Intelligent blood sampling personnel behavior monitoring system

Abstract

The invention relates to the technical field of behavior monitoring, in particular to an intelligent blood sampling personnel behavior monitoring system which specifically comprises a multi-modal data processing module, a key quality parameter extraction module, a quality prediction module and a quality early warning module. The method comprises the following steps: synchronously acquiring multi-state original data of blood sampling, disinfection and uniform mixing of working behaviors of blood sampling personnel through multiple sensors in a time-space manner, constructing a hemolysis quality model through DTW time sequence alignment and mutual information verification, outputting hemolysis quality parameters, and screening and constructing a key quality feature set through a time-space feature matrix and transfer entropy; constructing a knowledge-random forest model, and outputting a work result quality prediction result and a mechanical damage index; and calculating a real-time hemolysis probability value, and dynamically adjusting a working result quality early warning threshold value. The problem that in the prior art, due to lack of blood sampling personnel behavior monitoring, the quality of work results is uneven is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of behavior monitoring, and is an intelligent behavior monitoring system for blood collection personnel. Background Art

[0002] In the blood collection scenario of outdoor blood donation vehicles, the quality of blood samples is directly related to their usability after being received by blood stations. However, the existing technologies have significant defects in multi-modal data fusion, real-time quality assessment, and embedded deployment, resulting in a high sample rejection rate. Existing systems mostly rely on a single sensor (such as a pressure or vision sensor) for local monitoring, lacking the collaborative analysis of multi-dimensional parameters such as puncture angle, blood vessel displacement, and mixing intensity. Moreover, due to the asynchronous hardware clocks, the timing data is misaligned, making it difficult to accurately capture the operation dynamics. For example, the phase deviation between the pressure peak and blood vessel displacement may mask abnormal blood vessel elasticity, leading to inaccurate hemolysis risk assessment. In addition, existing methods often construct feature sets using static statistics, ignoring the spatio-temporal interaction and causal mechanism between parameters. For example, the statistical correlation between the disinfection coverage rate and blood vessel visibility may mislead the model to learn false associations, failing to identify the true causal chain of "blood vessel condition → operation parameters → sample quality", resulting in poor model robustness and an increased false alarm rate in blood donation scenarios with significant individual differences. At the same time, most existing hemolysis risk assessment models are trained based on offline laboratory data and adopt a fixed threshold warning mechanism, unable to dynamically respond to real-time variables such as environmental temperature and blood vessel status. For example, blood vessel constriction caused by low temperature may mask abnormal operation force, resulting in the rejection of samples at blood stations due to abnormal hemolysis or coagulation. According to statistics, approximately 35% of the samples rejected by blood stations are due to such improper operation problems, and the existing technologies can only identify them during laboratory reexamination, causing waste of resources and a decline in the experience of blood donors. On the other hand, the existing algorithms have high complexity and are difficult to run in real time on the embedded devices equipped in blood donation vehicles, unable to support real-time feedback during the operation, restricting the feasibility of rapid detection of sample quality in outdoor scenarios. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to propose an intelligent behavior monitoring system for blood collection personnel aiming at the problem of uneven quality of work results caused by the lack of behavior monitoring of blood collection personnel in the existing technology.

[0004] To achieve the above object, an intelligent behavior monitoring system for blood collection personnel of the present invention includes the following modules: A multi-modal data processing module, a key quality parameter extraction module, a quality prediction module, and a quality warning module; The multi-modal data processing module synchronously collects multi-state raw data of the work behaviors of blood collection personnel, such as blood collection, disinfection, and mixing, through multi-sensor spatio-temporal synchronization, and generates a time-series correlation data set containing blood vessel elasticity features after DTW time-series alignment and mutual information verification; The key quality parameter extraction module is used to extract multi-dimensional quality parameters from time-series correlation data, construct a hemolysis quality model, output hemolysis quality parameters, and construct a key quality feature set through spatio-temporal feature matrix and transfer entropy screening; The quality prediction module is used to construct a knowledge-random forest model, integrate the key quality feature set into model training, and output the quality prediction result of the work achievement and the mechanical damage index; The quality warning module is used to calculate the real-time hemolysis probability value in real time based on edge computing, dynamically adjust the quality warning threshold of the work achievement, and perform hierarchical quality warning on the work achievement through hardware interaction and generate work achievement disposal suggestions.

