A system for the examination and analysis of whole blood and blood components

Abstract

The application belongs to the technical field of blood transfusion, and particularly relates to a whole blood and component blood examination and analysis system, which comprises a rack, a controller, a camera, a forward light source, a bottom backlight source and a turbidity acquisition module, which are respectively arranged on the rack and connected with the controller; when in use, the forward light source irradiates, the camera acquires a forward light image of a blood bag, the bottom backlight source irradiates, the camera acquires a backlight image of the blood bag, and the forward light image and the backlight image are classified; the turbidity acquisition module acquires turbidity data of the blood bag, and the turbidity data is classified; and the results of the turbidity classification and the image classification are comprehensively classified to finally classify the whole blood and components. Compared with the prior art, the application starts from two dimensions of image classification and turbidity classification, jointly judges the condition of the blood bag, has higher accuracy and consistency, and is analyzed and judged by the controller, so that the efficiency is also higher.

Description

Technical Field

[0001] This invention belongs to the field of blood transfusion and collection technology, and particularly relates to a whole blood and component blood testing and analysis system. Background Technology

[0002] In the blood transfusion industry, it is sometimes necessary to examine and analyze whole blood and blood components. The current common practice is to use a simple light box or ordinary light as the light source for the collected whole blood and blood component bags, and to use manual methods in conjunction with auxiliary judgment tools such as color charts, and to judge the status of whole blood and blood component bags as "normal, slight, moderate or severe" based on human experience.

[0003] In the existing technology, the method of using auxiliary tools, light boxes and positive light sources to judge the results of images based on human experience requires a relatively high level of professionalism from the operators and workers. The judgment results may vary depending on the individual's interpretation, and the reliability, consistency and efficiency of the analysis and judgment cannot meet the requirements.

[0004] Table 1: Example of Manual Judgment in Existing Technologies

[0005]

[0006] Specifically, the manual judgment is made according to the table above (Table 1).

[0007] In view of this, the present invention aims to provide a whole blood and component blood examination and analysis system, which can accurately and quickly analyze the condition of whole blood and component blood through the combination of ingenious hardware structure and analysis methods. Furthermore, it can jointly determine the classification status and output the results through two dimensions: image classification and turbidity classification, thereby greatly improving the accuracy, efficiency and consistency of the analysis. Summary of the Invention

[0008] The purpose of this invention is to provide a whole blood and component blood testing and analysis system to address the shortcomings of existing technologies. Through the combination of ingenious hardware structure and analysis methods, it can accurately and quickly analyze the condition of whole blood and component blood. Furthermore, by using two dimensions—image classification and turbidity classification—it can jointly determine the classification status and output the results, thereby greatly improving the accuracy, efficiency, and consistency of the analysis.

[0009] To achieve the above objectives, the present invention adopts the following technical solution:

[0010] A whole blood and component blood testing and analysis system includes a rack, a controller, a camera respectively mounted on the rack and connected to the controller, a front light source, a bottom backlight source, and a turbidity acquisition module.

[0011] When in use, the camera captures a frontal light image of the blood bag under the illumination of a front light source, and a backlight image of the blood bag under the illumination of a bottom backlight source. The images are then classified using the frontal light image and the backlight image.

[0012] Turbidity data of blood bags is obtained through the turbidity acquisition module, and turbidity is classified.

[0013] The results of turbidity classification and image classification are combined to perform a final classification of whole blood and blood components.

[0014] As an improvement to the whole blood and component blood testing and analysis system of the present invention, the turbidity acquisition module includes a turbidity acquisition sensor, a turbidity acquisition light source and a downward sliding stage module. The turbidity acquisition sensor is disposed on one side of the bottom backlight light source, the turbidity acquisition light source is slidably disposed on the downward sliding stage module, and the turbidity acquisition light source is disposed above the turbidity acquisition sensor.

[0015] As an improvement to the whole blood and component blood testing and analysis system of the present invention, the lower sliding stage module includes a fixed base, a lead screw disposed on the fixed base, and a motor connected to the lead screw, and the turbidity acquisition light source is connected to the lead screw through a slider.

[0016] As an improvement to the whole blood and component blood testing and analysis system of the present invention, the steps for obtaining turbidity data of blood bags are as follows:

[0017] The turbidity acquisition light source is near-infrared light. The turbidity acquisition sensor acquires the light amount signal of near-infrared light passing through the blood plasma, which is converted into a voltage signal by the controller, then converted into a digital signal by the A / D converter, and then the digital signal is calibrated to ensure that the light transmittance turbidity data of the same blood plasma is consistent for each device. Finally, the controller automatically classifies the data according to the light transmittance.

