Digital PCR liquid drop identification method and system based on artificial intelligence algorithm

Abstract

The invention relates to a digital PCR liquid drop identification method and system based on an artificial intelligence algorithm, and the method comprises the steps: generating liquid drops through a micro-fluidic chip, and placing the micro-fluidic chip in PCR equipment for PCR reaction circulation; after the PCR reaction circulation is finished, placing the micro-fluidic chip under a microscope for observation so as to capture a PCR liquid drop image; and preprocessing the PCR liquid drop image, inputting the preprocessed PCR liquid drop image into a pre-constructed liquid drop identification model, and outputting to obtain a PCR liquid drop detection result. Compared with the prior art, on one hand, the micro-fluidic chip is designed, liquid drops with stable structures and consistent sizes can be generated, on the other hand, by introducing the pre-trained artificial intelligence model, efficient identification of the number of the liquid drops and intelligent judgment of a threshold value are achieved, background noise can be effectively reduced, and meanwhile the detection accuracy is improved. And the accuracy of liquid drop identification is obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformation data processing, and in particular to a digital PCR droplet recognition method and system based on artificial intelligence algorithm. Background Art

[0002] Polymerase chain reaction (PCR) is an important in vitro nucleic acid amplification technology. Since the mid-1980s, it has played an important role in clinical diagnosis and biotechnology due to its high specificity, sensitivity and ease of operation. It can quickly amplify target DNA fragments millions of times. With technological advances, digital PCR (dPCR) has emerged, further addressing the limitations of traditional PCR in gene copy number and mutation analysis. dPCR achieves PCR amplification and absolute quantification at the single-molecule level by distributing samples into tens of thousands of independent reaction units, each of which may contain zero or one DNA template. This method provides a precise nucleic acid quantification method that does not rely on cycle thresholds by counting fluorescence signals.

[0003] Currently, dPCR devices mainly use microwell or microdroplet technology to divide samples and perform single-molecule PCR reactions. dPCR technology divides samples into thousands of nanoliter droplets, each containing zero or more target molecules, and determines the presence or absence of target sequences through PCR amplification and fluorescence detection.

[0004] In existing equipment, factors such as background noise in the instrument often interfere with the recognition and judgment of droplets, thereby reducing the accuracy of recognition. In addition, current droplet recognition methods mainly rely on manual observation or automated systems based on simple image processing algorithms. These methods have the problem of low efficiency when processing large amounts of data, and after finding the corresponding droplet shape structure, it is easy to misjudge whether the brightness of the droplet exceeds the threshold.

[0005] In existing droplet recognition research, for example, patent CN 112813152 B proposes a digital PCR droplet fluorescence detection method based on image recognition. This method calculates the fluorescence mean of each droplet based on the position parameter, thereby realizing the identification and judgment of the number of droplets in the PCR reaction chip. However, this method has not yet solved the problem of the influence of background noise on droplet recognition. Patent CN 117455922 B proposes an automatic counting and analysis method based on droplet motion images. This method has a high recognition accuracy when the droplets do not overlap. However, factors such as droplet overlap will lead to a decrease in recognition effect.

[0006] Regarding the material selection for the structure used for droplet generation, patent CN 209619360 U proposes a structure with one unit or two or more identical units, which is used to collect the generated droplets and perform PCR amplification reactions. The structure is injection molded with polycarbonate (PC) plastic or cycloolefin copolymer (COC) material; patent CN103451088A proposes a multi-way injection port structure to achieve the generation of digital PCR reactions. However, none of the above schemes consider the impact of droplet size differences on digital PCR structure recognition, which is not conducive to reliable droplet generation, and it is difficult to ensure the accuracy of subsequent droplet recognition. Summary of the invention

[0007] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a digital PCR droplet identification method and system based on artificial intelligence algorithm, which can improve the uniformity and stability of droplets and can efficiently and accurately identify digital PCR droplets.

[0008] The purpose of the present invention can be achieved by the following technical solution: A digital PCR droplet recognition method based on artificial intelligence algorithm, comprising the following steps:

[0009] S1, using a microfluidic chip to generate droplets, placing the microfluidic chip in a PCR device to perform PCR reaction cycles;

[0010] S2. After the PCR reaction cycle is completed, the microfluidic chip is placed under a microscope for observation to capture the PCR droplet image;

[0011] S3. Preprocess the PCR droplet image, input the preprocessed PCR droplet image into a pre-built droplet recognition model, and output the PCR droplet detection result.