[0005] Specifically, the multi-modal data alignment module is configured with a data acquisition architecture construction unit, a multi-modal data acquisition unit, a data preprocessing unit, and a redundant data removal unit; Preferably, the data acquisition architecture construction unit is used to run the following strategies: A11: By adopting a master-slave clock synchronization architecture, a timestamp with a 1ms interval is generated through a hardware timer (a high-precision timer built into the STM32H7 series microcontroller, with a resolution of 1ns), and the multi-state raw data is started by the synchronous pulse signal output by the clock subunit; A12: Implement a timestamp embedding mechanism in the sensor driver layer. Specifically: Set the data acquisition frequency of the angle sensor and the pressure sensor to once every 10ms, and add the current timestamp to the head of the data frame at the same time; At the same time, inject timestamps into the camera and the infrared imager through the FPGA; Use the sliding window algorithm to perform real-time calibration on the timestamp deviation received by the data processing center; A13: Divide the working behavior of the blood collection personnel into four working states: standby, action, feedback, and end, and set the switching trigger conditions between states, including: the first working state switching, the second working state switching, and the third working state switching; The trigger condition for the first working state switching includes: a falling edge appears in the pressure sensor data and the camera detects a change in blood color; The trigger condition for the second working state switching includes: the feedback state lasts for more than 3 seconds or the pressure sensor data is lower than the threshold of 0.5N; The trigger condition for the third working state switching includes: the behavior of the blood collection personnel enters the end state and all sensor data returns to the standby state; A14: In each working state, establish a parameter acquisition queue, including: The starting angle of the behavior, the starting pressure of the behavior, and the behavior pressure fluctuation rate in the action state; The blood feedback rate and the behavior feedback pressure in the feedback state; Behavior end angle change rate and behavior end pressure in the end state; Preferably, the multi-modal data acquisition unit collects multi-state raw data of the working behavior of blood collection personnel through intelligent hardware devices, including: action (puncture) state data, standby (disinfection) state data, mixing state data, and characteristic data of the blood collection personnel; The puncture state data includes: behavior angle, behavior pressure (puncture force), behavior pressure change rate, infrared blood vessel displacement data, and behavior feedback time (blood return time); The disinfection state data includes: calculating the disinfection coverage rate through a visual recognition algorithm; The mixing state data includes: mixing times and oscillation intensity; The characteristic data of the blood collection personnel includes: blood vessel visibility; The data preprocessing unit is used to run the following strategies: A21: Jointly align the pressure curve with the angle sensor data and infrared blood vessel displacement data, normalize the behavior start angle, behavior pressure, and infrared blood vessel displacement data, construct a three-dimensional feature vector, import the three-dimensional feature vector into the differential distance calculation strategy, and obtain the differential distance of the three-dimensional feature vector; Preferably, the differential distance calculation strategy is specifically: ; Among them, represents the value of the k-th dimension feature at times i and j; is the weight of the k-th dimension feature; is the historical data standard deviation of the k-th dimension feature; A22: Extract the differential distance of the three-dimensional feature vector, construct a differential distance matrix, and use the dynamic time warping path search algorithm to find the optimal path from the starting point to the ending point; It should be noted that the optimal path is the mapping relationship between two time series, which represents how to minimize the morphological difference between the two series through alignment (stretching / compressing); A23: Through the optimal path, establish a time mapping table between the behavior start pressure curve and the behavior feedback state. Through the gradient detection algorithm, find the local minimum point of the differential distance in the pressure sequence after path mapping, and mark it as the peak moment of the behavior start pressure curve; Calculate the blood vessel elasticity parameter of the blood collection personnel through the time mapping table ; ; Among them, is the starting moment of the behavior feedback state of the blood collection personnel, This is the peak moment of the stress curve for the blood drawer's behavior.

[0006] Specifically, the redundant data removal unit is used to run the following strategy: A31: The correlation between disinfection coverage and vascular visibility was quantified by probability algorithm, and the mutual information value between the two was calculated; The calculation strategy of the mutual information value is: ; in, They are disinfection coverage and blood vessel visibility; Indicates the time window and The joint probability density of Respectively represent the time window , The marginal probability density of ; A32: Extract the mutual information value and compare it with the preset correlation threshold. When the mutual information value is greater than the preset correlation threshold, it is judged that the disinfection coverage is highly correlated with the vascular visibility. Before extracting the key quality parameters, the redundant parameters are eliminated.

[0007] Specifically, the key quality parameter extraction module includes: a hemolysis quality estimation unit and a key quality feature set screening unit; The hemolysis quality estimation unit is configured with the following strategy: B11: Construct a red blood cell shear damage prediction model, calculate the shear stress of red blood cells during the mixing operation by the blood collector through the shear stress calculation strategy, input the shear stress into the red blood cell shear damage prediction model to output the cumulative amount of shear damage ; Preferably, the shear stress calculation strategy is: ; in, The shear stress on red blood cells caused by blood collection personnel during mixing operation; K is the blood viscosity of the person whose blood is drawn (characterizing the viscosity characteristics of the fluid); It is the intensity of the shaking during the mixing operation of the blood collector; is the apparent viscosity of blood (related to G); r is the non-Newtonian index; Preferably, the output formula of the red blood cell shear injury prediction model is: ; in, is the cumulative amount of shear damage; is the critical rupture stress; is an indicator function, when it is the case, it is 1, otherwise it is 0; B12: Incorporate the number of action states of the blood collection personnel , the tourniquet usage time and the cumulative amount of shear damage into the hemolysis quality prediction strategy to predict the hemolysis quality parameters of the blood collection personnel's work behavior ; The specific hemolysis quality prediction strategy is as follows: .

[0008] Specifically, the key quality feature set screening unit is configured with the following strategy: B21: Obtain the multi-state raw data of the 12-dimensional work behavior of the blood collection personnel, establish a spatio-temporal feature matrix, extract the statistical features of the mean, variance and peak value through a sliding window, and form a feature vector containing time-series dynamic information, where the sliding window is 5s; B22: Analyze the directional influence between each element in the 12-dimensional spatio-temporal feature matrix and the hemolysis quality parameter through the transfer entropy algorithm to obtain a transfer entropy sequence, and screen the first 50% of the transfer entropy sequence corresponding to the multi-state raw data as the key quality feature set.