[0018] As an improvement to the whole blood and component blood testing and analysis system of the present invention, five sets of transmissivity turbidity data are collected for each bag of plasma. The maximum and minimum abnormal values ​​are removed, and the remaining normal turbidity values ​​are classified into the corresponding classification intervals. This is turbidity classification.

[0019] As an improvement to the whole blood and component blood testing and analysis system of the present invention, the condition of the blood bag under test is classified according to the front light image and back light image acquired by the camera, which is called image classification.

[0020] As an improvement to the whole blood and component blood examination and analysis system of the present invention, the method for classifying blood plasma images includes the following steps:

[0021] The first step is to use image binarization and closing operation image preprocessing algorithms to find the location of blood plasma in the image when the camera is taking pictures.

[0022] Second, the controller extracts the color data features of the area where the plasma is located;

[0023] The third step involves the controller transmitting color data to a pre-trained AI model, which then provides a judgment result.

[0024] As an improvement to the whole blood and component blood examination and analysis system of the present invention, the pre-trained AI model uses a machine learning MLP model, and the steps for color recognition using MLP are as follows:

[0025] (1) Data preparation: Prepare a set of images labeled "normal", "mild", "moderate" and "severe" in advance, and then extract the color data features of blood bags and use the RGB data of the images as feature inputs;

[0026] (2) Design the MLP structure: Select an appropriate number of hidden layers and neurons, as well as an appropriate activation function;

[0027] (3) Training the network: Use labeled color data to train the MLP model so that it can recognize and distinguish different colors.

[0028] As an improvement to the whole blood and component blood testing and analysis system of this invention, the final classification method is as follows:

[0029] If the results of turbidity classification and image classification are the same, the controller directly outputs the judgment result.

[0030] If the turbidity classification falls within the overlapping area of ​​adjacent classifications, output the image classification result;

[0031] If the turbidity classification differs from the image classification, a pop-up window will prompt manual judgment.

[0032] Compared to existing technologies, this invention uses both image classification and turbidity classification to jointly determine the condition of blood bags, resulting in higher accuracy and consistency. Furthermore, the analysis and judgment are performed using a controller, which also improves efficiency. Attached Figure Description

[0033] Figure 1 This is one of the structural schematic diagrams of the hardware of this invention (the turbidity acquisition module is in a non-working state).

[0034] Figure 2 This is the second structural schematic diagram of the hardware of the present invention (the turbidity acquisition module is in working state).

[0035] Figure 3 This is a schematic diagram of the turbidity acquisition light source and the downward sliding stage module in this invention (the turbidity acquisition module is in a non-working state).

[0036] Figure 4This is a schematic diagram of the turbidity acquisition light source and the downward sliding stage module in this invention (the turbidity acquisition module is in working condition).

[0037] Figure 5 This is one of the cross-sectional structural schematic diagrams of the present invention.

[0038] Figure 6 This is the second cross-sectional structural schematic diagram of the present invention.

[0039] Figure 7 This is a three-dimensional structural diagram of the blood bag, infrared light source, and turbidity acquisition sensor in this invention.

[0040] Figure 8 This is a cross-sectional structural diagram of the blood bag, infrared light source, and turbidity acquisition sensor in this invention.

[0041] Figure 9 This invention relates to the self-developed data classification and analysis of whole blood and component blood collection. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0043] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of the components in a specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0044] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. If the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0045] like Figures 1 to 8As shown, a whole blood and component blood testing and analysis system includes a frame (not shown), a controller (not shown), a camera 1 mounted on the frame and connected to the controller, a forward light source 2, a bottom backlight source 3, and a turbidity acquisition module 4. The camera 1 is an industrial camera with a lens at the front end. The bottom backlight source 3 is used to detect foreign matter inside normal component blood and lightly polluted whole blood and component blood. The forward light source 2 is used for AOI (Automated Optical Inspection) of normal, lightly, moderately, and severely polluted whole blood and component blood. The turbidity acquisition module 4 is used for acquiring turbidity data of normal, lightly, moderately, and severely polluted whole blood and component blood.

[0046] In use, under the illumination of the forward light source 2, the camera 1 captures the front light image of the blood bag 5, and under the illumination of the bottom backlight source 3, the camera 1 captures the backlight image of the blood bag 5. The images are classified by the front light image and the backlight image.

[0047] Turbidity data of blood bags are obtained through turbidity acquisition module 4, and turbidity is classified.

[0048] The results of turbidity classification and image classification are combined to perform a final classification of whole blood and blood components.