[0012] Furthermore, the microfluidic chip includes an oil phase injection port, a water phase injection port, a droplet size adjustment channel, and a droplet collection chamber. The oil phase injection port is used to inject silicone oil, the water phase injection port is used to inject a mixture of PCR samples and reaction reagents, the droplet size adjustment channel is used to adjust the size of generated oil-in-water droplets to obtain droplets of uniform size, and the droplet collection chamber is used to collect and contain droplets.

[0013] Furthermore, the oil phase injection port, the water phase injection port, and the droplet size adjustment channel are all provided with micro-column arrays. The micro-column arrays in the oil phase injection port and the water phase injection port are used to eliminate bubbles in the injected liquid, and the micro-column array in the droplet size adjustment channel is used to adjust the sizes of multiple droplets so that the sizes of the droplets are consistent.

[0014] Furthermore, the microcolumn array in the oil phase injection port and the water phase injection port includes a plurality of cylinders arranged in a trapezoidal shape, the diameters of the cylinders in the same row are consistent, the diameters of the cylinders located at the bottom edge of the trapezoid to the cylinders located at the top edge of the trapezoid are in a decreasing relationship from row to row, and the center spacing between the cylinders in the same row is 3 times the radius of the cylinder.

[0015] Furthermore, in the microcolumn array in the oil phase injection port and the water phase injection port, the diameter of the cylinder located at the bottom of the trapezoid is 50-70 μm, and the diameter of the cylinder located at the top of the trapezoid is 20-45 μm.

[0016] Furthermore, the micro-column array in the droplet size adjustment channel includes a plurality of cylinders arranged in a rectangular shape, the diameters of the plurality of cylinders are consistent, and the distances between the centers of the cylinders are consistent.

[0017] Furthermore, in the micro-column array in the droplet size adjustment channel, the diameter of the cylinder is 20 to 80 μm.

[0018] Furthermore, the process of preprocessing the PCR droplet image in step S3 includes:

[0019] For the PCR droplet images taken by the microscope, the images were read based on the cv2 database and converted into grayscale images in RGB three-channel format. Then, Gaussian blur technology was used to reduce the basic noise in the image, and the droplets were processed using the threshold segmentation method to extract the contour information.

[0020] Furthermore, the process of constructing the droplet recognition model in step S3 includes:

[0021] S31. Build a neural network model architecture, including:

[0022] The first convolutional layer has 32 3x3 convolution kernels, a stride of 1, and a ReLU activation function.

[0023] The first pooling layer is max pooling with a 2x2 pooling window and a stride of 2.

[0024] The second convolutional layer has 64 3x3 convolution kernels, a stride of 1, and a ReLU activation function.

[0025] The second pooling layer uses maximum pooling, a 2x2 pooling window, and a stride of 2.

[0026] Flatten layer, used to flatten multidimensional input into one dimension;

[0027] The first fully connected layer has 64 neurons and uses the ReLU activation function;

[0028] The second fully connected layer has 1 neuron and uses Sigmoid activation function for binary classification output;

[0029] S32, collecting sample data, collecting multiple droplet images captured from the microscope, and performing grayscale conversion, Gaussian blur, threshold segmentation, and independent labeling processing on each image;

[0030] S33, dividing the sample data processed in step S32 into a training set, a validation set and a test set, for training and optimizing the neural network model architecture, and constructing a droplet recognition model.

[0031] Furthermore, the marking process in step S32 includes:

[0032] Mark the location and contour of the droplets;

[0033] Classify and label the droplets as positive or negative based on their morphology or color characteristics;

[0034] Unrecognizable or defective droplets are marked as exceptions.

[0035] Furthermore, the PCR droplet detection result outputted in step S3 includes the number of droplets and the proportion of positive droplets.

[0036] A digital PCR droplet recognition system based on artificial intelligence algorithm, comprising a liquid injection device, a microfluidic chip, a microscope and a computer, wherein the liquid injection device is used to inject silicone oil, PCR samples and reaction reagents into the microfluidic chip respectively;

[0037] The microfluidic chip is used to generate PCR droplets with stable structure;

[0038] The microscope is used to capture PCR droplet images;

[0039] The computer is provided with an image processing module and a pre-built droplet recognition model. The image processing module is used to perform grayscale conversion, Gaussian blur and threshold segmentation processing on the PCR droplet image to obtain droplet contour information. The droplet recognition model predicts and calibrates the droplet contour information and outputs the PCR droplet detection result.

