A method for controlling the concentration of platelet-rich plasma by suspension centrifugation

Abstract

The present invention discloses a method for controlling the concentration of platelet-rich plasma by suspended centrifugal separation, which belongs to the technical field of data analysis and processing. The method uses deep learning technology to learn historical data in the platelet separation process, thereby determining the nonlinear separation law in the platelet separation process, and then controls the subsequent platelet separation process according to the acquired nonlinear separation law, thereby achieving precise control of the platelet-rich plasma concentration, so as to improve the efficiency and accuracy of platelet separation. The method is suitable for clinical medicine, bioengineering and other fields.

Description

Technical Field

[0001] The invention belongs to the technical field of data analysis and processing, and in particular relates to a method for controlling the concentration of platelet-rich plasma separated by suspension centrifugation. Background Art

[0002] Platelet-Rich Plasma (PRP) is a type of plasma that contains high concentrations of platelets extracted from autologous blood by centrifugation. It is rich in a variety of growth factors and proteins. Platelets are key components of blood coagulation and hemostasis, and their effectiveness is extremely important in clinical applications in the medical field.

[0003] Platelet-rich plasma is of great value in clinical treatment. In the field of orthopedics, it can promote bone tissue repair and regeneration, and effectively treat fracture nonunion, bone defects and other diseases; in oral and maxillofacial surgery, it can accelerate wound healing and promote periodontal tissue regeneration; in the field of cosmetology, it can stimulate skin collagen production, improve skin texture and reduce wrinkles. At the same time, platelets also play an important role in clinical fields such as blood diseases and trauma treatment.

[0004] However, in the process of platelet separation, accurately controlling the concentration of platelet-rich plasma has always been a technical challenge. Existing technologies mainly rely on manual secondary operation configuration after collection to meet clinical concentration requirements, and rely on experience judgment, which has problems such as low efficiency and large errors. Therefore, it is of great significance to develop an efficient and accurate platelet separation concentration control method. Summary of the invention

[0005] The present invention provides a method for controlling the concentration of platelet-rich plasma separated by suspension centrifugation, which is used to solve the problems of low efficiency, large error, etc. in the prior art that the clinical concentration requirement and empirical judgment are controlled by manual secondary operation configuration after collection.

[0006] A method for controlling the concentration of platelet-rich plasma by suspension centrifugation, comprising:

[0007] Acquire historical platelet separation concentration data; wherein the historical platelet separation concentration data includes historical platelet-rich plasma separation system parameters and historical platelet-rich plasma concentrations acquired based on the historical platelet-rich plasma separation system parameters;

[0008] A platelet separation concentration prediction model is constructed by using a deep learning algorithm, and the platelet separation concentration prediction model is trained using historical platelet-rich plasma separation system parameters and historical platelet-rich plasma concentrations obtained based on the historical platelet-rich plasma separation system parameters to obtain a trained platelet separation concentration prediction model;

[0009] Obtaining a target platelet-rich plasma concentration input by a medical staff, and encoding the platelet-rich plasma separation system parameters multiple times to determine multiple system parameter codes;

[0010] Using the system parameter code as a platelet separation concentration prediction model after training, determining the predicted platelet-rich plasma concentration corresponding to each system parameter code, and determining the loss function value corresponding to each system parameter code according to the target platelet-rich plasma concentration and the predicted platelet-rich plasma concentration;

[0011] With the goal of minimizing the loss function value, an intelligent optimization algorithm is used to optimize multiple system parameter codes, determine the global optimal system parameter code, and use the global optimal system parameter code as the final platelet-rich plasma separation system parameter;

[0012] Based on the final platelet-rich plasma separation system parameters, the platelet-rich plasma separation system operation is controlled, and the platelet-rich plasma separation system parameters are adjusted in real time to achieve platelet separation concentration control.

[0013] Further, the historical platelet-rich plasma separation system parameters include current platelet-rich plasma concentration, separation speed, total separation time, and elapsed separation time;

[0014] The historical platelet-rich plasma separation system parameters include the platelet-rich plasma separation system parameters at the initial moment of separation and the platelet-rich plasma separation system parameters during the separation process.

[0015] Furthermore, a platelet separation concentration prediction model is constructed by a deep learning algorithm, including: using a BP neural network algorithm to construct a platelet separation concentration prediction model.

