Platelet function indicator evaluation method and apparatus

Abstract

Disclosed in the present invention are a platelet function indicator evaluation method and apparatus. The platelet function indicator evaluation method comprises: collecting a platelet to be tested; using an image and video device to collect swirling feature data of said platelet, wherein the swirling feature data comprises swirling feature image data and / or swirling feature video data; and evaluating a platelet function indicator on the basis of the swirling feature data of said platelet. The present invention can implement accurate, rapid, and low-cost testing of platelet function indicators, and solve the problems of long time consumption, high costs, and large result errors in conventional platelet function testing methods.

Description

A method and apparatus for evaluating platelet performance indicators Technical Field

[0001] This invention relates to the field of blood testing technology, and in particular to a method and apparatus for evaluating platelet performance indicators. Background Technology

[0002] Platelets are small cytoplasmic fragments formed by the degranulation of megakaryocytes. Through adhesion, aggregation, release, and contraction, they play a crucial role in hemostasis and coagulation. In clinical treatment, platelet transfusion is widely used to prevent or treat bleeding disorders caused by reduced platelet counts and functional abnormalities induced by trauma, surgery, leukemia, bone marrow suppression after radiotherapy or chemotherapy, aplastic anemia, infection, etc. Timely platelet transfusion can buy patients more treatment time, thereby improving prognosis and even saving lives. However, in recent years, platelet transfusion ineffectiveness has frequently occurred, seriously delaying patient treatment. In-depth research has found that defects in the quality of the transfused platelets are one of the main reasons. Therefore, performance evaluation of platelets before transfusion is particularly important.

[0003] Currently, clinical platelet performance testing mainly relies on traditional laboratory methods, such as platelet aggregation tests, thromboelastography, and dynamic monitoring of coagulation and platelet function. While existing platelet performance testing methods address some clinical needs to a certain extent, they still have the following problems:

[0004] (1) The results are not accurate enough. Traditional platelet performance testing requires multiple steps. If a problem occurs in any one of these steps, it will affect the accuracy of the entire result and fail to reflect the patient's true coagulation status, thus misleading subsequent treatment. Taking thromboelastography as an example, the reaction cup must first be manually loaded, and then blood sample, kaolin, and calcium ions are added separately. Then the reaction cup is pushed to the testing section, and the timing is started by pressing the button. The equipment should be kept absolutely still throughout the entire testing process. Even slight vibrations will affect the accuracy of the results or cause the cup to fall off, requiring a retest.

[0005] (2) Time-consuming. Traditional platelet performance testing methods typically take more than an hour per sample from pre-test preparation to result delivery, seriously delaying patient treatment. This is especially important for patients with acute massive bleeding who require immediate platelet transfusion, as well as for emergency surgery, acute coronary syndrome, and other situations requiring rapid assessment of platelet function. For example, when a hypocoagulable patient receives procoagulant therapy, they are very likely to change from hypocoagulability to hypercoagulability in a short period of time, resulting in coagulation dysfunction. Even the latest traditional test results can only reflect the patient's hypocoagulable state an hour ago and cannot detect hypercoagulability in real time, seriously misleading subsequent treatment. This is not only detrimental to the patient's recovery but can even lead to death.

[0006] (3) Excessive cost. Traditional testing requires professional personnel, equipment, reagents, consumables, etc., resulting in high costs during use. For example, a routine thromboelastography test costs more than 300 RMB, while a platelet cup monitoring system costs more than 1,000 RMB. This not only increases patients' medical expenses but also further burdens the national medical insurance system.

[0007] In summary, traditional platelet performance testing methods suffer from inaccurate results, are time-consuming, and costly, failing to meet the practical needs for accurate, rapid, and cost-effective evaluation of platelet performance before transfusion. Summary of the Invention

[0008] This invention overcomes the shortcomings of the prior art and provides a method and device for evaluating platelet performance indicators, which can accurately, quickly and at low cost detect platelet performance indicators, solving the problems of inaccurate results, long time consumption and high cost of traditional platelet performance testing.

[0009] According to one aspect of the present invention, a method for evaluating platelet performance indicators is proposed, the method comprising:

[0010] Collect platelets for testing;

[0011] The cloud-like feature data of the platelets to be detected is acquired using image and video equipment, and the cloud-like feature data includes cloud-like feature image data and / or cloud-like feature video data;

[0012] Platelet performance indicators were evaluated using the cloud-like characteristic data of the platelets to be tested.

[0013] In one possible implementation, evaluating platelet performance indicators using the cloud-like feature data of the platelets to be detected includes:

[0014] Frequency domain analysis and gradient analysis are performed on the cloud-like feature data of each frame of platelets to be detected to obtain a comprehensive evaluation index of the cloud-like feature data of platelets to be detected.

[0015] The duration of the cloud-like feature of the platelets to be detected is recorded to obtain the duration of the cloud-like feature data of the platelets to be detected.

[0016] Based on the comprehensive evaluation index and duration of the cloud-like feature data of the platelets to be tested, the significance features of the cloud-like feature data of the platelets to be tested are obtained.

[0017] The platelet performance indicators are evaluated based on the significant characteristics of the cloud-like feature data of the platelets to be tested and the correspondence between them and the platelet performance indicators.

[0018] In one possible implementation, the comprehensive evaluation index of the cloud-like feature data of the platelets to be detected includes the spectral energy ratio, the average gradient amplitude, and the stripe direction score.

[0019] In one possible implementation, the evaluation of platelet performance indicators based on the correspondence between the saliency features of the cloud-like feature data of the platelets to be detected and the platelet performance indicators includes:

[0020] Determine the most significant feature of the cloud-like characteristic data of the platelets to be tested;

[0021] The performance of platelets is evaluated based on the correspondence between the most significant feature of the cloud-like characteristic data of the platelets to be tested and the platelet performance index level.

[0022] In one possible implementation, the platelet performance indicators are divided into different levels based on the platelet's hemostatic function.

[0023] In one possible implementation, evaluating the platelet performance indicators using the cloud-like feature data of the platelets to be detected further includes:

[0024] The cloud-like feature data of the platelets to be detected is input into the trained platelet performance index evaluation model PEM to obtain platelet performance index evaluation data.

[0025] In one possible implementation, the platelet performance index evaluation model (PEM) is trained based on platelet performance index data and platelet cloud-like feature data.

[0026] In one possible implementation, the platelet performance index evaluation model (PEM) is trained based on platelet performance index data and platelet cloud-like feature data, including:

[0027] Collect platelets for training;

[0028] Cloud-like feature data of training platelets are collected using image and video equipment, and the cloud-like feature data includes cloud-like feature image data and / or cloud-like feature video data;

[0029] Detect performance metrics data of platelets used in training;

[0030] The cloud-like feature data of the training platelets are preprocessed to obtain cloud-like feature preprocessed data of the training platelets that meet the input requirements of the platelet performance index evaluation model (PEM).

[0031] The cloud-like feature preprocessed data of the training platelets is input into the platelet performance index evaluation model PEM to obtain the predicted data of the performance index of the training platelets.

[0032] Based on the platelet performance index detection data and prediction data used for training, a platelet performance index evaluation model PEM is trained.

[0033] In one possible implementation, the preprocessing of the cloud-like feature data of training platelets to obtain cloud-like feature preprocessed data of training platelets that meets the input requirements of the Platelet Performance Evaluation Model (PEM) includes:

[0034] Keyframe images are extracted from the cloud-like feature video data of the training platelets;

[0035] K-means clustering algorithm was used to perform cluster analysis on keyframe images to obtain cloud-like feature keyframe images of training platelets with different cluster centers;

[0036] The cloud-like feature keyframe images of training platelets for different cluster centers are subjected to denoising, resolution adjustment, image enhancement and normalization to obtain cloud-like feature keyframe images of training platelets.

[0037] The cloud-like feature keyframe image of the training platelets is the cloud-like feature preprocessed data of the training platelets that meets the input requirements of the platelet performance evaluation model (PEM).

[0038] In one possible implementation, the platelet performance evaluation model (PEM) includes a convolutional neural network (CNN) and a recurrent neural network (RNN).