[0040] Compared with the prior art, the present invention has the following advantages:

[0041] The present invention utilizes a microfluidic chip to generate droplets, which can ensure the generation of droplets with stable structure and consistent size, and is beneficial to accurately confine the PCR reaction inside a tiny droplet; by preprocessing the PCR droplet image captured by a microscope, the background noise can be effectively reduced, so that the droplet contour in the image is clearly discernible; and the preprocessed PCR droplet image is then input into a pre-constructed droplet recognition model to output the PCR droplet detection result, thereby realizing efficient recognition of the number of droplets and intelligent judgment of thresholds based on artificial intelligence technology, which can significantly improve the efficiency and accuracy of droplet recognition.

[0042] The microfluidic chip designed by the present invention includes an oil phase injection port, a water phase injection port, a droplet size adjustment channel, and a droplet collection chamber, wherein the oil phase injection port, the water phase injection port, and the droplet size adjustment channel are all provided with micro-column arrays, the micro-column arrays in the oil phase injection port and the water phase injection port are used to eliminate bubbles in the injected liquid, and the micro-column array in the droplet size adjustment channel is used to adjust the sizes of multiple droplets so that the sizes of the droplets are consistent. Thus, the generation and size adjustment of droplets are accurately controlled by introducing the micro-column array, so that the sizes of the droplets finally generated have a certain consistency.

[0043] The present invention designs the microcolumn array in the oil phase injection port and the water phase injection port to include a plurality of cylinders arranged in a trapezoidal shape, the diameters of the cylinders in the same row are consistent, the diameters of the cylinders located at the bottom edge of the trapezoid to the cylinders located at the top edge of the trapezoid are in a decreasing relationship from row to row, and the center spacing between the cylinders in the same row is 3 times the radius of the cylinder; the microcolumn array in the droplet size adjustment channel is also designed to include a plurality of cylinders arranged in a rectangular shape, the diameters of the plurality of cylinders are consistent, and the center spacing between the cylinders is consistent. This structural design takes into account the impact of the difference in droplet size on the recognition efficiency, which is conducive to improving the accuracy of subsequent droplet recognition.

[0044] The present invention performs preprocessing on the PCR droplet image captured by the microscope, including conversion to grayscale, application of Gaussian blur and threshold segmentation. Among them, grayscale conversion helps to remove the interference of color information and simplify data; Gaussian blur is used to smooth the image, reduce noise, and make the outline of the droplet more distinct; threshold segmentation technology further separates the droplet from the background, thereby extracting accurate droplet outline information.

[0045] In the process of constructing the droplet recognition model, the present invention uses an independent annotation method to annotate the collected images: mark the position and outline of each droplet, and at the same time, classify and annotate the droplets as positive or negative. In addition, droplets that are difficult to identify or have defects in the image are individually marked as abnormal. This can ensure the accuracy of the annotation and the reliability of the subsequent model construction.

[0046] The present invention sets a second fully connected layer in the architecture of the droplet recognition model for binary classification output, and in the sample data, pre-classifies and labels the droplets as positive or negative based on the morphological or color features of the droplets in the image, so that the subsequent droplet recognition results not only include the number of droplets, but also the positive and negative classification results of the droplets, thereby improving the specificity and sensitivity of the detection and ensuring the accuracy and reliability of the digital PCR analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0048] Figure 2 Schematic diagram of the structure of the microfluidic chip in the present invention;

[0049] Figure 3 Schematic diagram of the micro-pillar array of the oil phase injection port in the microfluidic chip;

[0050] Figure 4 Schematic diagram of a microcolumn array for the water phase injection port in a microfluidic chip;

[0051] Figure 5 Schematic diagram of a micropillar array that adjusts the flow path for droplet size in a microfluidic chip;

[0052] Figure 6 This is a schematic diagram of the droplet recognition process;

[0053] Figure 7 This is a schematic diagram of the neural network structure of the droplet recognition model;

[0054] Figure 8 It is a schematic diagram of the system structure of the present invention;

