Method for revealing time sequence enzyme-linked immunosorbent assay signal in advance

Abstract

The invention relates to a method for revealing a time sequence enzyme-linked immunosorbent assay signal in advance, and the method comprises the following steps: training time sequence colorimetric analysis data, utilizing an improved YOLOv8n-cls deep learning model, and analyzing a partial reaction process video of a chromogenic reaction in a time sequence enzyme-linked immunosorbent assay to obtain a final enzyme-linked immunosorbent assay signal; the partial reaction process video is a partial reaction video after the colorimetric change begins to occur. According to the method, training is carried out by using the time sequence data of the video, the workload is smaller compared with construction of an image data set in a traditional image training method, and by capturing dynamic subtle changes in the video, the accuracy and sensitivity of the model are further improved compared with the traditional image training method. By adopting the method designed by the invention, the reaction waiting time can be greatly reduced, and the detection efficiency is remarkably improved.

Description

Technical Field

[0001] The present application relates to the field of microfluidic analysis and detection, and in particular to a method for revealing sequential enzyme-linked immunosorbent assay signals in advance. Background Art

[0002] As a portable point-of-care test (POCT) tool, microfluidic paper-based analytical devices (μPADs) have shown great potential in early disease diagnosis and health monitoring. μPADs use the capillary force of the microchannel to drive fluid flow, and can self-drive to complete the detection work without an external power pump. It also has the characteristics of low manufacturing cost, easy operation and portability. μPADs combined with enzyme-linked immunosorbent assay (ELISA) based on colorimetric reaction can be used for the detection of disease markers in biological samples such as human blood and urine, such as neurological biomarkers Aβ42 and P-Tau, and myocardial markers troponin I (cTnI) and calcitonin T (cTnT). Through the final color change detected by ELISA, users can intuitively obtain the test results.

[0003] However, the ELISA color reaction usually requires a waiting time of about 20 to 30 minutes. The long waiting time limits the application of μPADs in ELISA rapid detection. Therefore, it is urgent to propose a method that can reduce the reaction waiting time and reveal the enzyme-linked immunosorbent assay signal in advance. Summary of the invention

[0004] The purpose of the present application is to solve the problem of long waiting time in sequential enzyme-linked immunosorbent assay in the prior art.

[0005] In a first aspect, the present application provides a method for revealing a time-sequential enzyme-linked immunosorbent assay signal in advance in a microfluidic analysis platform, the method comprising:

[0006] Based on the improved YOLOv8n-cls model, the final signal of the time-sequential enzyme-linked immunosorbent assay was obtained by analyzing the partial reaction process video of the colorimetric reaction in the time-sequential enzyme-linked immunosorbent assay; the partial reaction process video was the partial reaction video after the colorimetric change began to occur.

[0007] Further specifically, the improved YOLOv8n-cls model includes a backbone network CSPDarknet for extracting image feature information, a feedforward neural network for processing the feature information, and a classification layer for outputting the probability of each category of the final image signal.

[0008] More specifically, the partial reaction video accounts for 10% to 20% of the length of the total color reaction video.

[0009] More specifically, the classification module adopts a softmax activation function.

[0010] More specifically, in the training stage of the improved YOLOv8n-cls model, a training set is obtained by image processing and ROI extraction of a process video of a color development reaction in a sequential enzyme-linked immunosorbent assay in a microfluidic analysis platform.

[0011] More specifically, the image processing includes grayscale conversion, Gaussian filtering, background removal and boundary smoothing.

[0012] More specifically, the ROI includes the area where the color development reaction generates a colorimetric signal.

[0013] More specifically, the training stage includes data enhancement training, and the data enhancement training includes at least one of image rotation, image distance change, image scaling, image brightness change, image mirror conversion and image clarity adjustment.

[0014] In the second aspect, the present application provides a device used in the model training in the above-mentioned method, including a main body with a U-shaped opening, a plurality of U-shaped grooves opened on the side walls of the U-shaped opening, a first tray placed in the U-shaped grooves for placing test paper chips, and a second tray arranged on the top of the main body and fixed with a shooting device; the plurality of U-shaped grooves are arranged parallel to the second tray, the U-shaped grooves extend along one end of the side wall of the U-shaped opening to the other end of the side wall, the second tray is provided with a shooting port corresponding to the first tray, and the shooting device has a model training program for performing the above-mentioned method.