[0039] In one possible implementation, the step of inputting the preprocessed data of the cloud-like features of training platelets into the platelet performance index evaluation model (PEM) to obtain predicted data of the platelet performance index for training; and training the platelet performance index evaluation model (PEM) based on the detected and predicted data of the platelet performance index for training, includes:

[0040] P1: Input the keyframe image of the cloud-like feature of the training platelets into the convolutional neural network CNN of the platelet performance index evaluation model PEM to obtain the keyframe image features of the cloud-like feature of the training platelets.

[0041] P2: Input the keyframe image features of the cloud-like characteristics of the training platelets into the recurrent neural network (RNN) and output the predicted data of the performance indicators of the training platelets.

[0042] P3: Based on the platelet performance index detection data and prediction data used for training, calculate the loss value of the platelet performance index evaluation model PEM;

[0043] P4: Update the weight parameters and bias parameters of the platelet performance index evaluation model PEM using the loss value and gradient backpropagation algorithm.

[0044] P5: Repeat the process from P1 to P4 until the preset number of training rounds are reached to train and obtain the platelet performance index evaluation model (PEM).

[0045] In one possible implementation, the platelet performance data includes one or more of the following: adhesion rate, maximum aggregation rate, platelet factor 4 (PF4) concentration, clot retraction rate, time to carotid artery thrombosis in mice, time to tail bleeding in mice, percentage of platelets transfused into the body, phosphatidylserine (PS) positivity rate, P-selectin (CD62P) positivity rate, and mean fluorescence intensity of platelet glycoprotein Ibα (CD42b).

[0046] According to another aspect of the present invention, a platelet performance index evaluation device is provided, the device comprising:

[0047] The collection device is used to collect platelets for testing;

[0048] An image and video acquisition device is used to acquire cloud-like feature data of the platelets to be detected, wherein the cloud-like feature data includes cloud-like feature image data and / or cloud-like feature video data;

[0049] The evaluation module is used to evaluate the platelet performance indicators using the cloud-like feature data of the platelets to be tested.

[0050] According to another aspect of the present invention, an electronic device is provided, the device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described above.

[0051] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0052] The platelet performance evaluation method of the present invention involves collecting platelets to be tested; acquiring cloud-like feature data of the platelets using an image and video device, wherein the cloud-like feature data includes cloud-like feature image data and / or cloud-like feature video data; and evaluating platelet performance indicators using the cloud-like feature data of the platelets to be tested. This method can accurately, quickly, and cost-effectively detect platelet performance indicators, solving the problems of inaccurate results, long testing times, and high costs associated with traditional platelet performance testing.

[0053] Other optional features and technical effects of the embodiments of the present invention are partly described below and partly apparent from reading this document. Attached Figure Description

[0054] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. The elements shown are not limited to the scale shown in the drawings, and the same or similar reference numerals in the drawings denote the same or similar elements, wherein:

[0055] Figure 1 shows a flowchart of a platelet performance index evaluation method according to an embodiment of the present invention;

[0056] Figure 2 shows a schematic diagram of the cloud-like features of platelets according to an embodiment of the present invention;

[0057] Figure 3 shows a flowchart of step S3 in a platelet performance index evaluation method according to an embodiment of the present invention.

[0058] Figure 4 shows a schematic diagram of the frequency domain spectrum of a cloud-like feature image of platelets according to an embodiment of the present invention;

[0059] Figure 5 shows a schematic diagram of the gradient amplitude of a cloud-like feature image of platelets according to an embodiment of the present invention;

[0060] Figure 6 shows a flowchart of the platelet performance index evaluation model PEM training method according to another embodiment of the present invention;

[0061] Figure 7a shows a scatter plot of the true and predicted values ​​of platelet performance index - adhesion rate according to an embodiment of the present invention;

[0062] Figure 7b shows a scatter plot of the actual and predicted values ​​of the platelet performance index - maximum aggregation rate - according to an embodiment of the present invention;

[0063] Figure 7c shows a scatter plot of the true and predicted values ​​of platelet performance index - PF4 concentration - according to an embodiment of the present invention;

[0064] Figure 7d shows a scatter plot of the true and predicted values ​​of the platelet performance index - clot shrinkage rate - according to an embodiment of the present invention;

[0065] Figure 7e shows a scatter plot of the true and predicted values ​​of platelet performance index - mouse carotid artery thrombosis time according to an embodiment of the present invention;

[0066] Figure 7f shows a scatter plot of the actual and predicted values ​​of platelet performance index - mouse tail clipping bleeding time - according to an embodiment of the present invention;

[0067] Figure 7g shows a scatter plot of the actual and predicted values ​​of platelet performance indicators—the percentage of platelets introduced into the body—according to an embodiment of the present invention.

[0068] Figure 7h shows a scatter plot of the true and predicted values ​​of the platelet performance index - PS positivity rate - according to an embodiment of the present invention;

[0069] Figure 7i shows a scatter plot of the true and predicted values ​​of the platelet performance index - CD62P positivity rate - according to an embodiment of the present invention;

[0070] Figure 7j shows a scatter plot of the true and predicted values ​​of the average fluorescence intensity of CD42b, a platelet performance index, according to an embodiment of the present invention.

[0071] Figure 8 shows a structural diagram of a platelet performance index evaluation device based on the cloud-like characteristics of platelets according to an embodiment of the present invention.

[0072] Figure 9 shows a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0073] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0074] The term "comprising" and its variations as used herein signify open inclusion, i.e., "including but not limited to". Unless otherwise stated, the term "or" means "and / or". The term "based on" means "at least partially based on". The terms "one example embodiment" and "one embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first", "second", etc., may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0075] Furthermore, the steps illustrated in the flowcharts of the accompanying drawings can be executed in a computer, such as a set of computer-executable instructions. Also, although a logical order is shown in the flowcharts, in some cases the steps shown or described may be performed in a different order than that presented here.

[0076] Figure 1 shows a flowchart of a platelet performance index evaluation method according to an embodiment of the present invention. As shown in Figure 1, the method may include:

[0077] Step S1: Collect platelets to be tested.

[0078] Platelets to be tested can be collected using blood collection equipment, such as blood cell separators, fully automated apheresis machines, and blood cell analyzers, etc., without limitation. For example, according to GB18467-2011 "Requirements for Health Examination of Blood Donors", a blood cell separator (Amicus 4R4580, Fenwal, USA) can be used to collect platelets from healthy volunteers. Furthermore, the platelets to be tested can also be pre-sampled platelets, without limitation.

[0079] The collected platelets can be used to prepare leukocyte-depleted apheresis platelets, apheresis platelets, concentrated platelets, and multi-person mixed concentrated platelets. The cloud-like features of leukocyte-depleted apheresis platelets, apheresis platelets, concentrated platelets, multi-person mixed concentrated platelets, and leukocyte-depleted apheresis platelets can be captured using image acquisition equipment.

[0080] The following explanation uses leukocyte-reduced apheresis platelets as an example. Depending on the actual needs, leukocyte-reduced apheresis platelets, single-donor platelets, platelet concentrate, multi-donor mixed platelet concentrate, etc., can be selected; no specific limitation is made here. For example, leukocyte-reduced apheresis platelets can be prepared by collecting platelets, where each volunteer donates two therapeutic doses (each therapeutic dose is 225-275 mL, containing no less than 2.5 × 10⁻⁶ mmol / L). 11 Leukocyte-reduced apheresis platelets (1 platelet) are placed in a platelet shaker for continuous shaking and storage. The temperature range of the shaker is 20℃-24℃, and no specific temperature limit is imposed.

[0081] Step S2: Use image and video equipment to acquire cloud-like feature data of the platelets to be detected, the cloud-like feature data including cloud-like feature image data and / or cloud-like feature video data.

[0082] Figure 2 shows a schematic diagram of a cloud-like feature image based on platelets to be detected according to an embodiment of the present invention.

[0083] Image and video devices can include video recorders, camcorders, cameras, webcams, as well as devices with photo and video recording functions such as mobile phones and computers, without being limited to any one of them.