[0055] Markings in the figure: 1. Oil phase injection port, 2. Water phase injection port, 3. Droplet size adjustment channel, 4. Droplet collection chamber. DETAILED DESCRIPTION

[0056] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] Example

[0058] like Figure 1 As shown, a digital PCR droplet recognition method based on artificial intelligence algorithm includes the following steps:

[0059] S1, using a microfluidic chip to generate droplets, placing the microfluidic chip in a PCR device to perform a PCR reaction cycle;

[0060] S2. After the PCR reaction cycle is completed, the microfluidic chip is placed under a microscope for observation to capture the PCR droplet image;

[0061] S3. Preprocess the PCR droplet image, input the preprocessed PCR droplet image into a pre-built droplet recognition model, and output the PCR droplet detection result.

[0062] The microfluidic chip designed in this scheme is Figure 2 As shown, it includes an oil phase injection port 1, a water phase injection port 2, a droplet size adjustment channel 3, and a droplet collection chamber 4 with a support structure, wherein the oil phase injection port 1 is used to inject silicone oil, the water phase injection port 2 is used to inject a mixture of PCR samples and reaction reagents, the droplet size adjustment channel 3 is used to adjust the size of the generated oil-in-water droplets to obtain droplets of uniform size, and the droplet collection chamber 4 is used to collect and contain droplets. The oil phase injection port 1, the water phase injection port 2, and the droplet size adjustment channel 3 are all provided with micro-column arrays, the micro-column arrays in the oil phase injection port 1 and the water phase injection port 2 are used to eliminate bubbles in the injected liquid, and the micro-column array in the droplet size adjustment channel 3 is used to adjust the size of multiple droplets so that the sizes of the droplets are consistent.

[0063] like Figure 3 and Figure 4 As shown, the microcolumn array in the oil phase injection port 1 and the water phase injection port 2 includes a plurality of cylinders arranged in a trapezoidal shape, the diameters of the cylinders in the same row are consistent, the diameters of the cylinders located at the bottom edge of the trapezoid to the cylinders located at the top edge of the trapezoid decrease row by row, and the center spacing between the cylinders in the same row is 3 times the radius of the cylinder.

[0064] like Figure 5 As shown, the micro-column array in the droplet size adjustment channel 3 includes a plurality of cylinders arranged in a rectangular shape, the diameters of the plurality of cylinders are consistent, and the center distances between the cylinders are consistent.

[0065] In actual application, the production process of microfluidic chips mainly relies on computer-aided design (CAD) tools for graphic design, and then the chip template is made by performing steps such as coating (SU8 3050), photolithography, and development on a four-inch silicon wafer through semiconductor processing technology. Subsequently, PDMS (Polydimethylsiloxane) is used for casting to prepare a microfluidic structure. The prepared PDMS structure needs to be bonded to a glass slide to form a closed fluid system. In addition, the drilling of the oil phase injection port 1 and the water phase injection port 2 are completed by a punch.

[0066] In step S3, if Figure 6 As shown, the PCR droplet image is first preprocessed, and the process includes:

[0067] For the PCR droplet images obtained by microscope, the images were read based on the cv2 database and converted into grayscale images in RGB three-channel format. Then, Gaussian blur technology was used to reduce the basic noise in the image, and the droplets were processed using the threshold segmentation method to extract the contour information.

[0068] The extracted contour information is then input into the droplet recognition model, and the PCR droplet detection result is output. The construction process of the droplet recognition model includes:

[0069] S31. Build a neural network model architecture, such as Figure 7 As shown, including:

[0070] The first convolutional layer has 32 3x3 convolution kernels, a stride of 1, and a ReLU activation function.

[0071] The first pooling layer is max pooling with a 2x2 pooling window and a stride of 2.

[0072] The second convolutional layer has 64 3x3 convolution kernels, a stride of 1, and a ReLU activation function.

[0073] The second pooling layer uses maximum pooling, a 2x2 pooling window, and a stride of 2.

[0074] Flatten layer, used to flatten multidimensional input into one dimension;

[0075] The first fully connected layer has 64 neurons and uses the ReLU activation function;

[0076] The second fully connected layer has 1 neuron and uses Sigmoid activation function for binary classification output;

[0077] S32, collecting sample data, collecting multiple droplet images captured from the microscope, and performing grayscale conversion, Gaussian blur, threshold segmentation, and independent labeling processing on each image (marking the position and contour of the droplet; classifying and labeling the droplet as positive or negative based on the morphology or color characteristics of the droplet; and marking the droplets that cannot be identified or have defects as abnormal);

[0078] S33, dividing the sample data processed in step S32 into a training set, a validation set and a test set, for training and optimizing the neural network model architecture, and constructing a droplet recognition model.