[0016] Furthermore, the platelet separation concentration prediction model is trained using historical platelet-rich plasma separation system parameters and historical platelet-rich plasma concentrations obtained based on the historical platelet-rich plasma separation system parameters to obtain a trained platelet separation concentration prediction model, including:

[0017] Initializing model parameters of a platelet separation concentration prediction model, encoding the model parameters, determining the model parameter encoding, and obtaining a plurality of model parameter encodings;

[0018] For any model parameter code, after applying the model parameters contained in the model parameter code to the platelet separation concentration prediction model, using the historical platelet-rich plasma separation system parameters as the input of the platelet separation concentration prediction model, using the platelet-rich plasma concentration as the expected output, and obtaining the error function value corresponding to the model parameter code;

[0019] Obtain the optimal model parameter encoding according to the error function value corresponding to the model parameter encoding;

[0020] For the model parameter coding, a random matching sinusoidal information interaction strategy is used to perform neighborhood exploration on the model parameter coding, and the model parameter coding after the neighborhood exploration is determined;

[0021] For the model parameter coding after neighborhood exploration, according to the optimal model parameter coding, an information sharing strategy combined with random jumps is used to perform adaptive jump exploration on the model parameter coding to obtain the model parameter coding after adaptive jump exploration;

[0022] For the model parameter encoding after adaptive jump exploration, the information interaction strategy affected by the boundary is used to explore the unfamiliar area of ​​the model parameter encoding, and the model parameter encoding after the unfamiliar area exploration is obtained;

[0023] For the model parameter coding after exploring the unfamiliar area, according to the optimal model parameter coding, an adaptive optimal guidance strategy is used to quickly search the model parameter coding in the optimal direction to obtain the model parameter coding after the rapid search;

[0024] Determine whether the current number of training times is equal to or greater than the maximum number of training times. If so, re-obtain the optimal model parameter coding based on the model parameter coding after the quick search, and use the optimal model parameter coding as the final parameter of the platelet separation concentration prediction model to obtain the platelet separation concentration prediction model after training. Otherwise, return to the step of obtaining the error function value corresponding to the model parameter coding.

[0025] Furthermore, for the model parameter coding, a random matching sinusoidal information interaction strategy is used to perform neighborhood exploration on the model parameter coding, and the model parameter coding after the neighborhood exploration is determined, including:

[0026] For any model parameter code, randomly match the model parameter code with another model parameter code to obtain a first target model parameter code;

[0027] The model parameter code is obtained to perform information interaction with the corresponding first target model parameter code, and the first information interaction item is determined as: ;in, Indicates the tth training process i Model parameter encoding, Indicates i Model parameter encoding The corresponding first target model parameter encoding, i =1,2,...,L, L represents the total number of model parameter encodings;

[0028] The first information interaction item is processed by using a sine function, and the first information interaction item after processing is obtained as follows: ;in, represents the first random number in (0,1), represents pi, and sin represents the sine function;

[0029] The model parameter encoding is subjected to neighborhood exploration according to the first information interaction item, and the model parameter encoding after the neighborhood exploration is determined to be: ;in, Represents the model parameter encoding after neighborhood exploration .

[0030] Furthermore, for the model parameter coding after the neighborhood exploration, according to the optimal model parameter coding, an information sharing strategy combined with random jumps is used to perform adaptive jump exploration on the model parameter coding, and the model parameter coding after the adaptive jump exploration is obtained, including:

[0031] Get the random jump coefficient as: ;in, represents the random jump coefficient, Indicates compliance N (0, ) a second random number from a normal distribution, Indicates compliance N (0,1) The third random number from normal distribution, represents the intermediate parameter, represents a constant term, and , represents the gamma function, and π represents the circumference of a circle;

[0032] Based on the current training times, the random jump intensity is determined as: ;in, represents the random jump intensity, represents the fourth random number in (0,1), and T represents the preset maximum number of training times;

[0033] For any model parameter encoding after neighborhood exploration, randomly match the model parameter encoding with another model parameter encoding to obtain a second target model parameter encoding;

[0034] For the model parameter encoding after any neighborhood exploration, the random jump term is determined according to the second target model parameter encoding, the random jump coefficient and the random jump strength: ;in, Indicates n The model parameter encoding after neighborhood exploration, n =1,2,...,L, L represents the total number of model parameter encodings, Indicates n Model parameter encoding after neighborhood exploration The corresponding second target model parameter encoding;

[0035] According to the random jump term and the optimal model parameter encoding, the model parameter encoding is adaptively jumped and explored, and the model parameter encoding after the adaptive jump exploration is obtained as follows: ;in, Represents the model parameter encoding after adaptive jump exploration , represents the fifth random number in (0,1), represents the sixth random number in (0,1), Represents the optimal model parameter encoding.