[0009] Specifically, the quality prediction module includes: a prior knowledge embedding unit, a model structure optimization unit and a quality result output unit; The prior knowledge embedding unit is configured with the following operating strategy: Construct a random forest model. During the construction of the decision tree in the random forest, rewrite the information gain calculation function by reconstructing the weights. During the growth of the decision tree, perform split feature priors to obtain a knowledge-random forest model; Among them, the split feature prior includes: when the node satisfies the condition that the blood vessel visibility is less than 0.3, use the behavior pressure of the blood collection personnel and the number of action states of the blood collection personnel as split features; Preferably, the rewriting of the information gain calculation function specifically includes: when the input feature of the decision tree node includes blood vessel visibility and the blood vessel visibility is less than 0.3, reconstruct the weights corresponding to the behavior pressure of the blood collection personnel and the number of action states of the blood collection personnel to 1.2 times the initial weight; The model structure optimization unit is configured with the following operating strategy: C21: Add supplementary feature parameters to the key quality features to obtain 10-dimensional feature parameters. The supplementary feature parameters include: the average value of the starting angle of the behavior during the historical work process of the current blood collection personnel, the variance of the starting angle of the behavior, the standard deviation of the oscillation intensity, and the blood vessel elasticity parameter of the blood collection person. C22: Expand the input layer of the knowledge-random forest model and add the calculation of the mechanical damage index to the output layer of the knowledge-random forest model. Among them, the expansion of the input layer includes: dividing the 10-dimensional feature parameters into three feature categories: static, time series, and event, and performing feature integration. Preferably, the mechanical damage index The calculation strategy is: ; Among them, are the pressure damage coefficient and the oscillation change rate coefficient respectively; is the behavior pressure of the blood collection personnel; Specifically, the quality result output unit is configured with the following operation strategy: C31: Perform the same standardization processing on the spatio-temporal feature vectors in the validation dataset as on the training data. Specifically, the validation dataset includes: true labels including hemolysis probability, coagulation marker, contamination level, and corresponding 10-dimensional spatio-temporal feature vectors. C32: Input the preprocessed feature vectors into the knowledge-random forest model, and output the three-dimensional quality index prediction results and the mechanical damage index based on the trained parameters and decision rules.

[0010] Preferably, the three-dimensional quality index prediction results include: hemolysis probability, coagulation marker, and contamination level; Specifically, the quality warning module includes: a real-time hemolysis probability calculation unit and a disposal suggestion output unit; The real-time hemolysis probability calculation unit is used to calculate and obtain the real-time hemolysis probability value of the work results under the multi-state work behaviors of the blood collection personnel. Preferably, the real-time hemolysis probability value The calculation strategy is: ; Among them, is the number of action behaviors of the blood collection personnel; Deviation degree of the starting angle of the blood collection personnel's behavior; It should be noted that. It reflects the amplification effect of rapid needle insertion on the hemolysis probability under low blood vessel visibility; It should be noted that, Indicates the mixing stability penalty, which is the standard deviation of the oscillation intensity. The larger its value, the more unstable the mixing operation of the blood collection personnel; The disposal suggestion output unit is configured with the following operation strategies: D11: Extract the real-time hemolysis probability value, the visibility and elasticity parameters of the blood vessel of the person being blood collected, as well as the change rate of the behavioral pressure and the number of action states of the blood collection personnel; D12: According to D11, conduct quality risk early warning of the work results of the blood collection personnel, including: When the blood vessel visibility is less than 0.3 and the blood vessel elasticity parameter is less than 50 ms, ignore the early warning threshold of the real-time hemolysis probability value, directly mark the work result of the blood collection personnel as a yellow early warning, and output the disposal suggestion of giving priority to the quality inspection of the work result for warehousing and increasing the microscopic examination of red blood cell morphology; When the number of action states of the blood collection personnel is greater than or equal to 2 and the change rate of the behavioral pressure of the blood collection personnel > 1.5 N / s, ignore the early warning threshold of the real-time hemolysis probability value, directly trigger a red early warning, and output the disposal suggestion of re-conducting the blood collection work; When the real-time hemolysis probability value is less than 0.3 and there are no spatio-temporal abnormal features, directly mark the work result of the blood collection personnel as a green early warning, and output the disposal suggestion of normal transportation.

[0011] Compared with the prior art, the technical effects of the present invention are as follows: Through multi-modal data synchronous acquisition and three-dimensional spatio-temporal alignment, the present invention significantly improves the fusion accuracy and reliability of blood collection operation parameters, and overcomes the time series deviation problem of traditional single-sensor monitoring; The causal feature screening mechanism based on transfer entropy of the present invention effectively eliminates the interference of redundant parameters, and identifies key causal chains such as "puncture force - blood vessel response" through dynamic mutual information calculation. While achieving a 40% reduction in feature dimension on the embedded platform (STM32H7), 95% of the effective information is retained, and the operation efficiency is greatly improved, ensuring the high-efficiency quality assessment of the rapid blood collection process. In the outdoor blood donation vehicle scenario, when it is detected that the blood vessel elasticity parameter is lower than 50 ms or the pressure mutation rate > 1.5 N / s, a hierarchical early warning is immediately triggered and disposal suggestions (such as priority microscopic examination or re-blood collection) are generated, reducing the false alarm rate compared with the traditional fixed threshold model, increasing the detection rate of on-site quality defects of blood collection samples, and reducing the rejection rate of blood stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to these drawings. Among them: Figure 1 This is a schematic structural diagram of an intelligent blood collection personnel behavior monitoring system according to the present invention. Specific embodiments

[0013] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings of the specification.

[0014] In the following description, many specific details are set forth to facilitate a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0015] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that excludes other embodiments.