[0084] Concentrated platelets consist of a liquid medium and platelets. The platelets are suspended and aggregated in the medium. When shaken appropriately, concentrated platelets will exhibit a cloud-like appearance. This cloud-like appearance is referred to as the cloud-like characteristic of platelets.

[0085] The leukocyte-reduced platelet collection bags obtained and stored in step S1 were subjected to standardized collection. The standardized collection process involved setting the ambient temperature to approximately 22°C and illuminating the platelets from below using a light source. The platelets were fixed on a shaking table, with the maximum deviation distance of the shaking amplitude set to 15 cm and the shaking frequency to 4 Hz. Shaking was initiated, and the image and video equipment was activated at the start of shaking. Shaking was stopped after 6 seconds. After shaking stopped, platelet images or videos were captured using a video acquisition device for a preset duration. The preset duration could be set appropriately based on the platelet texture characteristics to be detected, preferably 5 seconds. Video recording was stopped after the cloud-like texture of the platelets dissipated. The cloud-like feature data of the platelets to be detected includes cloud-like feature image data and / or cloud-like feature video data, where the cloud-like texture of the platelets to be detected is shown in Figure 2. Standardized collection of the cloud-like feature data of the platelets to be detected minimizes the errors caused by differences in the collection process between the cloud-like feature data of the platelets to be detected.

[0086] Step S3: Evaluate platelet performance indicators using the cloud-like characteristic data of the platelets to be tested.

[0087] The platelet performance data may include: adhesion rate, maximum aggregation rate, platelet factor 4 (PF4) concentration, blood clot contraction rate, time to carotid artery thrombosis in mice, time to tail bleeding in mice, percentage of platelets transfused into the body, phosphatidylserine (PS) positivity rate, P-selectin (CD62P) positivity rate, and mean fluorescence intensity of platelet glycoprotein Ibα (CD42b).

[0088] For example, when blood vessels are damaged, platelets respond immediately: they adhere to the damaged site through adhesion; they rapidly aggregate to form a thrombus to block the ruptured blood vessel; and they further enhance hemostasis through release and contraction. However, platelets suffer storage damage during in vitro storage, specifically manifested in: changes in the glycoprotein structure on the platelet surface, resulting in impaired adhesion; platelet apoptosis, leading to reduced release of bioactive substances; platelet activation, inducing rapid clearance by the body; and ultimately, impaired hemostatic function and shortened in vivo survival. Therefore, the adhesion, aggregation, release, and contraction functions of platelets are crucial for their hemostatic effect. The stronger these functions and the less severe the storage damage, the stronger the hemostatic function of platelets, specifically reflected in shorter carotid artery thrombosis time and tail-cutting bleeding time in mice. Therefore, evaluating the quality of these platelet performance indicators has important guiding significance for precise platelet transfusion in clinical medicine.

[0089] The platelet performance evaluation method of the present invention can accurately, quickly and at low cost detect platelet performance indicators, solving the problems of long time consumption, high cost and large result error of traditional platelet performance detection methods.

[0090] Example 1

[0091] Figure 3 shows a flowchart of step S3 of a platelet performance index evaluation method according to an embodiment of the present invention.

[0092] In one example, as shown in 3, step S3 may include:

[0093] Step S311: Perform frequency domain analysis and gradient analysis on the cloud-like feature data of each frame of platelets to be detected to obtain a comprehensive evaluation index of the cloud-like feature data of the platelets to be detected.

[0094] Step S312: Record the duration of the cloud-like feature of the platelet to be detected to obtain the duration of the cloud-like feature data of the platelet to be detected.

[0095] Step S313: Based on the comprehensive evaluation index and duration of the cloud-like feature data of the platelets to be tested, obtain the significance features of the cloud-like feature data of the platelets to be tested;

[0096] Step S314: Evaluate platelet performance indicators based on the significant characteristics of the cloud-like feature data of the platelets to be tested and the correspondence between platelet performance indicators.

[0097] Figures 4 and 5 respectively show the frequency domain spectrum diagram and gradient amplitude diagram of the cloud-like feature image of the platelet to be detected according to an embodiment of the present invention.

[0098] The comprehensive evaluation index of the cloud-like feature data of the platelets to be detected in step S311 includes the spectral energy ratio, the mean gradient amplitude, and the stripe direction fraction. Therefore, the comprehensive evaluation index S of the cloud-like feature data of the platelets to be detected is:

[0099] (1)

[0100] Among them, R freq G is the spectral energy ratio. mean H is the mean of the gradient magnitude. s α is the fringe direction fraction; α is the frequency domain feature weight, which can be 0.5; β is the gradient feature weight, which can be 0.3; γ is the direction feature weight, which can be 0.2.

[0101] Figure 4 shows a schematic diagram of the frequency domain spectrum of a cloud-like feature image of platelets to be detected according to an embodiment of the present invention.

[0102] Calculate the spectral energy ratio R freq :

[0103] Spectral energy ratio R freqThis refers to the proportion of high-frequency component energy (related to stripe characteristics) relative to the total energy in the cloud-like feature map of the platelet being tested (high-frequency component proportion). The proportion of high-frequency component energy (related to stripe characteristics) relative to the total energy in the cloud-like feature map of the platelet being tested can be measured using the frequency domain characteristics of the cloud-like feature map. The specific calculation process is as follows:

[0104] Step L1: Convert the input RGB image of the cloud-like feature map of the platelets to be detected into a grayscale image of the cloud-like feature map of the platelets to be detected, as shown in Figure 2. The grayscale image of the cloud-like features of the platelets to be detected is transformed from the spatial domain to the frequency domain using discrete Fourier transform:

[0105] (2)

[0106] Where M x N is the grayscale image size of the cloud-like features of the platelets to be detected, and M and N are positive integers.

[0107] Step L2: The fft.fftshift function from the NumPy library, as shown in equation (3), can be used to move the low-frequency part of the frequency domain features of the cloud-like feature map of the platelet to be detected to the center of the spectrum:

[0108] (3)

[0109] Step L3: Extract the amplitude of the frequency components of the frequency domain features of the cloud-like feature map of the platelets to be detected, ignoring phase information:

[0110] (4)

[0111] in, and These are the real and imaginary parts of the Fourier transform, respectively.

[0112] Step L4: Using equation (5), take the logarithm of the amplitude of the frequency component of the cloud-like feature map of the platelet to be detected, and obtain the frequency domain spectrum as shown in Figure 4. Using the logarithm of the amplitude of the frequency component to compress the dynamic range can enhance the display effect.

[0113] (5)

[0114] Step L5: Calculate the high-frequency energy of the frequency domain features of the cloud-like feature map of the platelets to be detected. For example, frequency components outside the frequency radius R of the spectrum center can be selected as high-frequency components (i.e., high-frequency regions), and the frequency radius r (u, v) of the spectrum center can be defined as follows:

[0115] (6)

[0116] Frequency radius from the center of the spectrum The frequency components are represented as high-frequency components (high-frequency region), thus yielding high-frequency energy:

[0117] (7)

[0118] Step L6: Calculate the sum of the energies of all high-frequency components in the frequency domain features of the cloud-like feature map of the platelet to be detected:

[0119] (8)

[0120] Step L7: Calculate the spectral energy ratio R freq (Percentage of high-frequency energy):

[0121] (9).

[0122] Figure 5 shows a schematic diagram of the gradient amplitude of a cloud-like feature image of platelets to be detected according to an embodiment of the present invention.

[0123] Calculate the mean gradient magnitude G mean :

[0124] The significance of pixel intensity changes in the cloud-like feature image of the platelets to be detected can be measured by the gradient features of the frequency domain map of the cloud-like feature image. The significance of these pixel intensity changes is related to the sharpness of the stripe edges. The mean gradient magnitude G is calculated. mean The specific process is as follows:

[0125] Step Q1: Use the Sobel operator (image gradient extraction operator) to calculate the horizontal gradient of the grayscale image of the cloud-like features of the platelets to be detected. and vertical gradient :

[0126] (10)

[0127] (11)

[0128] in, This indicates a convolution operation.