[0036] Furthermore, for the model parameter encoding after the adaptive jump exploration, the information interaction strategy affected by the boundary is used to perform unfamiliar area exploration on the model parameter encoding, and the model parameter encoding after the unfamiliar area exploration is obtained, including:

[0037] Based on the current number of training times, the boundary influence factor is obtained as: ;in, represents the boundary impact factor, t represents the current number of training times, T represents the maximum number of training times, and L represents the total number of model parameter encodings;

[0038] According to the boundary impact factor, the boundary impact code is obtained as: ;in, Indicates that the boundary affects the encoding, ub The upper limit encoding representing the upper limit composition of all dimensional model parameters, lb The lower bound encoding representing the lower bound composition of all dimensional model parameters;

[0039] For any model parameter encoding after the adaptive jump exploration, randomly match another model parameter encoding for the model parameter encoding to obtain a third target model parameter encoding;

[0040] According to the third target model parameter coding and the boundary influence coding, the model parameter coding is explored in the unfamiliar area, and the model parameter coding after the unfamiliar area exploration is obtained as follows: ;in, Indicates t The model parameter encoding after the mth adaptive jump exploration in the training process, represents the sixth random number in (0,1), Represents the model parameter encoding after exploring unfamiliar areas , represents the seventh random number in (0,1), Represents model parameter encoding The corresponding third target model parameter encoding, Represents the eighth random number in (0,1).

[0041] Furthermore, for the model parameter coding after the exploration of the unfamiliar area, according to the optimal model parameter coding, an adaptive optimal guidance strategy is used to quickly search the model parameter coding in the optimal direction, and the model parameter coding after the rapid search is obtained, including:

[0042] Based on the current number of training times, the adaptive inertia weight factor is determined as: ;in, Indicates t The adaptive inertia weight factor during the training process, represents the maximum value corresponding to the adaptive inertia weight factor, represents the minimum value corresponding to the adaptive inertia weight factor, cos represents the cosine function, represents pi, T represents the maximum number of training times;

[0043] According to the optimal model parameter encoding and the adaptive inertia weight factor, the adaptive update speed is determined as: ;in, Indicates t During the training process k The update speed of the model parameter encoding after exploring a strange area, Indicates t +1 training session k The update speed of the model parameter encoding after exploring a strange area, represents the first pilot coefficient, represents the second guidance coefficient, c represents the guidance range control factor, Indicates t During the training process k The model parameter encoding after exploring unfamiliar areas, Represents model parameter encoding The corresponding historical optimal value is dsit kp Represents model parameter encoding The best historical value The Euclidean distance between them, e represents a natural constant, represents the optimal model parameter encoding, dsit kbest Represents model parameter encoding Encoding with optimal model parameters The Euclidean distance between

[0044] According to the update speed , perform a quick search on the model parameter encoding in the optimal direction, and obtain the model parameter encoding after the quick search as: ;in, Indicates the model parameter encoding after fast search , represents the balance factor, and , sin represents the sine function.

[0045] Further, the target platelet-rich plasma concentration input by the medical staff is obtained, and the platelet-rich plasma separation system parameters are encoded multiple times to determine multiple system parameter codes, including:

[0046] Obtain target platelet-rich plasma concentration entered by healthcare provider;

[0047] The platelet-rich plasma separation system parameters are randomly generated within the upper and lower limits of the platelet-rich plasma separation system parameters, and the generated platelet-rich plasma separation system parameters are combined into a system parameter code, and a plurality of different system parameter codes are repeatedly obtained.

[0048] Furthermore, with the goal of minimizing the loss function value, an intelligent optimization algorithm is used to optimize multiple system parameter encodings to determine the global optimal system parameter encoding, including:

[0049] According to the loss function value, determine the system parameter encoding with the smallest loss function value as the optimal system parameter encoding;

[0050] Based on the optimal system parameter encoding, a particle swarm optimization algorithm is used to optimize the system parameter encoding;

[0051] The system parameter encoding is optimized repeatedly until the optimization end condition is met, and the global optimal system parameter encoding is determined.

[0052] The present invention provides a method for controlling the concentration of platelet-rich plasma by suspended centrifugal separation. The method uses deep learning technology to learn historical data in the platelet separation process, thereby determining the nonlinear separation law in the platelet separation process, and then controls the subsequent platelet separation process with the acquired nonlinear separation law, thereby achieving precise control of the platelet-rich plasma concentration, thereby improving the efficiency and accuracy of platelet separation, and is suitable for clinical medicine, bioengineering and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0054] Figure 1A flow chart of a method for controlling the concentration of platelet-rich plasma by suspension centrifugation provided in an embodiment of the present invention.