[0016] Embodiment 1: As Figure 1 shown, an intelligent blood collection personnel behavior monitoring system according to an embodiment of the present invention, as Figure 1 shown, includes the following modules: A multi-modal data processing module, a key quality parameter extraction module, a quality prediction module, and a quality warning module; The multi-modal data processing module synchronously collects multi-state raw data of the blood collection personnel's work behaviors of blood collection, disinfection, and mixing through multi-sensors in space and time, and generates a time-series correlation data set containing vascular elasticity characteristics through DTW time-series alignment and mutual information verification; The multi-modal data alignment module is configured with a data acquisition architecture construction unit, a multi-modal data acquisition unit, a data preprocessing unit, and a redundant data removal unit; Preferably, the data acquisition architecture construction unit is used to run the following strategies: A11: By adopting a master-slave clock synchronization architecture, a time stamp with a 1ms interval is generated through a hardware timer (a high-precision timer built into the STM32H7 series microcontroller, with a resolution of 1ns), and the multi-state raw data is started by the synchronous pulse signal output by the clock sub-unit; Exemplarily, in this embodiment, each sensor accesses the synchronous clock signal through the SPI / I2C bus, and the camera (such as IMX219, supporting 120fps high-speed shooting) and the infrared imager (FLIR Lepton 3.5, 9Hz frame rate) adopt an external trigger mode; A12: Implement a timestamp embedding mechanism in the sensor driver layer, specifically: set the data acquisition frequencies of the angle sensor and the pressure sensor to once every 10 ms, and at the same time add the current timestamp to the data frame header; at the same time, inject timestamps into the camera and the infrared imager through the FPGA; Use a sliding window algorithm to perform real-time calibration on the timestamp deviation received by the data processing center; A13: Divide the working behaviors of the blood collection personnel into four working states: standby, action (needle insertion), feedback (blood return), and end, and set the switching trigger conditions between the states, including: the first working state switching, the second working state switching, and the third working state switching; The trigger condition for the first working state switching includes: the pressure sensor data has a falling edge and the camera detects a change in blood color; The trigger condition for the second working state switching includes: the feedback state lasts for more than 3 seconds or the pressure sensor data is lower than the threshold of 0.5 N; The trigger condition for the third working state switching includes: the behavior of the blood collection personnel enters the end state and all sensor data returns to the standby state; A14: In each working state, establish a parameter acquisition queue, including: The starting angle of the behavior, the starting pressure of the behavior, and the behavior pressure fluctuation rate in the action state; The blood feedback rate and the behavior feedback pressure in the feedback state; The rate of change of the ending angle of the behavior and the ending pressure of the behavior in the end state; Exemplarily, in this embodiment, it should be noted that for the working process of the blood collection personnel, the action state is the state of needle insertion. The starting angle of the behavior reflects the angle change of the blood collection needle at each moment during the needle insertion process. Accurate angle information is crucial for evaluating the standardization and accuracy of the needle insertion operation; the starting pressure of the behavior reflects the pressure on the blood collection needle in the state of needle insertion, and the behavior pressure fluctuation rate reflects the speed of pressure change. In the state of needle insertion, the pressure and its pressure fluctuation rate can be used to judge whether the working behavior (needle insertion) of the blood collection personnel is normal, which helps the blood collection personnel adjust the operation in time and provides a basis for the behavior monitoring of the blood collection personnel; Exemplarily, in this embodiment, it should be noted that for the working process of the blood collection personnel, the feedback state is the state of blood return. The blood feedback rate can characterize whether the working behavior of the blood collection personnel is proceeding normally; the behavior feedback pressure reflects the pressure on the blood collection needle in the blood vessel. The change of its pressure is related to factors such as the elasticity of the blood vessel wall and blood flow. By monitoring this data of the working behavior of the blood collection personnel, the blood vessel state and the stability of the working behavior of the blood collection personnel can be judged, providing a reference for quality evaluation; Exemplarily, in this embodiment, it should be noted that for the working process of the blood collection personnel, the end state is the state of needle withdrawal. The end - state angle rate reflects the smoothness and standardization of the behavior of the blood collection personnel in the end (needle - withdrawal) state. It should be noted that a reasonable angle change rate helps to reduce the damage to blood vessels and surrounding tissues. Exemplarily, if the angle change rate is too large, it will cause complications such as blood vessel tearing or bleeding. In this embodiment, monitoring this data can provide behavioral feedback for the behavior monitoring of the blood collection personnel; the end - state pressure reflects the magnitude and change of the resistance received by the behavior of the blood collection personnel in the end (needle - withdrawal) state. The change in pressure indicates whether there are abnormal situations such as tissue adhesion caused by the behavior of the blood collection personnel, and can help the blood collection personnel judge whether the behavior in the end (needle - withdrawal) state is smooth.

[0017] Preferably, the multi - modal data acquisition unit collects multi - state raw data of the working behavior of the blood collection personnel through intelligent hardware devices, including: action (puncture) state data, standby (disinfection) state data, mixing state data, and characteristic data of the blood - donor; The puncture state data includes: behavior angle, behavior pressure (puncture force), behavior pressure change rate, infrared blood vessel displacement data, and behavior feedback time (blood return time); The disinfection state data includes: calculating the disinfection coverage rate through a visual recognition algorithm; The mixing state data includes: the number of mixing times and the oscillation intensity; The characteristic data of the blood - donor includes: blood vessel visibility; The data pre - processing unit is used to run the following strategies: A21: Jointly align the pressure curve with the angle sensor data and the infrared blood vessel displacement data, normalize the behavior start angle, behavior pressure, and infrared blood vessel displacement data, construct a three - dimensional feature vector, and import the three - dimensional feature vector into the differential distance calculation strategy to obtain the differential distance of the three - dimensional feature vector; Preferably, the differential distance calculation strategy is specifically: ; where, represents the value of the k - th dimension feature at times i and j; is the weight of the k - th dimension feature; Exemplarily, in this embodiment, ; is the standard deviation of the historical data of the k - th dimension feature; A22: Extract the difference distance of the three-dimensional feature vectors, construct a difference distance matrix, and use the dynamic time warping (DTW) path search algorithm to find the optimal path from the starting point to the ending point; It should be noted that the optimal path is the mapping relationship between two time series, which indicates how to minimize the morphological differences between the two sequences through alignment (stretching / compressing); A23: Through the optimal path, establish a time mapping table between the behavior start pressure curve and the behavior feedback (blood return) state. Through the gradient detection algorithm, find the local minimum point of the difference distance in the pressure sequence after path mapping, and mark it as the peak moment of the behavior start pressure curve; Exemplarily, in this embodiment, it should be noted that the local minimum point of the difference distance corresponds to the maximum resistance point of the blood collection personnel's work behavior; Calculate the blood vessel elasticity parameter of the person being blood collected through the time mapping table ; ; Among them, is the starting moment of the behavior feedback (blood return) state of the blood collection personnel, is the peak moment of the behavior start pressure curve of the blood collection personnel.