[0049] The turbidity acquisition module 4 includes a turbidity acquisition sensor 41, a turbidity acquisition light source 42, and a downward sliding stage module 43. The turbidity acquisition sensor 41 is located on one side of the bottom backlight source 3, and the turbidity acquisition light source 42 is slidably mounted on the downward sliding stage module 43, with the light source positioned above the sensor 41. When the blood bag 5 is placed on the bottom backlight source 3, the turbidity acquisition module 4 receives an acquisition command from the controller, and the downward sliding stage module 43 drives the turbidity acquisition light source 42 to move down to the acquisition position, where the controller acquires turbidity data. The turbidity acquisition sensor 41 and the turbidity acquisition light source 42 are set at a fixed distance, ensuring complete contact with the blood bag 5 and preventing light transmission. The turbidity acquisition light source 42 includes a light source 421 and a support 422.

[0050] The lower sliding stage module 43 includes a fixed base 431, a lead screw 432 mounted on the fixed base 431, and a motor 433 connected to the lead screw 432. The turbidity acquisition light source 42 is connected to the lead screw 432 via a slider 434.

[0051] In use, the blood bag 5 is placed on the platform of the bottom backlight source 3. Imaging is achieved through an industrial camera. The controller uses an embedded PC and a high-resolution terminal display. A visual algorithm classifies the images based on symptoms and symptom states. The PC's functions are briefly described as follows: it uses an x86 system; employs deep learning machine vision algorithms; acquires turbidity data; uses blood bag imaging and a turbidity data model to compare and judge the current test; and performs stand-alone storage and database queries. The high-resolution terminal display can show the following information: the current image of the blood bag; the computer's automatic judgment results; the display of identification information; and storage of the current examination results and the examination results corresponding to the current identity.

[0052] The UI of the acquisition device corresponds to three acquisition click buttons: 1) backlight image, 2) front light image, and 3) turbidity data. The basic process of this invention is as follows: (1) Power on the device; (2) The device completes self-test and turns on the front light and backlight; (3) Place the blood bag 5; (4) Click the start button on the inspection screen; (5) Take a picture of the backlight image; (6) Take a picture of the front light image; (7) Acquire turbidity data; (8) The device images and outputs the inspection results; (9) The device automatically saves the inspection results.

[0053] For each blood bag level, two images and one copy of turbidity data are collected and automatically saved to a preset storage path. The storage path is named independently using the current date as the folder name. For example, images and turbidity data collected on June 10, 2023, will be stored in the folder 20230610.dir. Within this folder, the images and turbidity data will be named using the current date, time, and assessment result. For example, images and turbidity data collected at 4:10 PM on June 10, 2023, will be named 202306101610-Mild. And so on.

[0054] To improve automation, a sorting robot can be added to sort the blood bags after they have been inspected.

[0055] The steps to obtain turbidity data from blood bags are as follows:

[0056] The turbidity acquisition light source 42 is near-infrared light. The turbidity acquisition sensor 41 acquires the light amount signal of near-infrared light passing through the blood plasma. The controller converts it into a voltage signal, and then converts it into a digital signal through an A / D converter. The digital signal is then calibrated to ensure that the turbidity data of the blood plasma transmitted through each device is consistent. Finally, the controller automatically classifies the data based on the amount of light transmitted.

[0057] Five sets of transmissivity turbidity data were collected from each bag of plasma. The largest and smallest outliers were removed, and the remaining normal turbidity values ​​were classified into the corresponding classification intervals. This is called turbidity classification.

[0058] Image AI filtering: The acquired images are automatically classified by finding the most similar images in the classification library. That is, the condition of the blood bags being tested is classified based on the front-lit and back-lit images acquired by the camera; this is image classification.

[0059] The method for classifying plasma images (AI classification of plasma images) includes the following steps:

[0060] The first step is to use image binarization and closing operation image preprocessing algorithms to find the location of blood plasma in the image when camera 1 takes a picture.

[0061] Second, the controller extracts color (RGB, CIE, HSV) data features of the region where the plasma is located;

[0062] The third step involves the controller transmitting color data to a pre-trained AI model, which then provides a judgment result.

[0063] The pre-trained AI model uses a machine learning MLP model. The steps for using MLP to recognize colors are as follows:

[0064] (1) Data preparation: Prepare a set of images labeled “normal”, “mild”, “moderate” and “severe” in advance (collect dozens to hundreds of images for each). Then, extract the color features of the blood bags and use the RGB data of the images as feature inputs. RGB (Red Green Blue) refers to the intensity values ​​of red, green and blue light in the color mode, which is usually used to describe and process color information in digital images.