[0129] Step Q2: Horizontal gradient of the grayscale image based on the cloud-like features of the platelets to be detected calculated in Step Q1. and vertical gradient The gradient magnitude of each pixel in the grayscale image of the cloud-like features of the platelets to be detected is calculated according to equation (12):

[0130] (12)

[0131] Step Q3: Calculate the mean gradient magnitude G of all pixels in the grayscale image of the cloud-like feature of the platelet to be detected. mean , used to represent the average degree of pixel intensity variation in the overall grayscale image of the cloud-like features of the platelets to be detected. Gradient magnitude mean G mean for:

[0132] (13)

[0133] The average gradient magnitude G mean Mapping the color range of the cloud-like feature image of the platelet to be detected yields the gradient amplitude map of the cloud-like feature image of the platelet to be detected, as shown in Figure 5.

[0134] Calculate the histogram entropy of the gradient direction distribution:

[0135] Step M1: Obtain the grayscale image of the cloud-like feature image of the platelets to be detected. gradient direction of each pixel gradient direction Expressed in radians, the range of values ​​is Then the gradient direction for:

[0136] (14)

[0137] Step M2: Grayscale image of the cloud-like feature image of the platelets to be detected. gradient direction of each pixel Discretization is performed to extract the grayscale image of the continuous cloud-like feature image of the platelets to be detected. gradient direction of each pixel Evenly divided into There are 3 directional intervals, where N is a positive integer.

[0138] (15)

[0139] in, This represents the i-th directional interval.

[0140] Step M4: Statistically analyze the grayscale image of the cloud-like feature image of the platelets to be detected. The frequencies of each gradient direction are used to construct a grayscale image of the cloud-like feature image of the platelets to be detected. The histogram of gradient orientation distribution. Each item in the gradient orientation histogram H represents the grayscale image of the cloud-like feature image of the platelet to be detected. The number of pixels belonging to this direction range:

[0141] (16)

[0142] in, The Kronecker delta function (an inner-chain binary function) is used to determine the gradient direction. Whether it falls within the i-th direction interval, bin(θ) is the interval number corresponding to direction θ.

[0143] Step M5: Normalize the grayscale image of the cloud-like feature image of the platelets to be detected. The histogram of gradient directions represents the probability distribution.

[0144] (17)

[0145] in, , where j represents the interval in the j-th direction.

[0146] Step M6: Calculate the grayscale image of the cloud-like feature image of the platelets to be detected according to the Shannon entropy (information entropy) formula. Entropy of the gradient direction histogram:

[0147] (18)

[0148] Step M7: Calculate the fringe direction fraction:

[0149] (19)

[0150] The entropy of the histogram of the gradient direction distribution in the grayscale image of the cloud-like feature image of the platelets to be detected can be used to measure the distribution of the gradient direction in the grayscale image of the cloud-like feature image of the platelets to be detected, which can effectively reflect the salience of the image stripes. For example, a higher entropy value usually indicates that the gradient direction distribution in the grayscale image of the cloud-like feature image of the platelets to be detected is more uniform, and the stripe directions in the cloud-like feature image of the platelets to be detected are more diverse and dispersed; a lower entropy value indicates that the gradient direction distribution in the grayscale image of the cloud-like feature image of the platelets to be detected is more concentrated, and the stripe directions in the cloud-like feature image of the platelets to be detected are consistent. That is, by calculating the distribution entropy of the gradient direction distribution in the grayscale image of the cloud-like feature image of the platelets to be detected, the directional complexity of the stripes in the cloud-like feature image of the platelets to be detected can be evaluated.

[0151] The above steps allow for the calculation of a comprehensive evaluation index S for the cloud-like feature data of each frame of platelets to be detected. The duration T of this cloud-like feature data is recorded. Multiplying the comprehensive evaluation index S by the duration T yields the significance index for each frame of the cloud-like feature data. In one example, platelet performance indicators can be evaluated based on the correspondence between the significance of the cloud-like feature data and platelet performance indicators. Specifically, the maximum significance of the cloud-like feature data is determined, and the quality of the platelet performance indicators is evaluated based on the correspondence between the maximum significance of the cloud-like feature data and the platelet performance indicator level.

[0152] For example, platelet performance indicators can be classified into different grades based on their hemostatic function. These grades can be categorized from 1 to 5. Grade 5 indicates the best platelet performance and hemostatic function; grades 4, 3, 2, and 1 indicate progressively decreasing platelet performance; and grade 1 indicates the worst platelet performance and hemostatic function. Accurate evaluation of platelet performance indicators is crucial for precise platelet transfusion in clinical practice. For instance, when a doctor requests concentrated platelets from the blood transfusion department to effectively stop bleeding in a patient with acute hemorrhage, if the platelet performance indicator grade is 5 (best performance), 2 units of concentrated platelets would be sufficient for effective hemostasis. Conversely, if the grade is 1 (worst performance), 10 units of concentrated platelets would be needed to achieve the same therapeutic effect. Furthermore, it plays an important role in guiding the use of antiplatelet drugs, including aspirin and clopidogrel. For example, a potential stroke patient with a platelet count of grade 5 before using aspirin should have their platelet count drop below grade 5 after taking 100 mg of aspirin daily. If the platelet count does not decrease, it indicates that the treatment dose is ineffective and the dose should be increased. If the platelet count still does not decrease after increasing the aspirin dose to 300 mg, it indicates that aspirin is not effective for the patient and another type of antiplatelet drug should be used.

[0153] Based on extensive experimental analysis, the correspondence between the significance values ​​of the cloud-like feature data of the platelets to be tested and the platelet performance indicators can be obtained. For example, the significance value of the cloud-like feature data of the platelets to be tested is in the range [0, 5), corresponding to a platelet performance indicator level of 1; the significance value of the cloud-like feature data of the platelets to be tested is in the range [5, 8), corresponding to a platelet performance indicator level of 2; the significance value of the cloud-like feature data of the platelets to be tested is in the range [8, 13), corresponding to a platelet performance indicator level of 3; the significance value of the cloud-like feature data of the platelets to be tested is in the range [13, 17), corresponding to a platelet performance indicator level of 4; and the significance value of the cloud-like feature data of the platelets to be tested is greater than or equal to 17, corresponding to a platelet performance indicator level of 5.

[0154] When continuously photographing or recording images of the same platelet sample using image and video equipment, the duration T of the cloud-like feature of the platelet sample is constant. At this time, the comprehensive evaluation index S of each frame of the platelet cloud-like video can be calculated using formula (1). The largest comprehensive evaluation index S is selected, and the largest comprehensive evaluation index S is multiplied by the duration T of the cloud-like feature of the corresponding platelet sample to obtain the maximum significant feature index of the platelet sample. At this time, the cloud-like features (texture, edge, etc.) of the platelet sample frame image are the most significant. Based on the magnitude of the significant feature index of the platelet sample and its correspondence with the platelet performance index level, the level of the platelet performance index can be found. The quality of platelet performance is judged by the platelet performance index level. It can efficiently and quickly evaluate the quality of platelet performance index and has an important guiding role in the precise transfusion of clinical platelets and the application of antiplatelet drugs.

[0155] Experimental verification:

[0156] The correlation coefficient R is a statistic that measures the correlation between platelet performance indicators and the significance of the cloud-like characteristic data of the platelets being tested. It represents the degree of fit of the correlation between platelet performance indicators and the significance of the cloud-like characteristic data of the platelets being tested.

[0157] Correlation coefficient (20)

[0158] in, , It is the standard deviation of variable X. X is the standard deviation of variable Y. In this embodiment, variable X is a platelet performance index, and variable Y is the significance characteristic of the cloud-like feature data of the platelets to be tested.

[0159] The correlation coefficient R ranges from -1 to 1. The larger the absolute value of the correlation coefficient R, the stronger the correlation between variables X and Y, and the better the model fit. As shown in Table 1, the R values ​​are all above 0.8, and some R values ​​are above 0.9, indicating a stronger correlation between platelet performance indicators and the significance of the cloud-like characteristic data of the platelets being tested.