[0015] More specifically, an LED lamp is arranged on the side wall of the U-shaped opening.

[0016] More specifically, the second tray is arranged on the top of the main body through a height adjustment structure.

[0017] More specifically, a background plate is provided at the bottom of the U-shaped opening, and the background plate corresponds to the first tray.

[0018] More specifically, it includes a slot arranged on the other side of the U-shaped opening, and a third tray arranged in the slot, wherein the third tray has a folding bracket.

[0019] In the third aspect, the present application also provides an ELISA determination software, the software includes a user login interface, a main interface, an image analysis interface, a video analysis interface, a determination result interface, a history record interface and a help interface. When the user opens the software, the login interface is displayed first, and the software will jump to the main interface after the user enters the login information. The main interface includes an ImageAnalysis module, a Video Analysis module, a History module and an FAQs module. The image analysis interface includes a ChangeImage module, an Upload module, a Sensing Target module and a Start module. The determination result interface includes a result display area, a Download module, a Share module and a RE-Test module. The video analysis interface includes a Change Video module, an Upload module, a Parameters module, a Sensing Target module and a Start module.

[0020] This application proposes a method based on an improved YOLOv8n-cls model to reveal the signal of a sequential ELISA in advance by analyzing a partial reaction process video of the color reaction in the sequential ELISA. The model is trained using the time series data of the video, which has a smaller workload than the construction of the image data set in the traditional image training method; by capturing the subtle changes in the dynamics in the video, the accuracy and sensitivity of the model are also improved compared to the traditional image training method. The method designed by this application can greatly reduce the reaction waiting time and significantly improve the detection efficiency. For rabbit IgG ELISA determination, the waiting time can be reduced by 33%, and for cTnI ELISA determination, the waiting time can be reduced by 60%. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the experimental process of this application;

[0022] Figure 2 It is a schematic diagram of the analysis process of the model of this application;

[0023] Figure 3 It is a three-dimensional schematic diagram of the model training device of the present application;

[0024] Figure 4 It is an exploded schematic diagram of the model training device of the present application;

[0025] Figure 5 is a schematic diagram of the porous paper chip of the present application;

[0026] Figure 6 This is the rabbit IgG test standard curve of this application;

[0027] Figure 7is the confusion matrix and ROC curve of the rabbit IgG test video analysis in this application;

[0028] Figure 8 is the confusion matrix and ROC curve of the rabbit IgG test image analysis in this application;

[0029] Fig. 9 is the cTnl test standard curve in this application;

[0030] Fig.10 is the confusion matrix and ROC curve of the cTnl test video analysis in this application;

[0031] Fig.11 This is the program interface diagram in this application.

[0032] In the figure: 100, model training device; 101, height adjustment button; 102, U-shaped groove; 103, LED light; 104, height adjustment structure; 110, first tray; 120, porous paper chip; 130, second tray; 131, shooting port; 140, third tray. DETAILED DESCRIPTION

[0033] In order to make the purpose, technical scheme and advantages of the implementation of this application clearer, the specific implementation methods of this application are clearly and completely described below. Obviously, the described embodiments are part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0034] In the description of the present application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0035] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances. In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0036] In paper-based ELISA testing, the final test result is based on the final colorimetric signal of the color development reaction, and the color development reaction usually requires waiting for about 30 minutes to obtain a reliable colorimetric signal, which limits the need for rapid detection of paper-based ELISA. Figure 1 As shown, the present application provides a method capable of revealing the signal of a sequential ELISA in advance. The method is based on an improved YOLOv8n-cls model. By analyzing a portion of the video after the 50% reaction process segment of the color development reaction in the sequential ELISA, the final signal of the sequential ELISA can be obtained, which greatly shortens the long waiting time in traditional paper-based ELISA.