[0055] Figure 2 A schematic diagram of a suspension centrifugation method provided in an embodiment of the present invention.

[0056] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0057] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0058] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0059] like Figure 1 As shown, an embodiment of the present invention provides a method for controlling the concentration of platelet-rich plasma by suspension centrifugation, comprising:

[0060] S101, obtaining historical platelet separation concentration data; wherein the historical platelet separation concentration data includes historical platelet-rich plasma separation system parameters and historical platelet-rich plasma concentrations obtained based on the historical platelet-rich plasma separation system parameters;

[0061] Historical platelet-rich plasma separation system parameters may include previous platelet-rich plasma concentration, separation speed, total separation time and separation time, and may also include other blood parameters of the patient, such as HCT (hematocrit) value, PLT (platelet count) baseline value, PLT multiples, total PRP (platelet-rich plasma) requirement, number of collection layers, etc.

[0062] The historical platelet-rich plasma separation system parameters affect the historical platelet-rich plasma concentration, and the historical platelet-rich plasma concentration is real data. Therefore, the relationship between these data can form prediction rules through deep learning.

[0063] S102, constructing a platelet separation concentration prediction model through a deep learning algorithm, and training the platelet separation concentration prediction model using historical platelet-rich plasma separation system parameters and historical platelet-rich plasma concentrations obtained based on the historical platelet-rich plasma separation system parameters to obtain a trained platelet separation concentration prediction model;

[0064] For example, a convolutional neural network algorithm, a BP (Back Propagation) neural network algorithm or other classification neural network algorithms can be used to construct a platelet separation concentration prediction model, and then the platelet separation concentration prediction model is used to identify the data relationship in the platelet separation concentration historical data, so as to form a prediction rule, that is, to obtain the trained platelet separation concentration prediction model.

[0065] S103, obtaining a target platelet-rich plasma concentration input by a medical staff, and encoding the platelet-rich plasma separation system parameters multiple times to determine multiple system parameter codes;

[0066] The target platelet-rich plasma concentration is the ultimate goal of the platelet separation task, and the platelet-rich plasma concentration is affected by the platelet-rich plasma separation system parameters. Therefore, to achieve accurate control of the platelet-rich plasma concentration, it is necessary to adjust the platelet-rich plasma separation system parameters. During the adjustment process, the concentration can be predicted using the trained platelet separation concentration prediction model, so that a better system parameter setting can be achieved, thereby improving the separation accuracy. Therefore, the embodiment of the present invention encodes the platelet-rich plasma separation system parameters into system parameter encoding, and optimizes based on this to determine the global optimal system parameter encoding.

[0067] S104, using the system parameter code as the platelet separation concentration prediction model after training, determining the predicted platelet-rich plasma concentration corresponding to each system parameter code, and determining the loss function value corresponding to each system parameter code according to the target platelet-rich plasma concentration and the predicted platelet-rich plasma concentration;

[0068] The loss function value corresponding to the system parameter encoding can be the absolute value of the target platelet-rich plasma concentration minus the predicted platelet-rich plasma concentration. The smaller the absolute value is, the better the combination of the system parameter encoding is.

[0069] S105, with the goal of minimizing the loss function value, using an intelligent optimization algorithm to optimize multiple system parameter codes, determine the global optimal system parameter code, and use the global optimal system parameter code as the final platelet-rich plasma separation system parameter;

[0070] For example, an intelligent optimization algorithm such as a particle swarm algorithm can be used to optimize multiple system parameter encodings, so that the global optimal system parameter encoding can be determined quickly and efficiently, and ultimately the control of the platelet-rich plasma separation system can be achieved.

[0071] S106. Based on the final platelet-rich plasma separation system parameters, control the operation of the platelet-rich plasma separation system, and adjust the platelet-rich plasma separation system parameters in real time to achieve platelet separation concentration control.

[0072] The above-mentioned platelet separation concentration historical data may include parameters in the separation process, so that the platelet separation concentration prediction in the separation process can be realized. During the separation process, the platelet separation concentration can be predicted in real time. If the prediction does not meet the standard, the platelet-rich plasma separation system parameters are readjusted to keep the loss function value always minimized, thereby ultimately achieving high-precision separation concentration control.

