnosing according to the feature vector of the reaction completed to obtain the diagnostic result of the analyte in the sample to be tested; Input the fluorescence intensity signal in the reaction in progress stage collected into the signal preprocessing model for time domain discrimination and labeling, the signal feature extraction model for signal feature vector extraction, and the time series prediction model for predicting the feature vector of the reaction completed, and finally obtain the detection result of the analyte in the sample to be tested through the detection result output model.

2. The rapid reading and analysis method of the fluorescence immunochromatographic test strip based on deep learning according to claim 1, wherein The collection of the fluorescence spectral data of the sample to be tested on the fluorescence immunochromatographic test strip specifically includes the following steps: Drop the sample to be tested onto the fluorescence immunochromatographic test strip, and read the fluorescence intensity signal at each position on the fluorescence immunochromatographic test strip at a preset frequency until the reaction is completed to obtain the fluorescence spectral data.

3. The rapid reading and analysis method of the fluorescence immunochromatographic test strip based on deep learning according to claim 1, wherein The signal of the reaction not started refers to the fluorescence spectral signal read before the sample to be tested migrates to the detection line of the fluorescence immunochromatographic test strip; the signal of the reaction completed refers to the signal read when the change rate of the fluorescence spectral signal is lower than the change rate threshold within a preset time period; the signal of the reaction in progress refers to the fluorescence spectral signal read during the time period between the reaction not started and the reaction completed.

4. The rapid reading and analysis method of the fluorescence immunochromatographic test strip based on deep learning according to claim 1, wherein, The construction of the signal preprocessing model specifically includes the following steps: According to the reaction kinetic characteristics of the fluorescence immunochromatographic test strip, establish a reaction stage segmentation standard, where the reaction stages include the reaction not started, the reaction in progress, and the reaction completed; According to the established reaction stage segmentation standard, perform the distinction and labeling of the reaction stages on the collected fluorescence spectral data.

5. The rapid reading and analysis method of the fluorescence immunochromatographic test strip based on deep learning according to claim 1, characterized in that The construction of the signal feature extraction model specifically includes the following steps: Obtain the multi-dimensional eigenvalues of the fluorescence spectral data of the reaction in progress; Use the fluorescence spectral data of the reaction in progress as the input and the multi-dimensional eigenvalues as the output to train and obtain the signal feature extraction model.

6. The rapid reading and analysis method of the fluorescence immunochromatographic test strip based on deep learning according to claim 5, wherein, The multi-dimensional eigenvalues include but are not limited to the peak area of the T line, the peak area of the C line, the fluorescence intensity of the T line, the fluorescence intensity of the C line, the peak area ratio of the T / C line, the fluorescence intensity ratio of the T / C line, and the background fluorescence intensity.

7. The rapid reading and analysis method of the fluorescence immunochromatographic test strip based on deep learning according to claim 1, wherein, The construction of the detection result output model specifically includes the following steps: Set the concentration thresholds of the analytes corresponding to each diagnostic conclusion of the analyte in the blood sample; Obtain the complete fluorescence intensity signal feature vector of the analyte to be measured; According to the set diagnostic threshold of the analyte concentration, detect and determine the complete reaction feature vector, and obtain the detection result of the analyte in the sample to be measured.

8. The rapid reading and analysis method of the fluorescence immunochromatographic test strip based on deep learning according to claim 7, characterized in that, The complete reaction feature vector is the complete reaction feature vector extracted by the signal feature extraction model or the complete reaction feature vector obtained by time series prediction through the time series prediction model.

9. Use of a rapid reading and analysis method for a fluorescence immunochromatographic test strip based on deep learning according to any one of claims 1-8, characterized in that, It is used to detect troponin I in blood.

Images