[0037] In some specific embodiments, the improved YOLOv8n-cls model consists of three parts. The first part is the backbone network, which uses CSPDarknet, which extracts key features from the processed video images through convolutional layers and residual connections. One set of feasible parameter settings is as follows: the depth coefficient is 1, the width coefficient is 1.25, and the maximum number of channels is 1024. The second part is a feedforward neural network, which receives the key features extracted by the backbone network and further processes them, including a fully connected or convolutional layer to map the features to a specific category, and one set of feasible parameter settings is as follows: the number of hidden layers is 1, and the number of channels in the hidden layer is 1024. The third part is a classification module, which uses a softmax activation function to output the probability of the final signal category of each key feature to complete the classification task.

[0038] In some specific embodiments, the color reaction part video used for analysis uses the video of the 50% to 67% reaction process segment in the color reaction, and the method can reduce the waiting time of the color reaction by 33%. Specifically, taking the total color reaction time of 30 minutes as an example, only the process video of the 15th to 20th minute of the color reaction can be used for analysis, and a reliable final colorimetric signal can be quickly obtained after analysis, reducing the waiting time of the color reaction by 10 minutes.

[0039] In some specific embodiments, when the improved YOLOv8n-cls model is trained, a training set is obtained by image processing and ROI extraction of a process video of a color development reaction in a sequential enzyme-linked immunosorbent assay in a microfluidic analysis platform.

[0040] In some specific embodiments, the microfluidic analysis platform uses a porous paper chip, and the image processing uses the micropore area extracted from the color reaction process video automatically recorded by the open source library OpenCV as the region of interest (ROI). The ROI includes the circular detection hole that generates the colorimetric signal and its outer rectangle.

[0041] The video processing process specifically includes: video frame extraction, such as Figure 2 As shown, the video is segmented into individual frames at a speed of 30 frames per second (fps), each frame represents a snapshot image of the colorimetric reaction at a specific time point, and an image is extracted from the video every 2 seconds; grayscale conversion, the image is converted from RGB to grayscale using the formula Gray = 0.2989 × R + 0.5870 × G + 0.1140 × B, where R, G, and B correspond to the red, green, and blue color values ​​of the pixel, respectively; Gaussian filtering, Gaussian filtering is applied to remove grayscale image noise and smooth the image to facilitate the subsequent extraction of the circular detection hole contour; background removal, the grayscale image is converted into a binary (black and white) image by calculating the optimal threshold using the Otsu method to separate the foreground (circular detection hole) and the background (the area outside the circular detection hole of the paper chip) to maximize the inter-class variance. In the resulting binary image, the circular detection hole is set to white (255) and the background is set to black (0). Morphological operations (such as corrosion and dilation) are used to clean up the binary image, further remove small noise and smooth the boundaries of the detection hole to ensure clear edges. Canny edge detection is then applied to identify the edges of the circular detection holes, followed by the OpenCV findContours function to identify contours. These identified contours correspond to the ROIs of the circular detection holes, and finally these ROIs are extracted to form a training set.

[0042] In some specific embodiments, the model training process of the improved YOLOv8n-cls model also includes data enhancement training. Data enhancement training includes image rotation training, image distance change training, image scaling training, image brightness change training, image mirror conversion training and image clarity adjustment training. Image rotation training: During training, the image is randomly rotated by -90° to +90°. This training can help the model recognize feature images in different directions. Image distance change training: During training, the image is randomly translated by -10% to +10% in width or height. This training can help the model recognize feature images in different positions. Image scaling training: During training, the image is scaled to 80% to 120% of the original image size. This training can help the model recognize feature images of different scales. Image brightness change training: During training, the brightness ratio factor of the image is randomly adjusted between -0.1 and 0.1. This training can help the model recognize feature images under different lighting conditions. Image mirror conversion training: During training, the image is randomly flipped with a probability of 50%. This training can help the model recognize mirror feature images. Image clarity adjustment training: During training, Gaussian blur is applied to the image with a probability of 10%, using a 5×5 kernel and a standard deviation of 0.5 for blurring. This training can help the model recognize blurred feature images and adapt to feature images of different shooting qualities.

[0043] like Figure 3 As shown, the present application provides a device for improving the YOLOv8n-cls model for training. Figure 4 As shown in the exploded view, the model training device 100 includes a main body with a U-shaped opening, a plurality of U-shaped grooves 102 opened on the side walls of the U-shaped opening, a first tray 110 placed in the U-shaped grooves 102, and a second tray 130 arranged on the top of the main body; the plurality of U-shaped grooves 102 are arranged parallel to the second tray 130, the U-shaped grooves 102 extend along one end of the side wall of the U-shaped opening to the other end of the side wall, and the second tray 130 is provided with a shooting port 131 corresponding to the first tray 110.