[0160] Table 1: Significance characteristics of the cloud-like feature data of platelets to be tested and the correlation coefficients (R) with platelet performance indicators

[0161] Indicator Name Adhesion Rate % Maximum Aggregation Rate % PF4 Concentration ng Blood Clot Contraction Rate % Mouse Carotid Artery Thrombosis Time sec R 0.985091739 0.953127013 -0.938754006 0.966960555 -0.913477239

[0162] Indicator Name: Mouse Tail-Clapping Bleeding Time (sec); Percentage of Platelets Transfused into the Body (%); PS Positive Rate (%); CD62P Positive Rate (%); CD42b Average Fluorescence Intensity; R: -0.973567256; 0.874808581; -0.961571837; -0.963924807; 0.894226797

[0163] Example 2

[0164] According to another aspect of the present invention, evaluating platelet performance indicators using the cloud-like feature data of the platelets to be tested may include: inputting the cloud-like feature data of the platelets to be tested into a trained platelet performance indicator evaluation model (PEM) to obtain platelet performance indicator evaluation data. The platelet performance indicator evaluation model (PEM) is used to evaluate platelet performance to guide clinical practice in precise platelet transfusion, the application of antiplatelet drugs, and other anticoagulant or procoagulant treatments.

[0165] The Platelet Performance Evaluation Model (PEM) is trained based on a large amount of platelet performance index data and platelet cloud-like feature data. The PEM comprises a Convolutional Neural Network (CNN) and a Recurrent Neural Network (RNN). The CNN B-CNET is a selected existing CNN backbone model, capable of extracting the cloud-like features of platelets from cloud-like feature images or video frames. B-CNET includes a sequentially connected input layer, four convolutional layers, two pooling layers, a batch normalization layer, a 3D convolutional layer, and a fully connected layer. The RNN B-RNET is also a selected existing RNN backbone model, capable of handling the dynamic features of platelet cloud-like changes over time. B-RNET includes an input layer, hidden layers, and an output layer.

[0166] The platelet performance evaluation method in this embodiment may include: acquiring cloud-like feature data of the platelets to be tested using an image and video device, wherein the cloud-like feature data includes cloud-like feature image data and / or cloud-like feature video data; preprocessing the cloud-like feature data of the platelets to be tested to obtain preprocessed cloud-like feature data of platelets that meets the input requirements of the Platelet Performance Evaluation Model (PEM); inputting the preprocessed cloud-like feature data of platelets into the Platelet Performance Evaluation Model (PEM) to obtain platelet performance index data. The quality of platelet performance indexes is then evaluated based on the predicted platelet performance index data obtained from the Platelet Performance Evaluation Model (PEM).

[0167] Figure 6 shows a flowchart of the platelet performance index evaluation model PEM training method according to an embodiment of the present invention.

[0168] In one possible implementation, as shown in Figure 6, the platelet performance index evaluation model (PEM) is trained based on platelet performance index data and platelet cloud-like feature data, and may include:

[0169] S61: Collect platelets for training. Platelet collection can be achieved using the method described in Example 1, and will not be elaborated here.

[0170] S62: Acquire cloud-like feature data of training platelets using image and video equipment. The cloud-like feature data includes cloud-like feature image data and / or cloud-like feature video data. The acquisition of cloud-like feature data of platelets can be implemented using the scheme in Example 1, which will not be elaborated here.

[0171] S63: Performance metrics data for testing training platelets.

[0172] Platelet performance indicators can be detected by the following methods: adhesion rate, maximum aggregation rate, platelet factor 4 (PF4) concentration, clot contraction rate, time to carotid artery thrombosis in mice, time to tail bleeding in mice, percentage of platelets transfused into the body, phosphatidylserine (PS) positivity rate, P-selectin (CD62P) positivity rate, and mean fluorescence intensity of platelet glycoprotein Ibα (CD42b).

[0173] Platelet adhesion rate detection. Take 0.5 ml of the leukocyte-reduced apheresis platelets to be tested and count the platelet concentration (before contact) using a hematology analyzer (XP-100, Sysmex, Japan). Take 1.5 ml of the leukocyte-reduced apheresis platelets and place them in a spherical bottle. Place the spherical bottle on a rotating device and rotate it at 3 r / min for 15 min to ensure sufficient contact between the platelets and the bottle wall. Take 0.5 ml of the leukocyte-reduced apheresis platelets from the spherical bottle and count the platelet concentration (after contact) using a hematology analyzer. Platelet adhesion rate = (platelet concentration before contact - platelet concentration after contact) / platelet concentration before contact × 100%.

[0174] Platelet maximum aggregation rate assay. The platelet maximum aggregation rate can be detected using a hemagglutination analyzer (CS-2400, Sysmex, Japan). The platelet maximum aggregation rate is measured according to the instructions of the Platelet Aggregation Function (ADP) kit (Sysmex, Japan).

[0175] PF4 concentration detection. Take 1 ml of the leukocyte-depleted apheresis platelet sample, centrifuge at 1000 g for 5 min, obtain the supernatant, and detect the PF4 concentration according to the instructions of the PF4 ELISA kit (Boster Biological, China).

[0176] Clot retraction rate test. Aspirate 0.5 ml of the leukocyte-reduced platelet-only plasma to be tested, add calcium chloride to coagulate the plasma into a clot. After removing the plasma clot, read the volume of the remaining serum (i.e., the volume of precipitated serum). Clot retraction rate = (Leukocyte-reduced platelet-only volume - Volume of precipitated serum) / Leukocyte-reduced platelet-only volume × 100%.

[0177] Determination of carotid artery thrombosis time in mice. NOD-SCID mice aged 10-12 weeks were housed in a sterile environment. Anti-mouse CD42b monoclonal antibody (0.5 μL / g, Emfret, Germany) was injected via tail vein to deplete autologous platelets. Anesthesia was induced 24 h later with 1.25% tribromoethanol (30 μL / mg, Aibei Biotechnology, China). Rhodamine 6G solution (0.5 mg / mL, 8 μL / g, Sigma, USA) was injected intraperitoneally. Leukocyte-depleted apheresis platelets (1×10⁻⁶) were injected via tail vein.9 / mL, 18 μL / g). The sublingual glands were dissected to fully expose the common carotid artery. Filter paper soaked in FeCl3 solution (200 mg / mL, Aladdin, China) was placed over the exposed common carotid artery for 2 min. The process of common carotid artery thrombosis was observed using a stereofluorescence microscope (AXIO Zoom.V16, Zeiss, Germany), and the formation time was recorded.

[0178] Timing of bleeding at tail clipping in mice. Autologous platelets in NOD-SCID mice were depleted by intravenous injection of anti-mouse CD42b monoclonal antibody via tail vein. Twenty-four hours later, 1×10⁻⁶ leukocyte-reduced apheresis platelets were injected via tail vein. 9 / mL, 18 μL / g). Cut off the tip of the tail 3 mm from the tip, and quickly place the tailless mouse into a centrifuge tube containing preheated physiological saline (maintained at 37°C). Start timing when bleeding begins from the mouse's tail and stop timing when bleeding stops and no more bleeding occurs within 1 minute.

[0179] In vivo survival assay of transfused platelets. Leukocyte-depleted apheresis platelets (1×10⁻⁶) were injected into 12-week-old NOD-SCID mice via tail vein injection. 9 50 μL of peripheral blood was collected 2 h later (14 μL / g). After centrifugation at 250 g for 10 min, 1 mL of ACK lysis buffer (Boster Biological, China) was added to the collected peripheral blood pellet to lyse red blood cells. After 1 min, 2 mL of phosphate buffer was added and centrifuged again at 250 g for 10 min. The pellet was then fixed with 4% paraformaldehyde for 5 min. 1 mL of flow cytometry staining buffer was added and the pellet was washed again by centrifugation at 250 g for 10 min. Platelets were labeled with APC-conjugated anti-human CD42a and FITC-conjugated anti-mouse CD42a antibodies and analyzed using a flow cytometer (FACS Canto, BD, USA).