[0073] The present invention provides a method for controlling the concentration of platelet-rich plasma by suspended centrifugal separation. The method uses deep learning technology to learn historical data in the platelet separation process, thereby determining the nonlinear separation law in the platelet separation process, and then controls the subsequent platelet separation process with the acquired nonlinear separation law, thereby achieving precise control of the platelet-rich plasma concentration, thereby improving the efficiency and accuracy of platelet separation, and is suitable for clinical medicine, bioengineering and other fields.

[0074] In order to facilitate those skilled in the art to better understand the concept of the present invention, Figure 2 A brief introduction to the platelet separation process is given below. Figure 2 As shown, the general workflow of the suspension centrifugation method is: blood collection → centrifugal component separation → plasma collection → other blood cell transfusion → collection completed.

[0075] The specific working method is: the patient's HCT (hematocrit) value, PLT (platelet count) baseline value, PLT multiples, total PRP (platelet-rich plasma) demand, and number of collection layers are automatically input through the equipment. Disposable consumables are installed on the equipment, the equipment is started, the blood donor is punctured, and the equipment safety monitoring system (0#, 1#, 2#, 3# air detectors, pipeline pressure monitors, cuff pressure monitors, etc.) monitors the blood collection process in real time (if there is an abnormality, an alarm will be issued at any time and the machine will automatically shut down). The equipment automatically collects blood, and the whole blood with anticoagulant is collected into the centrifuge cup in the centrifuge to start the preparation of platelet-rich plasma.

[0076] Whole blood is transported to the bottom of the cup through the central pipeline. The centrifuge separates the components according to the different specific gravities of various blood components. Red blood cells are in the outermost layer and plasma is in the innermost layer. The centrifugal cup pressure monitor monitors the pressure in the cup in real time, and uses centrifugal force to push larger component cells outward, leaving plasma in the inner layer. In order to meet the multiple requirements, the software accounting system will automatically open the 2# valve according to the feedback information of the platinum photoelectric and red blood cell detectors, and overflow the excess plasma (PPP) to the 1# electronic scale pulp bag through the overflow channel. After meeting the standard requirements, the 2# valve will automatically close. The centrifugal cup is continuously concentrated by the software control system. The software system platinum photoelectric continuously tracks and monitors the concentration of plasma in the centrifugal cup, and the red blood cell detector continuously tracks the overflow plasma concentration. The information control system is used to control the speed changes of the centrifuge and pump in a timely manner. When the multiple is monitored to meet the set parameters, the 3# valve is automatically opened, and the platelet-rich plasma collection is completed according to the set parameters and the 2# electronic scale is timely feedback. After the collection is completed, the valve automatically closes.

[0077] At the end of the separation phase, the centrifuge cup stops rotating, and the remaining red blood cells and other formed elements in the cup are returned to the blood donor through the central pipeline, thus completing the blood collection work. During the entire collection process, if the blood donor is in discomfort, the clinical doctor can confirm that it is necessary to infuse fluid through the 4# valve.

[0078] In the embodiment of the present invention, the historical platelet-rich plasma separation system parameters include the current platelet-rich plasma concentration, separation speed, total separation time, and separation time;

[0079] The historical platelet-rich plasma separation system parameters include the platelet-rich plasma separation system parameters at the initial moment of separation and the platelet-rich plasma separation system parameters during the separation process.

[0080] In an embodiment of the present invention, a platelet separation concentration prediction model is constructed by a deep learning algorithm, including: using a BP neural network algorithm to construct a platelet separation concentration prediction model.

[0081] In the prior art, the parameters of the BP neural network algorithm are generally optimized by the gradient descent method, which is very easy to fall into the local optimum, resulting in a large deviation in the subsequent prediction of platelet separation concentration and failure to achieve precise control. Therefore, an embodiment of the present invention provides an optimization algorithm to achieve optimal training of the platelet separation concentration prediction model, thereby improving the prediction accuracy of the platelet separation concentration and ultimately improving the control accuracy of the platelet separation concentration.

[0082] In an embodiment of the present invention, a platelet separation concentration prediction model is trained using historical platelet-rich plasma separation system parameters and historical platelet-rich plasma concentrations obtained based on the historical platelet-rich plasma separation system parameters to obtain a trained platelet separation concentration prediction model, including:

[0083] Initializing model parameters of a platelet separation concentration prediction model, encoding the model parameters, determining the model parameter encoding, and obtaining a plurality of model parameter encodings;

[0084] For any model parameter code, after applying the model parameters contained in the model parameter code to the platelet separation concentration prediction model, using the historical platelet-rich plasma separation system parameters as the input of the platelet separation concentration prediction model, and using the platelet-rich plasma concentration as the expected output, the error function value corresponding to the model parameter code is obtained; for example, the root mean square error function value or the cross entropy error function value corresponding to the model parameter code can be obtained.