[0018] Exemplarily, in this embodiment, it should be noted that DTW in the prior art only aligns a single pressure curve. However, in the actual work process of blood collection personnel, there is a dynamic association among the behavior start pressure, behavior start angle, and blood vessel displacement. Moreover, the blood vessel conditions of different patients vary greatly, and fixed parameters are prone to misjudgment. Therefore, using three-dimensional feature vectors for joint alignment can identify some special situations in the work of blood collection personnel. For example, it can identify the situation where the force is excessive but the angle is correct. Another example is that for a person being blood collected with a relatively hard blood vessel, the pressure curve may be normal, but the blood vessel displacement amount will be significantly reduced. Joint alignment through three-dimensional feature vectors can early warn the work quality of blood collection personnel.

[0019] The redundant data removal unit is used to run the following strategy: A31: Quantify the correlation degree between the disinfection coverage rate and the blood vessel visibility through a probability algorithm, and calculate and obtain the mutual information value between the two; The calculation strategy of the mutual information value is: ; Among them, are the disinfection coverage rate and the blood vessel visibility respectively; represents within the time window (30 seconds) and is the joint probability density; respectively represent within the time window , The marginal probability density of A32: Extract the mutual information value, compare the mutual information value with a preset correlation threshold. When the mutual information value is greater than the preset correlation threshold, it is determined that there is a high correlation between the disinfection coverage rate and the blood vessel visibility. Before extracting the key quality parameters, redundant parameters are removed.

[0020] Exemplarily, in this embodiment, a strategy for removing redundant data is given. Specifically, when the calculated mutual information value is greater than 0.3 and it is determined that the visibility of blood vessels is reduced due to the reflection of the disinfected area, the core feature reflecting the blood vessel condition of the blood donor, that is, the blood vessel visibility, is preferentially retained, and the strongly correlated disinfection coverage rate is discarded, reducing data redundancy and improving the calculation efficiency of the system; The key quality parameter extraction module is used to extract multi-dimensional quality parameters from the time-series correlation data, construct a hemolysis quality model, output hemolysis quality parameters, and screen and construct a key quality feature set through a spatio-temporal feature matrix and transfer entropy; The key quality parameter extraction module includes: a hemolysis quality prediction unit and a key quality feature set screening unit; The hemolysis quality prediction unit is configured with the following strategy: B11: Construct a red blood cell shear damage prediction model, calculate the shear stress on red blood cells of the blood donor during the mixing operation through a shear stress calculation strategy, and input the shear stress into the red blood cell shear damage prediction model to output the cumulative shear damage ; Preferably, the shear stress calculation strategy is: ; Where is the shear stress on red blood cells of the blood donor during the mixing operation; K is the blood consistency of the blood donor (characterizing the fluid viscosity feature); is the oscillation intensity of the blood donor during the mixing operation; is the apparent blood viscosity (related to G); r is the non-Newtonian index; Exemplarily, in this embodiment, ; Preferably, the output formula of the red blood cell shear damage prediction model is: ; Where is the cumulative shear damage; is the critical rupture stress; is the indicator function. When , is 1, otherwise it is 0; B12: Incorporate the number of action states of the blood collection personnel , the tourniquet usage time and the cumulative amount of shear damage into the hemolysis quality prediction strategy to predict the hemolysis quality parameters of the blood collection personnel's work behavior ; The specific hemolysis quality prediction strategy is as follows: .

[0021] The key quality feature set screening unit is configured with the following strategy: B21: Obtain the multi-state raw data of the 12-dimensional work behavior of the blood collection personnel, establish a spatio-temporal feature matrix, extract the statistical features of mean, variance, and peak value through a sliding window, and form a feature vector containing time-series dynamic information, where the sliding window is 5s; Exemplarily, in this embodiment, the multi-state raw data of the 12-dimensional work behavior of the blood collection personnel includes: the number of behavioral actions, the starting angle of the behavior, the disinfection coverage rate, the tourniquet usage time, the behavior pressure change rate, the mixing and oscillation intensity; the blood vessel visibility, the infrared blood vessel displacement, the age of the blood donor, the BMI index, the diameter of the blood collection needle, and the working environment temperature of the blood collection personnel; B22: Analyze the directional influence between each element in the 12-dimensional spatio-temporal feature matrix and the hemolysis quality parameters through the transfer entropy algorithm to obtain a transfer entropy sequence, and screen the multi-state raw data corresponding to the first 50% of the transfer entropy sequence as the key quality feature set.