[0180] PS-positive platelet rate detection. Take 1 μL of leukocyte-reduced apheresis platelets (concentration 1×10⁻⁶). 9 ( / mL), referring to the instructions of the apoptosis detection kit (BD, USA), add 5 μL Annexin V-FITC and mix well. Incubate at room temperature in the dark for 15 min. Wash three times with buffer and analyze on a flow cytometer (FACS Canto, BD, USA).

[0181] CD62P-positive platelet count detection. Take 1 μL of leukocyte-reduced apheresis platelets (concentration 1×10⁻⁶) to be tested. 9Add 5 μL of FITC-conjugated anti-human CD62P (BD, USA) to the sample ( / mL), mix well, incubate at room temperature in the dark for 15 min, wash three times with flow cytometry buffer, and analyze by flow cytometry (FACS Canto, BD, USA).

[0182] Mean fluorescence intensity of CD42b in platelets was measured. 1 μL of leukocyte-reduced apheresis platelets (concentration 1×10⁻⁶) were collected for testing. 9 Add 5 μL of APC-conjugated anti-human CD42b (BD, USA) to the sample ( / mL), mix well, incubate at room temperature in the dark for 15 min, wash three times with flow cytometry buffer, and analyze by flow cytometry (FACS Canto, BD, USA).

[0183] Using the platelet performance index detection methods described above, platelet performance index evaluation datasets at different time points can be obtained. For example, platelet performance index data from 100 cases saved on day 1, 100 cases saved on day 2, 100 cases saved on day 3, 100 cases saved on day 4, and 100 cases saved on day 5 can be collected to form a 500-case platelet performance index dataset.

[0184] Similarly, using the method in step S2, cloud-like feature images and / or video data of platelets corresponding to the aforementioned 500-case platelet performance index evaluation dataset can be collected.

[0185] A platelet performance index dataset consisting of 500 cases and corresponding cloud-like feature images and / or video data of 500 platelets constitutes the platelet performance index evaluation dataset. The platelet performance evaluation dataset is then divided into training data T and test data C for the platelet performance index evaluation model PEM according to any ratio between 7:3 and 9:1.

[0186] S64: Preprocess the cloud-like feature data of training platelets to obtain cloud-like feature preprocessed data of training platelets that meet the input requirements of the platelet performance index evaluation model (PEM).

[0187] First, extracting keyframe images from the cloud-like feature video data of platelets can give the processed cloud-like feature video of platelets an appropriate time series length, reducing redundant information and video frame rate.

[0188] Keyframe extraction can be achieved by using the K-means clustering algorithm to cluster N cloud-like feature images within a preset time period. The most representative cloud-like feature image of each cluster center is selected as the keyframe image. C keyframe images of different categories form a fixed-length time series video V of platelet cloud-like features, where the number of clusters can be set to C. C is a positive integer, C≥1, and is not limited here.

[0189] Secondly, the resolution of keyframe images in the cloud-like feature video V is adjusted to ensure that the keyframe image data size of all platelet cloud-like features input into the convolutional neural network B-CNET of the platelet performance evaluation model PEM is a fixed size, such as 1024*768. Other resolutions can be selected as needed, without any limitation. Then, the keyframe images of platelet cloud-like features after resolution adjustment are subjected to denoising and image enhancement processing. Furthermore, during training, image enhancement, rotation, translation, scaling, flipping, contrast adjustment, and brightness adjustment are randomly performed on the keyframe image data of platelet cloud-like features to make it easier to capture the cloud-like features (texture features) in the cloud-like feature images or video frame data images of the training platelets.

[0190] Finally, the pixel values ​​of the keyframe image data of platelet cloud features are normalized (e.g., by minimizing the maximum value). The pixel values ​​of the keyframe image data of platelet cloud features are scaled to the range of 0 to 1 or standardized so that the distribution of the pixel values ​​of the keyframe image data of platelet cloud features follows a normal distribution. In this way, the preprocessed data of platelet cloud features that meet the input requirements of the platelet performance index evaluation model PEM is obtained.

[0191] Step S641 enables the processing of cloud-like feature images or video frame data images of platelets with different time series lengths to meet the input feature requirements of the platelet performance index evaluation model (PEM), thereby improving the training efficiency of the platelet performance index evaluation model (PEM).

[0192] S65: Input the preprocessed data of cloud-like features of training platelets into the platelet performance index evaluation model PEM to obtain the predicted data of training platelet performance index.

[0193] S66: Train a platelet performance index evaluation model PEM based on platelet performance index detection and prediction data used for training.

[0194] Before training the platelet performance evaluation model (PEM), the convolutional neural network (B-CNET) and recurrent neural network (B-RNET) in the PEM can be defined using the deep learning framework PyTorch, and then initialized. The initialization parameters for the PEM model are: learning rate of 0.01, training epochs of 50, and training batch size of 32.

[0195] The platelet performance evaluation model PEM is trained using the training data T of the platelet performance evaluation dataset that conforms to the normal distribution condition constructed in step S63. The specific process is as follows:

[0196] P1: Input the keyframe image of the cloud-like features of the training platelets into the convolutional neural network B-CNET of the platelet performance index evaluation model PEM, and output the keyframe image features of the cloud-like features of the platelets to be detected, such as the edge, texture, color, contour, corner, shape features and angular features of the keyframe image of the cloud-like features of the platelets to be detected, forming the keyframe image feature parameter matrix of the cloud-like features of the platelets to be detected.

[0197] P2: The keyframe image features of the cloud-like characteristics of the platelets to be detected are input into the recurrent neural network B-RNET, and the output is platelet performance index prediction data. Specifically, the keyframe image feature parameter matrix of the cloud-like characteristics of the platelets to be detected is input into the recurrent neural network B-RNET for feature association. The fully connected layer of the platelet performance index evaluation model PEM maps and classifies the keyframe image feature parameter matrix of the cloud-like characteristics of the platelets to be detected, transforming it into platelet performance index prediction data, and outputs the platelet performance index prediction data.

[0198] P3: Based on platelet performance index detection data and prediction data, calculate the loss value of the platelet performance index evaluation model PEM.

[0199] The loss value of the platelet performance index evaluation model PEM can be calculated using the L1 loss function of equation (21):

[0200] (twenty one)

[0201] in, This represents the true value of the m-th platelet performance index for the n-th sample in the platelet performance index evaluation model PEM. The platelet performance index prediction results are given by the platelet performance index evaluation model PEM. M and N are both positive integers, where M is the number of platelet performance indices and N is the number of training samples for the platelet performance index evaluation model PEM.

[0202] P4: Update the weight and bias parameters of the platelet performance index evaluation model PEM using the loss value and gradient backpropagation algorithm.

[0203] The weight parameters W and bias parameters b of the platelet performance evaluation model PEM can be updated using the SGD optimizer and gradient backpropagation algorithm, specifically:

[0204] (twenty two)

[0205] (twenty three)

[0206] Where W and b are the weights and bias parameters of the platelet performance evaluation model PEM before the update, respectively, and W' and b' are the weights and bias parameters of the platelet performance evaluation model PEM after the update. and Let be the gradients of the loss L with respect to the weight parameter W and the bias parameter b, respectively, and η be the learning rate.

[0207] P5: Repeat the process from P1 to P4 until the preset number of training epochs is reached, and the platelet performance index evaluation model PEM is obtained.

[0208] After steps S61-S66, a trained platelet performance index evaluation model (PEM) can be obtained. Finally, the cloud-like feature data of the platelets to be detected is input into the platelet performance index evaluation model (PEM) to obtain platelet performance index evaluation data.

[0209] According to another aspect of the present invention, a platelet performance index evaluation device is proposed. The device may include: an image and video device for acquiring cloud-like feature data of platelets to be tested, the cloud-like feature data including cloud-like feature image data and / or cloud-like feature video data; a preprocessing module for preprocessing the cloud-like feature data of platelets to be tested to obtain preprocessed cloud-like feature data of platelets that meets the input requirements of the Platelet Performance Index Evaluation Model (PEM); a model output module for inputting the preprocessed cloud-like feature data of platelets into the Platelet Performance Index Evaluation Model (PEM) and outputting platelet performance index data; and an evaluation module for evaluating the quality of platelet performance indexes based on the predicted platelet performance index data obtained from the Platelet Performance Index Evaluation Model (PEM).