[0044] The first tray 110 is used to place the porous paper chip 120 to be photographed. In a further improved solution, in order to facilitate the positioning of the porous paper chip 120, a groove corresponding to the porous paper chip 120 is provided on the first tray 110. In order to facilitate the flat placement of the porous paper chip 120 to prevent wrinkles, movement, etc., a clamping structure, such as a magnetic clamp, etc., can also be provided on the first tray 110.

[0045] In some specific embodiments, a plurality of U-shaped grooves 102 divide the side walls of the U-shaped opening into a first side wall, a second side wall, a third side wall, ... from top to bottom. In a feasible solution, an LED lamp 103 (5V-COB) is provided on the first side wall. When the device is running, the LED lamp 103 is turned on to evenly illuminate the porous paper chip 120 on the first tray 110. In a feasible solution, LED lamps 103 (5V-COB) are provided on the first side wall, the second side wall, the third side wall, ..., and when the device is running, the LED lamps 103 on the upper and lower adjacent side walls of the first tray 110 are turned on to evenly illuminate the back and front of the porous paper chip 120. In a further improved solution, the brightness of the LED lamp 103 is adjusted by the control system of the device. A background plate is provided at the bottom of the U-shaped opening, and the background plate corresponds to the first tray.

[0046] In some specific embodiments, the second tray 130 is arranged on the top of the main body through a height adjustment structure 104. The height adjustment structure 104 can be a height adjustment screw rod. The screw rod can be adjusted through the control system of the device, and can also be controlled to increase or decrease through the height adjustment button 101 on the side of the device. A shooting device is placed on the second tray 130. The shooting device can select a suitable mobile phone or other device according to the user's own equipment conditions. The size of the second tray 130 can also be adjusted according to the size of the selected device. The second tray 130 is also provided with a clamping device for clamping a mobile phone to enhance the stability and consistency of the captured video and avoid affecting the video quality due to shaking. In the following specific embodiments, the iPhone 13Pro produced by Apple is used for shooting. The camera of the iPhone 13Pro is aimed at the first tray 110 through the shooting port 131 to shoot the color reaction process video of the porous paper chip 120. The iPhone 13Pro has a program for video analysis and model training.

[0047] In some specific embodiments, the device further includes a third tray 140, which is disposed on the other side of the device opposite to the U-shaped opening. The third tray 140 can be inserted into or extended out of the device through a slot, and the third tray 140 is used to place paper chips to be processed. In a further improved scheme: to facilitate the positioning of the porous paper chips, the third tray 140 is provided with grooves corresponding to the porous paper chips. In order to facilitate the flat placement of the porous paper chips to prevent wrinkles and prevent the paper chips from moving during operation, a clamping structure, such as a magnetic clamp, may also be provided on the third tray 140. The third tray 140 also has a folding support frame. When the paper chips need to be processed, the third tray 140 is pulled out of the device, the folding support frame is opened, and the paper chips are placed on the third tray 140 for operations such as dripping reagents.

[0048] The device also has a charging port to ensure that the LED light 103 and the highly retractable structure 104 have an uninterrupted power supply during long experiments and model training.

[0049] In some specific embodiments, model training is performed using Figure 5 The porous microfluidic paper-based analytical device (μPAD) shown. The porous μPAD preparation method: The porous structure with a diameter of 5.6 mm was designed using AutoCAD software, and the hydrophobic pattern was printed onto Whatman No. 1 chromatography paper using a wax printer (S880DN, Xerox), and then the paper was heated at 120°C for 30 seconds to melt the wax to form the porous μPAD.