[0022] Exemplarily, in this embodiment, a specific calculation strategy for transfer entropy is provided as follows: ; It should be noted that t is the current time point; X and Y are two data variables whose correlation needs to be analyzed in the multi-state raw data of the 12-dimensional work behavior of the blood collection personnel; represents the state of data variable X at time t; respectively represent the states of data variable Y at time t and time t + 1; is the joint probability, representing and the probability that the three states appear simultaneously; is the conditional probability, predicting the future state of Y given the historical state of Y and the current state of X ; It should be noted that if X has a causal effect on Y, this probability should be higher than the predictive ability using only the history of Y itself.

[0023] is the conditional probability, representing using only the historical state of Y to predict the future state of Y (i.e., autoregressive prediction).

[0024] The quality prediction module is used to construct a knowledge - random forest model, integrate the key quality feature set into model training, and output the quality prediction result of the work achievement and the mechanical damage index; The quality prediction module includes: a prior knowledge embedding unit, a model structure optimization unit, and a quality result output unit; The prior knowledge embedding unit is configured with the following operating strategy: Construct a random forest model. During the construction of the decision trees in the random forest, rewrite the information gain calculation function by reconstructing weights. During the growth of the decision trees, perform split feature priors to obtain a knowledge - random forest model; Among them, the split feature prior includes: when the node satisfies the condition that the blood vessel visibility is less than 0.3, use the behavioral pressure of the blood collection personnel and the number of action states of the blood collection personnel as split features; Preferably, the rewriting of the information gain calculation function specifically includes: when the input feature of the decision tree node includes blood vessel visibility and the blood vessel visibility is less than 0.3, reconstruct the weights corresponding to the behavioral pressure of the blood collection personnel and the number of action states (puncture times) of the blood collection personnel to 1.2 times the initial weight; The model structure optimization unit is configured with the following operating strategy: C21: Add supplementary feature parameters to the key quality feature set to obtain 10 - dimensional feature parameters. The supplementary feature parameters include: the average value of the starting angle of the behavior, the variance of the starting angle of the behavior, the standard deviation of the oscillation intensity, and the blood vessel elasticity parameter of the blood donor; C22: Expand the input layer of the knowledge - random forest model, and at the same time add the calculation of the mechanical damage index to the output layer of the knowledge - random forest model; Among them, the expansion of the input layer includes: divide the 10 - dimensional feature parameters into three feature categories: static, time - series, and event, and perform feature integration; Preferably, the mechanical damage index The calculation strategy is: ; Among them, are the pressure damage coefficient and the oscillation change rate coefficient respectively; is the behavioral stress of blood collection personnel; Exemplarily, in this embodiment, it should be noted that the formula for the strain energy of the red blood cell membrane is: , the strain energy (leading to rupture) is proportional to the square of the pressure, so the square of the behavioral stress is used to characterize the non-linear damage; where U is the strain energy of the red blood cell membrane, which refers to the elastic potential energy stored by the red blood cell membrane due to deformation (tension, compression or shear), and is the energy source leading to membrane structure damage and even rupture. It should be noted that when red blood cells pass through narrow blood vessels (such as capillaries) or are subjected to fluid shear force, the membrane deforms and accumulates strain energy. If the energy exceeds the tolerance threshold of the membrane, the membrane will rupture (such as hemolysis).

[0025] is the stiffness coefficient of the red blood cell membrane, which characterizes the elastic properties of the red blood cell membrane and reflects the ability of the membrane to resist deformation; is the relative deformation of the membrane; its relationship with the force per unit area acting on the membrane is , according to Hooke's law, within the elastic range, the strain force and stress are proportional, so the strain energy (leading to rupture) is proportional to the square of the pressure; Exemplarily, in this embodiment, it should be noted that a rapidly changing shear rate (high ) causes a sudden drop in local viscosity, making the cells more easily torn and reducing the quality of the work results of blood collection personnel; The quality warning module is used to calculate the real-time hemolysis probability value in real time based on edge computing, dynamically adjust the quality warning threshold of the work results, and perform hierarchical quality warning on the work results through hardware interaction and generate work result disposal suggestions.

[0026] The quality result output unit is configured with the following operation strategies: C31: Perform the same standardization processing on the spatio-temporal feature vectors in the verification dataset as on the training data; Specifically, the verification dataset includes: real labels including hemolysis probability, coagulation markers, contamination levels, and corresponding 10-dimensional spatio-temporal feature vectors; C32: Input the preprocessed feature vectors into the knowledge-random forest model, and based on the trained parameters and decision rules, output the three-dimensional quality index prediction results and the mechanical damage index.