[0079] The final output of the PCR droplet detection results includes the number of droplets and the proportion of positive droplets.

[0080] This proposal also proposes a digital PCR droplet recognition system based on artificial intelligence algorithm, such as Figure 8As shown, it includes a liquid injection device, a microfluidic chip, a microscope and a computer, wherein the liquid injection device is used to inject silicone oil, PCR samples and reaction reagents into the microfluidic chip respectively;

[0081] Microfluidic chips are used to generate structurally stable PCR droplets;

[0082] A microscope was used to capture images of PCR droplets;

[0083] The computer is provided with an image processing module and a pre-built droplet recognition model. The image processing module is used to perform grayscale conversion, Gaussian blur and threshold segmentation on the PCR droplet image to obtain droplet contour information. The droplet recognition model predicts and calibrates the droplet contour information and outputs the PCR droplet detection result.

[0084] In the present embodiment, in the process of PCR detection of microfluidic chip, it is first necessary to use precision instruments such as injection pumps (i.e., injection devices) to inject silicone oil as the oil phase and the aqueous phase containing reagents such as PCR template reaction solution into the corresponding injection ports (i.e., oil phase injection port 1 and aqueous phase injection port 2) respectively. The purpose of this step is to form stable droplets in the microfluidic chip to lay the foundation for subsequent PCR reactions. When it is observed through a microscope that the droplet collection end has collected full droplets, the entire microfluidic chip is placed in a commercial PCR device for PCR reaction cycles. After the PCR cycle is completed, the microfluidic chip is placed under a microscope for observation. This step is to check the effect of the PCR reaction and confirm whether the DNA has been successfully amplified. Finally, the observed droplet image is sent to the computer, and the corresponding test results can be obtained by executing a preset program for analysis.

[0085] In terms of droplet generation and processing, this solution adopts an innovative microfluidic structure, which accurately controls the generation and size adjustment of droplets by introducing a microcolumn array. In practical applications, the microcolumn array can be a cylinder or an irregular body. This embodiment adopts a cylinder. Specifically, two microcolumn arrays are provided at the injection port, which are mainly used to support the chamber of the microfluidic structure and play a role in removing bubbles to a certain extent. In terms of size design, the diameter of these array structures decreases by 5μm from the injection port to the inside of the pipeline, and the outermost diameter is between 50 and 70μm, and this embodiment is designed to be 60μm. The innermost diameter is between 20 and 45μm, and this embodiment is designed to be 30μm. The combination of each cylinder is in the form of a trapezoidal array, and the diameter of each row of the array is consistent, and the distance between the centers of the columns in the same row is 3 times the radius. There are differences in the diameters of different rows. The number of columns at the injection port is about 5 to 6 columns. The number of each row is adjusted according to the size of the pipeline.

[0086] The main function of the microcolumn array at the droplet generation site is not only to support the microfluidic structure, but also to adjust the droplet size. First, a cylinder with a bottom diameter of 20 to 80 μm is designed. The bottom diameter of this embodiment is designed to be 50 μm. It is generated in a rectangular array, and the spacing between the outer walls of the microcolumns is between 25 and 140 μm. This embodiment is designed to be 50 μm (that is, the center spacing between the cylinders is 75 μm). During the generation process of the droplets, their size will change with the difference in flow rate between the water phase and the oil phase. Droplets with a diameter greater than 50 μm will deform after passing through these carefully designed arrays, and the size of multiple droplets will be adjusted to ensure that the size between the droplets has a certain consistency. The size of the generated oil-in-water droplets is adjusted using a microcolumn array, so that the amount of templates and reaction reagents contained in the PCR reaction system is more uniform, thereby reducing the impact on the reaction system during the liquid generation process, ensuring the uniformity and stability of the droplets, which is crucial to improving the accuracy of dPCR. This structural design takes into account the impact of differences in droplet size on recognition efficiency, thereby overcoming some limitations of existing technologies.