[0065] (2) Designing the MLP Structure: Selecting an appropriate number of hidden layers and neurons, as well as a suitable activation function; this requires prior debugging to determine various parameters. In this MLP (Multilayer Perceptron) model design, we adopted a specific configuration to optimize the color recognition process. The model contains one hidden layer with 10 neurons, a configuration sufficient to handle the complexity of color data while maintaining the model's efficiency. We chose 'softmax' as the activation function for the output layer, making the model particularly suitable for handling classification problems with multiple output classes. Furthermore, to improve the model's ability to process input data, we used 'principal component analysis' as a preprocessing method. This method helps improve the accuracy and efficiency of model training and classification by reducing data dimensionality.

[0066] In this model, softmax, as the activation function of the output layer, transforms the raw outputs of the neurons into a probability distribution, representing the probability of each class. Specifically, the softmax function works as follows: 1. It receives the raw output value from each neuron in the output layer. 2. It calculates the exponent (e raised to a power) of each output value to ensure all outputs are positive. 3. It divides the exponential output of each neuron by the sum of the exponential outputs of all neurons. In this way, the softmax function ensures that the sum of all output values ​​in the output layer equals 1, allowing each output value to be interpreted as a probability. In multi-class classification problems, each output represents the probability that the input data belongs to a specific class. The class corresponding to the output with the highest probability is selected as the model's prediction.

[0067] Principal component analysis primarily simplifies data while preserving important information as much as possible. This helps reduce the demand for computational resources and removes noise and redundant information, thereby improving the training efficiency and predictive performance of the model.

[0068] It works through the following steps:

[0069] a. Standardizing data: First, standardize each feature in the dataset (e.g., each pixel in image processing), usually by subtracting its mean and dividing by its standard deviation.

[0070] b. Calculate the covariance matrix; next, calculate the covariance matrix of the data. The covariance matrix reveals the relationships between different features in the dataset.

[0071] c. Solve for eigenvalues ​​and eigenvectors; solve for eigenvalues ​​and corresponding eigenvectors using the covariance matrix. Eigenvectors represent the main directions of the data, while eigenvalues ​​represent the "importance" of these directions or the degree of dispersion of the data in these directions.

[0072] d. Select principal components; based on the magnitude of the eigenvalues, select the top N eigenvectors as principal components. These principal components capture the most important variance in the data.

[0073] e. Transform the data to a new space. Finally, the original data is transformed (or projected) into a new space composed of these principal components. This transformed data typically has lower dimensionality but still retains most of the important information from the original data.

[0074] (3) Training the network: Use labeled color data to train the MLP model so that it can recognize and distinguish different colors.

[0075] The Multilayer Perceptron (MLP) classifier is a basic neural network used for classification tasks. It consists of an input layer, multiple hidden layers, and an output layer. Data propagates forward through the network, extracting features through the hidden layers and generating classification results at the output layer. MLPs introduce non-linearity using activation functions, enabling the network to learn complex data patterns. During training, loss functions and optimization algorithms are used to adjust the network's weights and biases to improve classification accuracy.

[0076] MLP classification algorithm description: (English words are parameters)

[0077] An MLP consists of three layers: an input layer with NumInput input variables (units, neurons); a hidden layer with NumHidden units; and an output layer with NumOutput output variables. An MLP performs the following steps to process the input data x. i (So-called feature vectors) calculate the activation value z of the hidden unit. j :

[0078]

[0079] z j =tanh(a j )

[0080] j = 1, ..., n h

[0081] Here, matrix w ji And vector b j These are the weights of the input layer (first layer) of the MLP. In the hidden layer (second layer), the activation value z... j First, the transformation is performed using a linear combination of variables, similar to the method described above:

[0082]

[0083] k = 1, ..., n o

[0084] Here, matrix w kj And vector b k These are the weights of the second layer of the MLP.

[0085] The activation function used in the output layer can be determined by setting OutputFunction. For OutputFunction = 'linear', the data will simply be copied:

[0086] y k =a k

[0087] k = 1, ..., no

[0088] This type of activation function is suitable for regression problems (function approximation). It is not suitable for classification problems.

[0089] For OutputFunction='logistic', the activation value is calculated as follows:

[0090]

[0091] k = 1, ..., n o

[0092] This type of activation function is applied to classification problems with multiple (NumOutput) independent logical properties as outputs. In practice, this type of classification problem is relatively rare.

[0093] For OutputFunction='softmax', the activation value is calculated as follows:

[0094]

[0095] k = 1, ..., n o

[0096] This activation function is applicable to common classification problems with multiple mutually exclusive categories as output (NumOutput). Specifically, when using pixel data for classification, OutputFunction = 'softmax' must be used.