[0027] Preferably, the three-dimensional quality index prediction results include: hemolysis probability, coagulation markers, and contamination levels; Exemplarily, in this embodiment, the labels in the knowledge-random forest model include: hemolysis probability, coagulation markers, and contamination levels; The value range of the hemolysis probability is [0,1], indicating the possibility of hemolysis of the blood collection sample; The coagulation marker is a binary variable with values of normal or abnormal, and is used to judge the coagulation state of the blood collection sample; The contamination level is a three-category variable with values of low, medium, and high, reflecting the degree of contamination of the blood collection sample; The quality warning module includes: a real-time hemolysis probability calculation unit and a disposal suggestion output unit; The real-time hemolysis probability calculation unit is used to calculate and obtain the real-time hemolysis probability value of the work result under the multi-state work behavior of the blood collection personnel; Preferably, the real-time hemolysis probability value The calculation strategy is: ; Among them, is the number of action behaviors of the blood collection personnel; The deviation degree of the starting angle of the blood collection personnel's behavior; It should be noted that. reflects the amplification effect of rapid needle insertion on the hemolysis probability under low vascular visibility; It should be noted that, represents the mixing stability penalty, is the standard deviation of the oscillation intensity, and the larger its value, the more unstable the mixing operation of the blood collection personnel; The disposal suggestion output unit is configured with the following operation strategy: D11: Extract the real-time hemolysis probability value, the vascular visibility and vascular elasticity parameters of the blood collection personnel, and the behavior pressure change rate and the number of action states of the blood collection personnel; D12: According to D11, perform quality risk warning on the work results of the blood collection personnel, including: When the vascular visibility is less than 0.3 and the vascular elasticity parameter is less than 50 ms, ignore the warning threshold of the real-time hemolysis probability value, directly mark the work result of the blood collection personnel as a yellow warning, and output the disposal suggestion of giving priority to the quality inspection of the work result storage and increasing the microscopic examination of red blood cell morphology; When the number of action states of the blood collection personnel is greater than or equal to 2 and the behavior pressure change rate of the blood collection personnel > 1.5 N / s, ignore the warning threshold of the real-time hemolysis probability value, directly trigger a red warning, and output the disposal suggestion of re-performing the blood collection work; When the real-time hemolysis probability value is less than 0.3 and there are no spatio-temporal abnormal features, directly mark the work result of the blood collection personnel as a green warning and output the disposal suggestion of being able to be transported normally.

[0028] Embodiment 2: This embodiment provides an electronic device, including: a processor and a memory, wherein the memory stores a computer program that can be called by the processor; The processor runs the above intelligent blood collection personnel behavior monitoring system by calling the computer program stored in the memory.

[0029] This electronic device may have relatively large differences due to different configurations or performances, and can include one or more processors (Central Processing Units, CPU) and one or more memories. Among them, at least one computer program is stored in the memory, and this computer program is loaded and run by the processor to implement an intelligent blood collection personnel behavior monitoring system provided by the above method embodiment. This electronic device can also include other components for implementing device functions. For example, this electronic device can also have components such as wired or wireless network interfaces and input / output interfaces for data input and output. This embodiment will not be elaborated here.

[0030] Embodiment 3: This embodiment proposes a computer-readable storage medium, on which a rewritable computer program is stored; When the computer program runs on a computer device, it causes the computer device to run the above intelligent blood collection personnel behavior monitoring system.

[0031] For example, the computer-readable storage medium can be a read-only memory (abbreviation: ROM), a random access memory (abbreviation: RAM), a compact disc read-only memory (abbreviation: CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0032] It should be understood that in various embodiments of the present application, the magnitude of the sequence numbers of the above processes does not mean the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0033] It should be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0034] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions according to the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired network or / and a wireless network. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that includes one or more collections of available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0035] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in the present invention can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. A professional person can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0036] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0037] In several embodiments provided by the present invention, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of units is only one way, and there can be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings, direct couplings, or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in an electrical, mechanical, or other form.

[0038] The unit described as a separation component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place or may be distributed across multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0039] In addition, in each embodiment of the present invention, each functional unit may be integrated in a processing unit, may exist physically separately for each unit, or two or more units may be integrated in one unit.

[0040] In the description of this specification, the description with reference to terms such as "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.

[0041] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and the description in the specification only illustrate the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent behavior monitoring system for blood collection personnel, characterized in that, The system includes: a multimodal data processing module, a key quality parameter extraction module, a quality prediction module and a quality early warning module; The multimodal data processing module collects multi-state raw data of blood collection, disinfection and mixing of blood collection personnel's work behavior through multi-sensor temporal and spatial synchronization, and generates a time series correlation data set through DTW time series alignment and mutual information verification; The key quality parameter extraction module is used to extract multi-dimensional quality parameters from time series correlation data, construct a hemolysis quality model, output hemolysis quality parameters, and construct a key quality feature set through spatiotemporal feature matrix and transfer entropy screening; The quality prediction module is used to construct a knowledge-random forest model, integrate key quality feature sets into model training, and output work product quality prediction results and mechanical damage index; The quality warning module is used to calculate the real-time hemolysis probability value based on edge computing, dynamically adjust the work result quality warning threshold, perform graded quality warning of the work results through hardware interaction, and generate work result disposal suggestions.

2. The intelligent blood collection personnel behavior monitoring system according to claim 1, wherein, The multimodal data alignment module is configured with a data acquisition architecture construction unit, a multimodal data acquisition unit, a data preprocessing unit and a redundant data removal unit; The data preprocessing unit is used to run the following strategy: A21: jointly align the pressure curve with the angle sensor data and the infrared vascular displacement data, normalize the behavior start angle, behavior pressure and infrared vascular displacement data, construct a three-dimensional feature vector, import the three-dimensional feature vector into the difference distance calculation strategy, and obtain the difference distance of the three-dimensional feature vector; A22: Extract the difference distance of the three-dimensional feature vectors, construct the difference distance matrix, and use the dynamic time warp path search algorithm to find the optimal path from the start point to the end point; A23: Through the optimal path, a time mapping table of the behavior start pressure curve and the behavior feedback state is established. Through the gradient detection algorithm, the local minimum point of the difference distance is found in the pressure sequence after the path mapping, and it is marked as the peak moment of the behavior start pressure curve; Calculate the vascular elasticity parameters of the blood donor through the time mapping table ; ； Among them, is the starting moment of the behavior feedback status of the blood collection personnel, is the peak moment of the starting pressure curve of the behavior of the blood collection personnel.