[0085] Obtain the optimal model parameter encoding according to the error function value corresponding to the model parameter encoding;

[0086] For the model parameter coding, a random matching sinusoidal information interaction strategy is used to perform neighborhood exploration on the model parameter coding, and the model parameter coding after the neighborhood exploration is determined;

[0087] For the model parameter coding after neighborhood exploration, according to the optimal model parameter coding, an information sharing strategy combined with random jumps is used to perform adaptive jump exploration on the model parameter coding to obtain the model parameter coding after adaptive jump exploration;

[0088] For the model parameter encoding after adaptive jump exploration, the information interaction strategy affected by the boundary is used to explore the unfamiliar area of ​​the model parameter encoding, and the model parameter encoding after the unfamiliar area exploration is obtained;

[0089] For the model parameter coding after exploring the unfamiliar area, according to the optimal model parameter coding, an adaptive optimal guidance strategy is used to quickly search the model parameter coding in the optimal direction to obtain the model parameter coding after the rapid search;

[0090] Determine whether the current number of training times is equal to or greater than the maximum number of training times. If so, re-obtain the optimal model parameter coding based on the model parameter coding after the quick search, and use the optimal model parameter coding as the final parameter of the platelet separation concentration prediction model to obtain the platelet separation concentration prediction model after training. Otherwise, return to the step of obtaining the error function value corresponding to the model parameter coding.

[0091] Optionally, after each training, the model parameters may be processed for out-of-bounds errors so that the model parameters always remain valid.

[0092] In an embodiment of the present invention, for model parameter coding, a random matching sinusoidal information interaction strategy is used to perform neighborhood exploration on the model parameter coding, and the model parameter coding after the neighborhood exploration is determined, including:

[0093] For any model parameter code, randomly match the model parameter code with another model parameter code to obtain a first target model parameter code;

[0094] The model parameter code is obtained to perform information interaction with the corresponding first target model parameter code, and the first information interaction item is determined as: ;in, Indicates the tth training process i Model parameter encoding, Indicates i Model parameter encoding The corresponding first target model parameter encoding, i =1,2,...,L, L represents the total number of model parameter encodings;

[0095] The first information interaction item is processed by using a sine function, and the first information interaction item after processing is obtained as follows: ;in, represents the first random number in (0,1), represents pi, and sin represents the sine function;

[0096] The model parameter encoding is subjected to neighborhood exploration according to the first information interaction item, and the model parameter encoding after the neighborhood exploration is determined to be: ;in, Represents the model parameter encoding after neighborhood exploration .

[0097] The random matching sinusoidal information interaction strategy provided in the embodiment of the present invention can enable the model parameter encoding to move forward along the trajectory of a sinusoidal function curve during the information interaction with other model parameter encodings, so that the model parameter encoding can find a better position in the neighborhood and ensure that no repeated exploration is performed, thereby improving the efficiency of neighborhood exploration.

[0098] In an embodiment of the present invention, for the model parameter coding after the neighborhood exploration, according to the optimal model parameter coding, an information sharing strategy combined with random jumps is used to perform adaptive jump exploration on the model parameter coding, so as to obtain the model parameter coding after the adaptive jump exploration, including:

[0099] Get the random jump coefficient as: ;in, represents the random jump coefficient, Indicates compliance N (0, ) a second random number from a normal distribution, Indicates compliance N (0,1) The third random number from normal distribution, represents the intermediate parameter, represents a constant term (which can be set to 1.5, for example), and , represents the gamma function, and π represents the circumference of a circle;

[0100] Based on the current training times, the random jump intensity is determined as: ;in, represents the random jump intensity, represents the fourth random number in (0,1), and T represents the preset maximum number of training times;

[0101] For any model parameter encoding after neighborhood exploration, randomly match the model parameter encoding with another model parameter encoding to obtain a second target model parameter encoding;

[0102] For the model parameter encoding after any neighborhood exploration, the random jump term is determined according to the second target model parameter encoding, the random jump coefficient and the random jump strength: ;in, Indicates n The model parameter encoding after neighborhood exploration, n =1,2,...,L, L represents the total number of model parameter encodings, Indicates n Model parameter encoding after neighborhood exploration The corresponding second target model parameter encoding;

[0103] According to the random jump term and the optimal model parameter encoding, the model parameter encoding is adaptively jumped and explored, and the model parameter encoding after the adaptive jump exploration is obtained as follows: ;in, Represents the model parameter encoding after adaptive jump exploration , represents the fifth random number in (0,1), represents the sixth random number in (0,1), Represents the optimal model parameter encoding.