[0097] Statistical analysis of self-developed data on transillumination turbidity of whole blood and blood components revealed that the turbidity data corresponding to "normal," "mild," "moderate," and "severe" turbidity levels have their own intervals and overlaps. Figure 9 As shown.

[0098] The final classification method is as follows:

[0099] If the results of turbidity classification and image classification are the same, the controller directly outputs the judgment result.

[0100] If the turbidity classification falls within the overlapping area of ​​adjacent classifications, output the image classification result;

[0101] If the turbidity classification differs from the image classification, a pop-up window will prompt manual judgment.

[0102] As shown in the table below (Table 2):

[0103] Table 2: The method for classifying whole blood and blood components according to the present invention

[0104]

[0105] This invention uses both image classification and turbidity classification to jointly determine the condition of blood bags, resulting in higher accuracy and consistency. Furthermore, the analysis and judgment are performed using a controller, which also improves efficiency.

[0106] Based on the disclosure and teachings of the foregoing specification, those skilled in the art can make changes and modifications to the above embodiments. Therefore, the present invention is not limited to the specific embodiments disclosed and described above, and some modifications and changes to the present invention should also fall within the protection scope of the claims of the present invention. Furthermore, although some specific terms are used in this specification, these terms are only for convenience of explanation and do not constitute any limitation on the present invention.

Claims

1. A whole blood and component blood testing and analysis system, characterized in that: The system includes a frame, a controller, cameras respectively mounted on the frame and connected to the controller, a front light source, a bottom backlight source, and a turbidity acquisition module. When in use, the camera captures a frontal light image of the blood bag under the illumination of a front light source, and a backlight image of the blood bag under the illumination of a bottom backlight source. The images are then classified using the frontal light image and the backlight image. Turbidity data of blood bags is obtained through the turbidity acquisition module, and turbidity is classified. Based on the combined results of turbidity classification and image classification, whole blood and blood components are finally classified. The turbidity acquisition module includes a turbidity acquisition sensor, a turbidity acquisition light source, and a downward sliding stage module. The turbidity acquisition sensor is disposed on one side of the bottom backlight light source, and the turbidity acquisition light source is slidably disposed on the downward sliding stage module, and the turbidity acquisition light source is disposed above the turbidity acquisition sensor. The downward sliding stage module includes a fixed base, a lead screw mounted on the fixed base, and a motor connected to the lead screw. The turbidity acquisition light source is connected to the lead screw via a slider. The steps to obtain turbidity data from blood bags are as follows: The turbidity acquisition light source is near-infrared light. The turbidity acquisition sensor acquires the light amount signal of near-infrared light passing through the blood plasma, which is converted into a voltage signal by the controller, then converted into a digital signal by the A / D converter, and then the digital signal is calibrated to make the light transmittance turbidity data of the same blood plasma consistent for each device. Finally, the controller automatically classifies the data according to the light transmittance. Five sets of transmittance turbidity data were collected from each bag of plasma. The largest and smallest outliers were removed, and the remaining normal turbidity values ​​were classified into the corresponding classification intervals. This is the turbidity classification. The final classification method is as follows: If the results of turbidity classification and image classification are the same, the controller directly outputs the judgment result. If the turbidity classification falls within the overlapping area of ​​adjacent classifications, output the image classification result; If the turbidity classification and the image classification differ, a pop-up window will prompt manual judgment. The turbidity sensor and the turbidity light source are set at a fixed distance, and are completely in contact with the blood bag without being opaque. The method for classifying blood plasma images includes the following steps: The first step is to use image binarization and closing operation image preprocessing algorithms to find the location of blood plasma in the image when the camera is taking pictures. Second, the controller extracts the color characteristics of the area where the plasma is located; The third step is for the controller to transmit the color data to the pre-trained AI model, and the AI ​​model will provide a judgment result. The pre-trained AI model uses a machine learning MLP model. The steps for using MLP to recognize colors are as follows: (1) Data preparation: Prepare a set of images labeled "normal", "mild", "moderate" and "severe" in advance, and then extract the color data features of the blood bags and use the RGB data of the images as feature inputs; (2) Design the MLP structure: Select an appropriate number of hidden layers and neurons, as well as an appropriate activation function; (3) Training the network: Use labeled color data to train the MLP model so that it can recognize and distinguish different colors.

2. The whole blood and component blood testing and analysis system according to claim 1, characterized in that: The condition of the tested blood bags is classified based on the front-lit and back-lit images acquired by the camera; this is called image classification.

Images