[0210] The platelet performance index evaluation model (PEM) training method of the present invention, which uses the cloud-like feature of platelets, is obtained by inputting the cloud-like feature data of platelets in the test data C of the platelet performance evaluation dataset into the platelet performance index evaluation model (PEM), and outputting predicted values ​​of platelet performance indexes. The predicted values ​​of platelet performance indexes output by the PEM are compared with the detection results (measured values ​​and true values) of platelet performance indexes corresponding to the cloud-like feature data in the test data C of the platelet performance evaluation dataset to verify the accuracy of the PEM.

[0211] Verification results:

[0212] The experimental environment was Ubuntu 18.04 and Python 3.7, with an Intel Core i7-8700K CPU, an Nvidia RTX 2080Ti GPU and 32GB of RAM. The deep learning framework used was PyTorch 1.4.0.

[0213] Figures 7a-7j show scatter plots of the actual and predicted values ​​of platelet performance indicators according to an embodiment of the present invention.

[0214] Multiple frames of cloud-like feature data of platelets to be detected are input into the Platelet Performance Evaluation Model (PEM). The predicted values ​​of multiple platelet performance indicators output by the PEM and their corresponding actual values ​​are represented by scatter plots, as shown in Figures 7a-7j. The horizontal axis represents the measured values ​​of platelet performance indicators, and the vertical axis represents the predicted values ​​of platelet performance indicators output by the PEM. As can be seen from Figures 7a-7j, most of the scatter points are concentrated around the x=y line, indicating that the closer the predicted values ​​of platelet performance indicators output by the PEM are to the measured values, the better the performance of the PEM. This also demonstrates the strong correlation between the cloud-like features of platelets and platelet performance indicators.

[0215] Using the carotid artery thrombosis time of mice in test data C of the platelet performance evaluation dataset as a benchmark, the correlation coefficient between the carotid artery thrombosis time of mice in test data C of the platelet performance evaluation dataset and other indicators in the platelet performance index is calculated by formula (20), as shown in the true value row in Table 2. At the same time, the correlation trend between the carotid artery thrombosis time of mice and other indicators in the platelet performance index in test data C of the platelet performance evaluation dataset can be obtained. The cloud-like feature data of platelets in test data C of the platelet performance evaluation dataset is input into the platelet performance index evaluation model PEM to obtain the platelet performance index test value. Using the predicted carotid artery thrombosis time of mice as a benchmark, the correlation coefficient between the predicted value of the carotid artery thrombosis time of mice and the predicted value of other indicators in the platelet performance index is calculated by formula (20), as shown in the test value row in Table 2. Similarly, the correlation trend between the predicted value of the carotid artery thrombosis time of mice and the predicted value of other indicators in the platelet performance index can be obtained.

[0216] Table 2. Correlation between platelet performance indicators

[0217] Adhesion rate % Maximum aggregation rate % PF4 concentration ng Blood clotting rapid contraction rate % Mouse carotid artery thrombosis time sec True value -0.76087 -0.79091 0.799696 -0.832651 Predicted value -0.91935 -0.97261 0.980405 -0.929441 Difference 0.158479 0.181697 0.18071 0.096790

[0218] Mouse tail-cropping bleeding time (sec), percentage of platelets transfused into the body (%), PS positivity rate (%), CD62P positivity rate (%), mean CD42b fluorescence intensity. Mouse carotid artery thrombosis time (sec): True value 0.837807, -0.80938, 0.826413, 0.864145, -0.7781583291; Predicted value 0.964048, -0.96446, 0.953871, 0.964389, -0.9872401051; Difference 0.126241, 0.155078, 0.127458, 0.100243, 0.2090817760.

[0219] As shown in Table 2, the correlation coefficients between carotid artery thrombosis time and other indicators of platelet performance in mice in test data C of the platelet performance evaluation dataset (actual values ​​in Table 2) and the correlation coefficients between carotid artery thrombosis time and other indicators of platelet performance in mice output by the platelet performance evaluation model PEM (predicted values ​​in Table 2) are all within 0.2. This indicates that the correlation trend between carotid artery thrombosis time and other indicators of platelet performance in mice in test data C of the platelet performance evaluation dataset is basically consistent with the correlation trend between carotid artery thrombosis time and other indicators of platelet performance in mice predicted by the platelet performance evaluation model PEM.

[0220] Table 3: Accuracy Evaluation Indicators of Platelet Performance Evaluation Model PEM

[0221] Indicator Name Adhesion Rate % Maximum Aggregation Rate % PF4 Concentration ng Blood Clotting Rapid Contraction Rate % Mouse Carotid Artery Thrombosis Time sec MAE 2.6857681015.13958564818.86407585.59504354542.28706278 R 0.8914980620.885989040.8814348580.9200884560.928128985

[0222] Indicator Name: Mouse Tail-Clipping Bleeding Time (sec); Percentage of Platelets Transfused into the Body (%); PS Positive Rate (%); CD62P Positive Rate (%); CD42b Mean Fluorescence Intensity (MAE): 45.9067700; 42.45547943; 0.95525646; 1.476948484197; 9.784264; R: 0.925059659; 0.915105767; 0.932614657; 0.941883352; 0.826005496

[0223] (twenty four)

[0224] in, This represents the predicted value from the platelet performance evaluation model PEM. This represents the true value of platelet performance indicators.

[0225] Mean Absolute Error (MAE) is a commonly used method for evaluating continuous data. MAE is used to assess the predictive accuracy of the platelet performance evaluation model PEM. A MAE value of 0 indicates that the PEM predictions of the platelet performance evaluation model are exactly the same as the actual values. The smaller the MAE value, the closer the PEM predictions are to the actual values, indicating higher accuracy. Table 2 shows that the MAE values ​​for each platelet performance indicator are very small compared to the normal values ​​(the MAE value for CD42b average fluorescence intensity is 1979.78, while the normal value is around 20,000 to 30,000, making this value very small). This indicates high predictive accuracy of the platelet performance evaluation model PEM and a strong correlation between the cloud-like characteristics of platelets and platelet performance.

[0226] In Table 3, the correlation coefficient R represents the predicted values ​​of the platelet performance index evaluation model (PEM) for variable X and the actual values ​​for variable Y for variable Y. This correlation coefficient indicates the degree of fit between the PEM model and the actual values ​​of the platelet performance index. The correlation coefficient R ranges from -1 to 1. A larger absolute value of R indicates a better fit between the PEM model and the actual values ​​of the platelet performance index. As shown in Table 3, the R values ​​are all above 0.8, with some values ​​above 0.9, indicating a high degree of fit between the PEM model and the actual values ​​of the platelet performance index. The cloud-like characteristics of platelets are strongly correlated with platelet performance indexes; therefore, the cloud-like characteristics of platelets can be used to evaluate platelet performance indexes.

[0227] The platelet performance evaluation method of this invention can detect platelet performance indicators based on the cloud-like characteristics of platelets, solving the problems of long time consumption, high cost, and large result errors of traditional platelet performance testing methods. By using image-video regression prediction of platelet cloud-like characteristics, manual operation steps are reduced. Deep neural networks can quickly process and analyze video data of platelet cloud-like characteristics, achieving real-time or near-real-time platelet performance evaluation, significantly shortening the detection time and improving detection efficiency. It eliminates the need for professional personnel and specialized reagents and consumables, greatly reducing detection costs and thus lowering medical expenses for patients.

[0228] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0229] Figure 8 shows a structural diagram of a platelet performance index evaluation device based on the cloud-like characteristics of platelets according to an embodiment of the present invention; as shown in Figure 8, the platelet performance index evaluation device may include:

[0230] The collection device 801 is used to collect platelets to be tested.

[0231] Image and video acquisition device 802 is used to acquire cloud-like feature data of the platelets to be detected, wherein the cloud-like feature data includes cloud-like feature image data and / or cloud-like feature video data;

[0232] Evaluation module 803 is used to evaluate the platelet performance indicators using the cloud-like feature data of the platelets to be tested.