[0104] The embodiment of the present invention provides an information sharing strategy combined with random jumps, which can learn other position information while learning the optimal position information, and is accompanied by random jumps that vary with the number of training times. In the early stage of the algorithm, random jump search can be performed globally, and the convergence accuracy of the algorithm can also be improved in the later stage of the algorithm.

[0105] In an embodiment of the present invention, for the model parameter coding after the adaptive jump exploration, an information interaction strategy affected by the boundary is used to perform unfamiliar area exploration on the model parameter coding to obtain the model parameter coding after the unfamiliar area exploration, including:

[0106] Based on the current number of training times, the boundary influence factor is obtained as: ;in, represents the boundary impact factor, t represents the current number of training times, T represents the maximum number of training times, and L represents the total number of model parameter encodings;

[0107] According to the boundary impact factor, the boundary impact code is obtained as: ;in, Indicates that the boundary affects the encoding, ub The upper limit encoding representing the upper limit composition of all dimensional model parameters, lb The lower bound encoding representing the lower bound composition of all dimensional model parameters;

[0108] For any model parameter encoding after the adaptive jump exploration, randomly match another model parameter encoding for the model parameter encoding to obtain a third target model parameter encoding;

[0109] According to the third target model parameter coding and the boundary influence coding, the model parameter coding is explored in the unfamiliar area, and the model parameter coding after the unfamiliar area exploration is obtained as follows: ;in, Indicates t The model parameter encoding after the mth adaptive jump exploration in the training process, represents the sixth random number in (0,1), Represents the model parameter encoding after exploring unfamiliar areas , represents the seventh random number in (0,1), Represents model parameter encoding The corresponding third target model parameter encoding, Represents the eighth random number in (0,1).

[0110] The information interaction strategy affected by boundaries provided by the embodiments of the present invention can realize the fusion of multiple information, thereby improving the exploration of unfamiliar areas and increasing the possibility of finding a global optimal solution.

[0111] In an embodiment of the present invention, for the model parameter coding after the exploration of the unfamiliar area, according to the optimal model parameter coding, an adaptive optimal guidance strategy is used to perform a rapid search in the optimal direction for the model parameter coding, and the model parameter coding after the rapid search is obtained, including:

[0112] Based on the current number of training times, the adaptive inertia weight factor is determined as: ;in, Indicates tThe adaptive inertia weight factor during the training process, represents the maximum value corresponding to the adaptive inertia weight factor, represents the minimum value corresponding to the adaptive inertia weight factor, cos represents the cosine function, represents pi, T represents the maximum number of training times;

[0113] According to the optimal model parameter encoding and the adaptive inertia weight factor, the adaptive update speed is determined as: ;in, Indicates t During the training process k The update speed of the model parameter encoding after exploring a strange area, Indicates t +1 training session k The update speed of the model parameter encoding after exploring a strange area, represents the first pilot coefficient, represents the second guidance coefficient, c represents the guidance range control factor, Indicates t During the training process k The model parameter encoding after exploring unfamiliar areas, Represents model parameter encoding The corresponding historical optimal value is dsit kp Represents model parameter encoding The best historical value The Euclidean distance between them, e represents a natural constant, represents the optimal model parameter encoding, dsit kbest Represents model parameter encoding Encoding with optimal model parameters The Euclidean distance between

[0114] According to the update speed , perform a quick search on the model parameter encoding in the optimal direction, and obtain the model parameter encoding after the quick search as: ;in, Indicates the model parameter encoding after fast search , represents the balance factor, and , sin represents the sine function.

[0115] The adaptive optimal guidance strategy provided by the embodiment of the present invention can realize rapid exploration towards the optimal position and can greatly improve the convergence accuracy of the algorithm. Combined with the previous jump exploration and unfamiliar area exploration, it can not only effectively avoid falling into the local optimum, but also realize high-precision exploration.

[0116] In an embodiment of the present invention, a target platelet-rich plasma concentration input by a medical staff is obtained, and the platelet-rich plasma separation system parameters are encoded multiple times to determine multiple system parameter codes, including:

[0117] Obtain target platelet-rich plasma concentration entered by healthcare provider;

[0118] The platelet-rich plasma separation system parameters are randomly generated within the upper and lower limits of the platelet-rich plasma separation system parameters, and the generated platelet-rich plasma separation system parameters are combined into a system parameter code, and a plurality of different system parameter codes are repeatedly obtained.