[0050] The above-mentioned multiporous μPAD was used to perform ELISA assays for cTnI and rabbit IgG. The selection and sources of each reagent are as follows: cTnI antigen (IAC00102), cTnI coated monoclonal antibody (L3C00401) and HRP labeled monoclonal antibody (L3C00402) were purchased from Linc-bio (Shanghai, China). Rabbit IgG (15006) was purchased from Sigma-Aldrich (Shanghai, China), ALP-conjugated anti-rabbit IgG (A0239) and BCIP / NBT colorimetric kit (C3206) were purchased from Biotech Research Institute (Haimen, China). Potassium periodate (>99.5%, 80106916) was purchased from Sinopharm Chemical Reagent Co., Ltd. (Shanghai, China). TMB substrate solution (CO4-03002) was purchased from Bios Antibodies (Beijing, China). 10X phosphate buffered saline (PBS, G4207), Tween-20 (G5058), and bovine serum albumin (BSA, G5001) were purchased from Servicebio (Wuhan, China).

[0051] Rabbit IgG ELISA assay experimental procedure: 3 μL of different concentrations of rabbit IgG antigen (6.7 μM, 670 nM, 67 nM, 6.7 nM, 670 pM, 67 pM, 6.7 pM) and 0 concentration were used as control experiments. Different concentrations of rabbit IgG antigen were fixed in the test wells, and after incubation at room temperature for 10 minutes, each well was treated with 3 μL of blocking buffer (consisting of 1% w / v BSA and 0.05% w / v Tween-20 in PBS) to block nonspecific binding sites. After drying at room temperature for 10 minutes, 3 μL of ALP-conjugated anti-rabbit IgG antibody (1:100 diluted from a stock solution from goat sources in PBS with 0.05% w / v Tween-20) was introduced into the test wells. Incubate on a 37°C hot plate for 1 minute, and then rinse the wells thoroughly twice with 10 μL of 1× PBS. Finally, 3 μL of BCIP / NBT substrate solution with a pH of 9.5 was added to each test well and incubated for 30 minutes to initiate the color development reaction. The time of adding the substrate solution was defined as Time 0 of the ELISA color development reaction, Time 30 refers to the time point 30 minutes after Time 0, and Time 15-20 refers to the period from 15 to 20 minutes after Time 0.

[0052] The above experimental process was used to test the above 8 different concentrations of rabbit IgG antigens, and the following results were obtained by curve fitting: Figure 6 The calibration curve shown in Figure 1 shows a detection limit of 43 pM (n=8). The 0-min process video of the color reaction was used, and the recording started from the 10th minute (the time when the colorimetric signal began to be generated). The 10-40-min video was divided into 6 process videos at a time interval of 5 minutes (10-15 minutes, 15-20 minutes, 20-25 minutes, 25-30 minutes, 30-35 minutes, and 35-40 minutes were obtained). The receiver operating characteristic (ROC) curve, confusion matrix, and quantitative indicators (accuracy, sensitivity, and specificity) were used to evaluate the performance of the model trained in each time period. Accuracy is defined as the ratio of correctly predicted samples (true positive TP and true negative TN) to the total number of samples (including false negative FN and false positive FP), as shown in Formula 1.

[0053]

[0054] Sensitivity is defined as the ratio of TP to the sum of TP and FN (as shown in Formula 2), which is used to measure the effectiveness of the model in identifying positive cases.

[0055]

[0056] Specificity is defined as the ratio of TN to the sum of TN and FP (as shown in Formula 3), which is used to measure the ability of the model to correctly identify negative samples.

[0057]

[0058] The confusion matrix and ROC curve of the 15-20min process video are as follows Figure 7 As shown. The ROC curve provides the area under the curve (AUC) value for each concentration. The closer the AUC is to 1, the better the classification performance of the model. The average AUC values ​​of different time periods in the ROC curve in this application are slightly different, with a difference of <1.66%, among which the maximum AUC is 0.981 and the minimum AUC is 0.965, indicating that the model performance of different time periods in this application is similar. Among them, the 35-40min process video provides the best average AUC value of 0.981, but compared with the average AUC value of 0.974 of the 15-20min process video, the improvement is only 0.72%, and the required color reaction waiting time has doubled. The model accuracy obtained by training the 15-20min process video is 94.1%, the accuracy of the 30-35min process video is 95.6%, and the accuracy of the 35-40min process video drops to 94.3%. As the color reaction proceeds, the evaporation of the reagent will gradually appear. Excessive reaction waiting time will cause the evaporation of the reagent to increase, thereby affecting the accuracy of the final detection. In the present application, the preferred process analysis video is a color reaction video of 15 to 20 minutes. Using the method designed in the present application, the final signal can be obtained in advance by analyzing only 10% to 20% of the video length after the color reaction begins to produce colorimetric changes. When a video of 15 to 20 minutes is used for analysis, the waiting time can be reduced by 10 minutes, and the detection efficiency is improved by 33.7%.