3. An intelligent blood collection personnel behavior monitoring system according to claim 2, wherein, The redundant data removal unit is used to run the following strategy: A31: The correlation between disinfection coverage and vascular visibility was quantified by probability algorithm, and the mutual information value between the two was calculated; The calculation strategy of the mutual information value is: ； Among them, are the disinfection coverage rate and the blood vessel visibility respectively; represent the joint probability density within the time window and ; respectively represent the marginal probability density within the time window , ; A32: Extract the mutual information value and compare it with the preset correlation threshold. When the mutual information value is greater than the preset correlation threshold, it is judged that the disinfection coverage is highly correlated with the vascular visibility. Before extracting the key quality parameters, the redundant parameters are eliminated.

4. The intelligent blood collection personnel behavior monitoring system according to claim 3, characterized in that, The key quality parameter extraction module includes: a hemolysis quality estimation unit and a key quality feature set screening unit; The hemolysis quality estimation unit is configured with the following strategy: B11: Construct a prediction model for red blood cell shear damage. Calculate the shear stress on red blood cells during the mixing operation of the blood collection personnel through a shear stress calculation strategy, and input the shear stress into the prediction model for red blood cell shear damage to output the cumulative shear damage ; B12: The number of action states of the blood collection personnel , the tourniquet usage time and the cumulative amount of shear damage are imported into the hemolysis quality prediction strategy to predict the hemolysis quality parameters of the blood collection personnel's work behavior ; The hemolysis quality estimation strategy is specifically as follows: 。 5. An intelligent blood collection personnel behavior monitoring system according to claim 4, characterized in that, The key quality feature set screening unit is configured with the following strategies: B21: Obtain the multi-state raw data of the working behavior of blood collectors in 12 dimensions, establish the spatiotemporal feature matrix, extract the statistical features of the mean, variance and peak value through the sliding window, and form a feature vector containing time series dynamic information, where the sliding window is 5s; B22: Analyze the directional influence between each element in the 12-dimensional spatio-temporal feature matrix and the hemolysis quality parameter through the transfer entropy algorithm, obtain the transfer entropy sequence, and screen the first 50% of the transfer entropy sequence corresponding to the multi-state raw data as the key quality feature set. Between them, obtain the transfer entropy sequence, and screen the first 50% of the transfer entropy sequence corresponding to the multi-state raw data as the key quality feature set.

6. An intelligent blood collection personnel behavior monitoring system according to claim 5, wherein The quality prediction module includes: a prior knowledge embedding unit, a model structure optimization unit, and a quality result output unit; The prior knowledge embedding unit is configured with the following operating strategy: Construct a random forest model. During the construction of the decision trees in the random forest, rewrite the information gain calculation function by reconstructing weights. During the growth of the decision trees, perform split feature priors to obtain a knowledge-random forest model; Among them, the split feature priors include: when the node satisfies the condition that the vascular visibility is less than 0.3, use the behavioral pressure of the blood collection personnel and the number of action states of the blood collection personnel as split features; The model structure optimization unit is configured with the following operating strategy: C21: Add supplementary feature parameters to the critical quality feature set to obtain 10-dimensional feature parameters. The supplementary feature parameters include: the average value of the starting angle of the behavior, the variance of the starting angle of the behavior, the standard deviation of the oscillation intensity, and the vascular elasticity parameter of the blood donor during the historical work process of the current blood collection personnel; C22: Expand the input layer of the knowledge-random forest model and add the calculation of the mechanical damage index to the output layer of the knowledge-random forest model; Among them, the expansion of the input layer includes: dividing the 10-dimensional feature parameters into three feature categories: static, time series, and event, and performing feature integration.

7. The intelligent blood collection personnel behavior monitoring system according to claim 6, wherein, The quality result output unit is configured with the following operating strategy: C31: Perform the same standardization processing on the spatio-temporal feature vectors in the validation dataset as on the training data; C32: Input the preprocessed feature vectors into the knowledge-random forest model, and based on the trained parameters and decision rules, output the three-dimensional quality index prediction result and the mechanical damage index.

8. An intelligent blood collection personnel behavior monitoring system according to claim 7, characterized in that, The quality warning module includes: a real-time hemolysis probability calculation unit and a disposal suggestion output unit; The real-time hemolysis probability calculation unit is used to calculate and obtain the real-time hemolysis probability value of the work results under the multi-state work behaviors of the blood collection personnel; The disposal suggestion output unit is configured with the following operating strategy: D11: Extract the real-time hemolysis probability value, the vascular visibility and vascular elasticity parameter of the blood donor, and the change rate of the behavioral pressure and the number of action states of the blood collection personnel; D12: According to D11, perform quality risk warnings on the work results of the blood collection personnel, including: When the vascular visibility is less than 0.3 and the vascular elasticity parameter is less than 50 ms, ignore the warning threshold of the real-time hemolysis probability value, directly mark the work results of the blood collection personnel as a yellow warning, and output the disposal suggestion of giving priority to the quality inspection of the work results in storage and increasing the microscopic examination of red blood cell morphology; When the number of action states of the blood collection personnel is greater than or equal to 2 and the change rate of the behavioral pressure of the blood collection personnel > 1.5 N / s, ignore the warning threshold of the real-time hemolysis probability value, directly trigger a red warning, and output the disposal suggestion of re-performing the blood collection work; When the real-time hemolysis probability value is less than 0.3 and there are no spatio-temporal abnormal features, directly mark the work results of the blood collection personnel as a green warning and output the disposal suggestion of normal transportation.

Images