[0087] In addition, in terms of identifying droplets, in order to achieve accurate identification of the number of droplets and reduce the interference of background noise, this program proposes an artificial intelligence-based program that processes the image data captured by the microscope through a series of carefully designed steps. First, for the image data obtained by the microscope, the cv2 (Open Source Computer Vision Library) database is used to read the image and convert the image into a grayscale image in RGB format to simplify the data and remove unnecessary color information. Next, the image noise is reduced by Gaussian blur technology, and then the threshold segmentation method is applied to identify the outline of the droplet. The identified outline will be input into a pre-trained neural network model (i.e., the droplet recognition model) for further prediction and calibration to improve the accuracy of recognition. Finally, the program can output the number of droplets detected and the proportion of positive droplets.

[0088] In this embodiment, the structure of the neural network model includes multiple convolutional layers and pooling layers, as well as fully connected layers. The model extracts features through 32 3x3 convolution kernels of convolution layer 1 and 64 3x3 convolution kernels of convolution layer 2, and each convolution layer is followed by a ReLU activation function. The two pooling layers are responsible for reducing the spatial dimension of the features and reducing the computational complexity for the subsequent fully connected layers. The flattening layer flattens the multidimensional features into one dimension, and the fully connected layer 1 contains 64 neurons, and the ReLU activation function is used again. Finally, the fully connected layer 2 performs a binary classification output through a neuron and a Sigmoid activation function to distinguish between positive and negative droplets.

[0089] In order to train this model, a high-quality dataset needs to be constructed, which involves a manual labeling process. First, data is collected from images taken with a microscope, and the image quality is ensured to meet the requirements and the droplet outline is clearly visible. The image data should be systematically organized according to the experimental conditions, and source data such as shooting time, droplet size, and light source conditions should be recorded. This information provides the necessary background for subsequent image processing and annotation.

[0090] Subsequently, the image is subjected to key preprocessing steps, including conversion to grayscale, application of Gaussian blur and threshold segmentation. Grayscale conversion helps to remove the interference of color information and simplify the data; Gaussian blur is used to smooth the image, reduce noise, and make the outline of the droplet more distinct; threshold segmentation technology further separates the droplet from the background, laying the foundation for accurate labeling. Image preprocessing is performed to reduce noise and highlight the droplet outline.

[0091] During the manual labeling stage, the operator uses the labeling tool to accurately mark the position and outline of each droplet, and classifies and labels them as positive or negative based on the morphology or color characteristics of the droplets. Droplets that are difficult to identify or defective should be individually marked as abnormal data for subsequent screening and processing. To ensure the accuracy of the labeling, each image is independently labeled by multiple operators, and consistency is ensured through cross-validation. After the labeling is completed, all data is saved in a unified format, such as XML or JSON files, which contain the image file name, the number of droplets, the outline coordinates of each droplet, and the classification label. Files in this format facilitate rapid access and use of neural network models during training, providing accurate labeled data for model training.

[0092] In order to meet the needs of model training, the data set also needs to be divided into training sets, validation sets, and test sets to ensure that the model can effectively learn and generalize. Data enhancement techniques, such as rotation, scaling, flipping, etc., can further expand the scale of the data set, increase data diversity, prevent model overfitting, and improve the robustness of the model in practical applications. Through these steps, a high-quality annotated data set can be created, laying a solid foundation for the training and optimization of neural network models.

[0093] In summary, this solution combines advanced image processing technology and neural network models with innovative microfluidic structure design to provide a more efficient, accurate and reliable dPCR solution. This not only improves the performance of PCR droplet detection, but also provides more powerful technical support for clinical diagnosis and biological research, and is expected to play an important role in the fields of precision medicine and molecular diagnosis.

Claims

1. A digital PCR droplet recognition method based on artificial intelligence algorithm, characterized in that: The following steps are involved: S1, using a microfluidic chip to generate droplets, placing the microfluidic chip in a PCR device to perform a PCR reaction cycle; S2. After the PCR reaction cycle is completed, the microfluidic chip is placed under a microscope for observation to capture the PCR droplet image; S3. Preprocess the PCR droplet image, input the preprocessed PCR droplet image into a pre-built droplet recognition model, and output the PCR droplet detection result.