[0059] Further compare the advantages of using dynamic process video data sets in this application with the traditional use of static image data sets for model training. Taking the above-mentioned c-ELISA experiments of 8 different concentrations of rabbit IgG as the sample set, in the image training, each concentration experiment is repeated 16 times, and 30 images are collected at Time 30, and 8 concentrations × 16 times × 30 = 3840 images are obtained, of which 90% of the images are used as training sets for model training, and the remaining 10% (384 images) are reserved for testing the model. In video training, each concentration experiment is repeated 16 times, and a 30-minute process video is collected. 30 frames of images are extracted from each minute of the process video using an image processing program, and 8 concentrations × 16 times × 30 images / minute × 30 minutes = 115200 images are obtained. In the two training methods, the image training method needs to collect 8 × 16 × 30 = 3840 times, while the video training method only needs to collect 8 × 16 = 128 times. The video training method can obtain a larger image training set in fewer acquisition times through an image processing program.

[0060] The video training was performed using a training set with the same number of images as the image training method, and the performance indicators of the two methods were compared. Figure 8 As shown in the figure, the average AUC value of the ROC curve of the image-based method is 0.975, which is close to the average AUC value of 0.974 for the 15-20min process video, but the image training method requires 50% more waiting time (from 20 minutes to 30 minutes). The performance of the two methods was further evaluated by confusion matrix analysis. The video training method achieved 76.3% sensitivity and 96.6% specificity, significantly reducing false positives and false negatives compared to the traditional image training method. The traditional method used the final 30-minute snapshot, with a sensitivity of 74.0% and a specificity of 96.3%. The confusion matrix of the time series method showed a more balanced performance, with a 9% reduction in false positives. This improvement in classification accuracy makes the model trained by the time series method not only have higher sensitivity, but also provide more reliable output results.

[0061] The present application also provides an ELISA detection example for human cardiac troponin I (cTnI). The increase in cTnI levels is crucial for the diagnosis of myocardial infarction, heart failure, myocarditis and other heart diseases, and can reflect the degree of myocardial damage. The critical value of cTnI is 500pg / mL.

[0062] First, cTnI ELISA was performed according to the above detection process, and the curve fitting method was used to generate the following Fig. 9The cTnI calibration curve shown in the figure has a detection limit LOD of 204 pg / mL. The data sets in the above embodiments were constructed to compare the performance of the improved YOLOv8n-cls model in different time periods. The complete color reaction time of the cTnI ELISA assay is 25 minutes, and the time when the colorimetric signal begins to be generated is the 5th minute. Using the video from the 5th to 25th minute of the color reaction, four time periods were extracted in sequence with a video length of 5 minutes, namely 5-10 min video, 10-15 min video, 15-20 min video, and 20-25 min video. The confusion matrix and ROC curve of the 5-10 min video are shown as follows: Fig.10 As shown. The use of 5-10 min video analysis has excellent quantitative indicators, with an accuracy of 98.1%, a sensitivity of 94.4%, and a specificity of 98.9%. The sensitivity and specificity of the method provided in this application for cTnI detection are comparable to the performance indicators of existing clinical cardiac biomarker assays. Prolonging the reaction waiting time can further improve the performance. For example, the accuracy is 99.4%, the sensitivity is 98.9%, and the specificity is 99.4% when using 15-20 min video, but the waiting time is doubled compared to 5-10 min video. The use of the method designed in this application to reveal the cTnI ELISA assay signal in advance can significantly improve the detection efficiency, and the use of 5-10 minutes of process video for analysis can reduce the waiting time by 60%.

[0063] like Fig.11 As shown, the present application also provides an ELISA assay software installed on a smart phone, the software including a user login interface, a main interface, an image analysis interface, a video analysis interface, an assay result interface, a history record interface and a help interface. The login interface includes login information to be entered. The main interface includes an Image Analysis module, a Video Analysis module, a History module and an FAQs module. The image analysis interface includes a Change Image module, an Upload module, a Sensing Target module and a Start module. The assay result interface includes a result display area, a Download module, a Share module and a RE-Test module. The video analysis interface includes a Change Video module, an Upload module, a Parameters module, a Sensing Target module and a Start module.