[0233] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0234] Figure 9 is a schematic diagram of the structure of the electronic device 3 provided in an embodiment of this application. As shown in Figure 9, the electronic device 3 of this embodiment includes: a processor 301, a memory 302, and a computer program 303 stored in the memory 302 and executable on the processor 301. When the processor 301 executes the computer program 303, it implements the steps in the various method embodiments described above. Alternatively, when the processor 301 executes the computer program 303, it implements the functions of each module / unit in the various device embodiments described above.

[0235] For example, computer program 303 may be divided into one or more modules / units, which are stored in memory 302 and executed by processor 301 to complete this application. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of computer program 303 in electronic device 3.

[0236] Electronic device 3 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 3 may include, but is not limited to, processor 301 and memory 302. Those skilled in the art will understand that FIG9 is merely an example of electronic device 3 and does not constitute a limitation on electronic device 3. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device may also include input / output devices, network access devices, buses, etc.

[0237] Processor 301 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.

[0238] The memory 302 can be an internal storage unit of the electronic device 3, such as a hard disk or RAM. The memory 302 can also be an external storage device of the electronic device 3, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card. Furthermore, the memory 302 can include both internal and external storage units of the electronic device 3. The memory 302 is used to store computer programs and other programs and data required by the electronic device. The memory 302 can also be used to temporarily store data that has been output or will be output.

[0239] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0240] The exemplary systems and methods of the present invention have been specifically shown and described with reference to the above embodiments, which are merely examples of the best mode for implementing the systems and methods. Those skilled in the art will understand that various changes can be made to the embodiments of the systems and methods described herein without departing from the spirit and scope of the invention as defined in the appended claims when implementing the systems and / or methods.

Claims

1. A method for evaluating platelet performance indicators, characterized in that, The method includes: Collect platelets for testing; The cloud-like feature data of the platelets to be detected is acquired using image and video equipment, and the cloud-like feature data includes cloud-like feature image data and / or cloud-like feature video data; Platelet performance indicators were evaluated using the cloud-like characteristic data of the platelets to be tested.

2. The platelet performance index evaluation method according to claim 1, characterized in that, The evaluation of platelet performance indicators using the cloud-like feature data of the platelets to be tested includes: Frequency domain analysis and gradient analysis are performed on the cloud-like feature data of each frame of platelets to be detected to obtain a comprehensive evaluation index of the cloud-like feature data of platelets to be detected. The duration of the cloud-like feature of the platelets to be detected is recorded to obtain the duration of the cloud-like feature data of the platelets to be detected. Based on the comprehensive evaluation index and duration of the cloud-like feature data of the platelets to be tested, the significance features of the cloud-like feature data of the platelets to be tested are obtained. The platelet performance indicators are evaluated based on the significant characteristics of the cloud-like feature data of the platelets to be tested and the correspondence between them and the platelet performance indicators.

3. The platelet performance index evaluation method according to claim 2, characterized in that, The comprehensive evaluation indicators for the cloud-like feature data of the platelets to be detected include the spectral energy ratio, the average gradient amplitude, and the stripe direction score.

4. The platelet performance index evaluation method according to claim 2, characterized in that, The evaluation of platelet performance indicators based on the significant characteristics of the cloud-like feature data of the platelets to be tested and the correspondence with platelet performance indicators includes: Determine the most significant feature of the cloud-like characteristic data of the platelets to be tested; The performance of platelets is evaluated based on the correspondence between the most significant feature of the cloud-like characteristic data of the platelets to be tested and the platelet performance index level.

5. The platelet performance index evaluation method according to claim 4, characterized in that, The platelet performance indicators are classified into different levels based on their hemostatic and coagulant functions.

6. The method for evaluating platelet performance indicators according to claim 1, characterized in that, The method of evaluating platelet performance indicators using the cloud-like feature data of the platelets to be tested also includes: The cloud-like feature data of the platelets to be detected is input into the trained platelet performance index evaluation model PEM to obtain platelet performance index evaluation data.

7. The method for evaluating platelet performance indicators according to claim 6, characterized in that, The platelet performance index evaluation model PEM is trained based on platelet performance index data and platelet cloud-like feature data.

8. The method for evaluating platelet performance indicators according to claim 7, characterized in that, The platelet performance index evaluation model PEM is trained based on platelet performance index data and platelet cloud-like feature data, including: Collect platelets for training; Cloud-like feature data of training platelets are collected using image and video equipment, and the cloud-like feature data includes cloud-like feature image data and / or cloud-like feature video data; Detect performance metrics data of platelets used in training; The cloud-like feature data of the training platelets are preprocessed to obtain cloud-like feature preprocessed data of the training platelets that meet the input requirements of the platelet performance index evaluation model (PEM). The cloud-like feature preprocessed data of the training platelets is input into the platelet performance index evaluation model PEM to obtain the predicted data of the performance index of the training platelets. Based on the platelet performance index detection data and prediction data used for training, a platelet performance index evaluation model PEM is trained.

9. The method for evaluating platelet performance indicators according to claim 8, characterized in that, The preprocessing of the cloud-like feature data of training platelets to obtain cloud-like feature preprocessed data of training platelets that meets the input requirements of the platelet performance index evaluation model (PEM) includes: Keyframe images are extracted from the cloud-like feature video data of the training platelets; K-means clustering algorithm was used to perform cluster analysis on keyframe images to obtain cloud-like feature keyframe images of training platelets with different cluster centers; The cloud-like feature keyframe images of training platelets for different cluster centers are subjected to denoising, resolution adjustment, image enhancement and normalization to obtain cloud-like feature keyframe images of training platelets. The cloud-like feature keyframe image of the training platelets is the cloud-like feature preprocessed data of the training platelets that meets the input requirements of the platelet performance evaluation model (PEM).

10. The method for evaluating platelet performance indicators according to claim 8, characterized in that, The platelet performance evaluation model PEM includes convolutional neural networks (CNN) and recurrent neural networks (RNN).

11. The method for evaluating platelet performance indicators according to claim 10, characterized in that, The preprocessed data of cloud-like features of training platelets are input into the platelet performance index evaluation model PEM to obtain predicted data of training platelet performance index. Based on the platelet performance index detection and prediction data used for training, a platelet performance index evaluation model PEM was trained, including: P1: Input the keyframe image of the cloud-like feature of the training platelets into the convolutional neural network CNN of the platelet performance index evaluation model PEM to obtain the keyframe image features of the cloud-like feature of the training platelets. P2: Input the keyframe image features of the cloud-like characteristics of the training platelets into the recurrent neural network (RNN) and output the predicted data of the performance indicators of the training platelets. P3: Based on the platelet performance index detection data and prediction data used for training, calculate the loss value of the platelet performance index evaluation model PEM; P4: Update the weight parameters and bias parameters of the platelet performance index evaluation model PEM using the loss value and gradient backpropagation algorithm. P5: Repeat the process from P1 to P4 until the preset number of training rounds are reached to train and obtain the platelet performance index evaluation model (PEM).

12. The method for evaluating platelet performance indicators according to claim 1, characterized in that, The platelet performance data include one or more of the following: adhesion rate, maximum aggregation rate, platelet factor 4 (PF4) concentration, blood clot contraction rate, time to carotid artery thrombosis in mice, time to tail bleeding in mice, percentage of platelets transfused into the body, phosphatidylserine (PS) positivity rate, P-selectin (CD62P) positivity rate, and mean fluorescence intensity of platelet glycoprotein Ibα (CD42b).

13. A platelet performance evaluation device, characterized in that, The device includes: The collection device is used to collect platelets for testing; An image and video acquisition device is used to acquire cloud-like feature data of the platelets to be detected, wherein the cloud-like feature data includes cloud-like feature image data and / or cloud-like feature video data; The evaluation module is used to evaluate the platelet performance indicators using the cloud-like feature data of the platelets to be tested.

14. An electronic device, characterized in that, The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method as described in any one of claims 1 to 12.

15. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 12.

Images