2. According to claim 1, a digital PCR droplet identification method based on artificial intelligence algorithm is characterized in that: The microfluidic chip comprises an oil phase injection port, a water phase injection port, a droplet size adjustment channel, and a droplet collection chamber. The oil phase injection port is used to inject silicone oil, the water phase injection port is used to inject a mixture of PCR samples and reaction reagents, the droplet size adjustment channel is used to adjust the size of generated oil-in-water droplets to obtain droplets of uniform size, and the droplet collection chamber is used to collect and contain droplets.

3. A digital PCR droplet identification method based on artificial intelligence algorithm according to claim 2, characterized in that: The oil phase injection port, the water phase injection port, and the droplet size adjustment channel are all provided with microcolumn arrays. The microcolumn arrays in the oil phase injection port and the water phase injection port are used to eliminate bubbles in the injected liquid, and the microcolumn array in the droplet size adjustment channel is used to adjust the sizes of multiple droplets so that the sizes of the droplets are consistent.

4. The digital PCR droplet identification method based on artificial intelligence algorithm according to claim 3 is characterized in that: The microcolumn arrays in the oil phase injection port and the water phase injection port include a plurality of cylinders arranged in a trapezoidal shape, the diameters of the cylinders in the same row are consistent, the diameters of the cylinders located at the bottom edge of the trapezoid to the cylinders located at the top edge of the trapezoid are in a decreasing relationship row by row, and the center spacing between the cylinders in the same row is 3 times the radius of the cylinder.

5. The digital PCR droplet identification method based on artificial intelligence algorithm according to claim 3 is characterized in that: The micro-column array in the droplet size adjustment channel includes a plurality of cylinders arranged in a rectangular shape, the diameters of the plurality of cylinders are consistent, and the center distances between the cylinders are consistent.

6. The digital PCR droplet identification method based on artificial intelligence algorithm according to claim 1 is characterized in that: The process of preprocessing the PCR droplet image in step S3 includes: For the PCR droplet images taken by the microscope, the images were read based on the cv2 database and converted into grayscale images in RGB three-channel format. Then, Gaussian blur technology was used to reduce the basic noise in the image, and the droplets were processed using the threshold segmentation method to extract the contour information.

7. The digital PCR droplet identification method based on artificial intelligence algorithm according to claim 1 is characterized in that: The process of constructing the droplet recognition model in step S3 includes: S31. Build a neural network model architecture, including: The first convolutional layer has 32 3x3 convolution kernels, a stride of 1, and a ReLU activation function. The first pooling layer is max pooling with a 2x2 pooling window and a stride of 2. The second convolutional layer has 64 3x3 convolution kernels, a stride of 1, and a ReLU activation function. The second pooling layer uses maximum pooling, a 2x2 pooling window, and a stride of 2. Flatten layer, used to flatten multidimensional input into one dimension; The first fully connected layer has 64 neurons and uses the ReLU activation function; The second fully connected layer has 1 neuron and uses Sigmoid activation function for binary classification output; S32, collecting sample data, collecting multiple droplet images captured from the microscope, and performing grayscale conversion, Gaussian blur, threshold segmentation, and independent labeling processing on each image; S33, dividing the sample data processed in step S32 into a training set, a validation set and a test set, for training and optimizing the neural network model architecture, and constructing a droplet recognition model.

8. The method for digital PCR droplet identification based on artificial intelligence algorithm according to claim 7, characterized in that: The marking process in step S32 includes: Mark the location and contour of the droplets; Classify and label the droplets as positive or negative based on their morphology or color characteristics; Unrecognizable or defective droplets are marked as exceptions.

9. The method for digital PCR droplet identification based on artificial intelligence algorithm according to claim 8, characterized in that: The PCR droplet detection result outputted in step S3 includes the number of droplets and the proportion of positive droplets.

10. A digital PCR droplet recognition system based on artificial intelligence algorithm, characterized in that: It includes a liquid injection device, a microfluidic chip, a microscope and a computer, wherein the liquid injection device is used to inject silicone oil, PCR samples and reaction reagents into the microfluidic chip respectively; The microfluidic chip is used to generate PCR droplets with stable structure; The microscope is used to capture PCR droplet images; The computer is provided with an image processing module and a pre-built droplet recognition model. The image processing module is used to perform grayscale conversion, Gaussian blur and threshold segmentation processing on the PCR droplet image to obtain droplet contour information. The droplet recognition model predicts and calibrates the droplet contour information and outputs the PCR droplet detection result.

Images