[0064] When the user opens the software, the login interface is displayed first. After the user enters the login information, the software will jump to the main interface. The main interface includes the Image Analysis module, Video Analysis module, History module and FAQs module. The user clicks the Image Analysis module on the main interface to jump to the image analysis interface. The image analysis interface includes the Change Image module, Upload module, Sensing Target module and Start module. After the user selects the image to be predicted and uploads it, click the Start module, and the interface will jump to the measurement result interface. The measurement result interface includes the result display area, Download module, Share module and RE-Test module. The result display area will display the final ELISA measurement result predicted by the software. The user can click the Download module to download the test report, click the Share module to share the test report, or click the RE-Test module to jump back to the image analysis interface. The user clicks the Video Analysis module to jump to the video analysis interface. The video analysis interface includes the Change Video module, Upload module, Parameters module, Sensing Target module and Start module. After the user selects the video to be predicted and uploads it, click the Start module, and the interface will jump to the measurement result interface. The measurement result interface includes the result display area, Download module, Share module and RE-Test module. The result display area will display the final ELISA measurement result predicted by the software. The user can click the Download module to download the test report, click the Share module to share the test report, or click the RE-Test module to jump back to the video analysis interface.

[0065] The above detailed description of the embodiments of the present application is provided, but the present application is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions and variations of these embodiments are made without departing from the principles and spirit of the present application, and still fall within the scope of protection of the present application.

Claims

1. A method for revealing a time-series enzyme-linked immunosorbent assay signal in advance, characterized in that: The method comprises: Based on the improved YOLOv8n-cls model, the final signal of the time-sequential enzyme-linked immunosorbent assay was obtained by analyzing the partial reaction process video of the colorimetric reaction in the time-sequential enzyme-linked immunosorbent assay; the partial reaction process video was the partial reaction video after the colorimetric change began to occur.

2. The method according to claim 1, characterized in that The improved YOLOv8n-cls model includes a backbone network for extracting feature information of video images, a feedforward neural network for processing the feature information, and a classification layer for outputting the probability of each category of the final signal of the image.

3. The method according to claim 1, characterized in that The partial reaction video accounts for 10% to 20% of the length of the total color development reaction video.

4. The method according to claim 1, characterized in that: The improved YOLOv8n-cls model training stage uses a process video of a color development reaction in a sequential enzyme-linked immunosorbent assay in a microfluidic analysis platform, and obtains a training set after image processing and ROI extraction.

5. The method according to claim 4, characterized in that The image processing includes grayscale conversion, Gaussian filtering, background removal and boundary smoothing.

6. The method according to claim 4, characterized in that The ROI includes the area where the colorimetric signal is generated by the colorimetric reaction.

7. The method according to claim 4, characterized in that The training stage includes data enhancement training, and the data enhancement training includes at least one of image rotation, image distance change, image scaling, image brightness change, image mirror conversion and image clarity adjustment.

8. A model training device applied to the method according to any one of claims 4 to 7, characterized in that: It includes a main body with a U-shaped opening, a plurality of U-shaped grooves opened on the side walls of the U-shaped opening, a first tray placed in the U-shaped grooves for placing test paper chips, and a second tray arranged on the top of the main body and fixed with a shooting device; the plurality of U-shaped grooves are arranged in parallel with the second tray, the U-shaped grooves extend from one end of the side wall of the U-shaped opening to the other end of the side wall, the second tray is provided with a shooting port corresponding to the first tray, and the shooting device has a model training program for performing the method described in any one of claims 4 to 7.

9. The model training device according to claim 8, characterized in that: The side wall of the U-shaped opening is provided with an LED lamp.

10. The model training device according to claim 8, characterized in that: The second tray is arranged on the top of the main body through a height adjustment structure.

11. The model training device according to claim 8, characterized in that: It comprises a slot arranged on the other side of the U-shaped opening, and a third tray arranged in the slot, wherein the third tray has a folding bracket.

Images