[0119] In the embodiment of the present invention, with the goal of minimizing the loss function value, an intelligent optimization algorithm is used to optimize multiple system parameter encodings to determine the global optimal system parameter encoding, including:

[0120] According to the loss function value, determine the system parameter encoding with the smallest loss function value as the optimal system parameter encoding;

[0121] Based on the optimal system parameter encoding, a particle swarm optimization algorithm is used to optimize the system parameter encoding;

[0122] The system parameter encoding is optimized repeatedly until the optimization end condition is met, and the global optimal system parameter encoding is determined.

[0123] For example, a maximum number of optimization times may be set, and when the number of optimizations for the system parameter encoding reaches the maximum number of optimizations, the global optimal system parameter encoding may be determined.

[0124] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A method for controlling the concentration of platelet-rich plasma by suspension centrifugation, characterized in that: include: Acquire historical platelet separation concentration data; wherein the historical platelet separation concentration data includes historical platelet-rich plasma separation system parameters and historical platelet-rich plasma concentrations acquired based on the historical platelet-rich plasma separation system parameters; A platelet separation concentration prediction model is constructed by using a deep learning algorithm, and the platelet separation concentration prediction model is trained using historical platelet-rich plasma separation system parameters and historical platelet-rich plasma concentrations obtained based on the historical platelet-rich plasma separation system parameters to obtain a trained platelet separation concentration prediction model; Obtaining a target platelet-rich plasma concentration input by a medical staff, and encoding the platelet-rich plasma separation system parameters multiple times to determine multiple system parameter codes; Using the system parameter code as a platelet separation concentration prediction model after training, determining the predicted platelet-rich plasma concentration corresponding to each system parameter code, and determining the loss function value corresponding to each system parameter code according to the target platelet-rich plasma concentration and the predicted platelet-rich plasma concentration; With the goal of minimizing the loss function value, an intelligent optimization algorithm is used to optimize multiple system parameter codes, determine the global optimal system parameter code, and use the global optimal system parameter code as the final platelet-rich plasma separation system parameter; Based on the final platelet-rich plasma separation system parameters, the platelet-rich plasma separation system is controlled, and the platelet-rich plasma separation system parameters are adjusted in real time to achieve platelet separation concentration control; The platelet-rich plasma separation system parameters include the current platelet-rich plasma concentration, separation speed, total separation time and separation time; The historical platelet-rich plasma separation system parameters include the platelet-rich plasma separation system parameters at the initial moment of separation and the platelet-rich plasma separation system parameters during the separation process; Constructing a platelet separation concentration prediction model through a deep learning algorithm, including: constructing a platelet separation concentration prediction model through a BP neural network algorithm; The platelet separation concentration prediction model is trained using historical platelet-rich plasma separation system parameters and historical platelet-rich plasma concentrations obtained based on the historical platelet-rich plasma separation system parameters to obtain a trained platelet separation concentration prediction model, including: Initializing model parameters of a platelet separation concentration prediction model, encoding the model parameters, determining the model parameter encoding, and obtaining a plurality of model parameter encodings; For any model parameter code, after applying the model parameters contained in the model parameter code to the platelet separation concentration prediction model, using the historical platelet-rich plasma separation system parameters as the input of the platelet separation concentration prediction model, using the platelet-rich plasma concentration as the expected output, and obtaining the error function value corresponding to the model parameter code; Obtain the optimal model parameter encoding according to the error function value corresponding to the model parameter encoding; For the model parameter coding, a random matching sinusoidal information interaction strategy is used to perform neighborhood exploration on the model parameter coding, and the model parameter coding after the neighborhood exploration is determined; For the model parameter coding after neighborhood exploration, according to the optimal model parameter coding, an information sharing strategy combined with random jumps is used to perform adaptive jump exploration on the model parameter coding to obtain the model parameter coding after adaptive jump exploration; For the model parameter encoding after adaptive jump exploration, the information interaction strategy affected by the boundary is used to explore the unfamiliar area of ​​the model parameter encoding, and the model parameter encoding after the unfamiliar area exploration is obtained; For the model parameter coding after exploring the unfamiliar area, according to the optimal model parameter coding, an adaptive optimal guidance strategy is used to quickly search the model parameter coding in the optimal direction to obtain the model parameter coding after the rapid search; Determine whether the current number of training times is equal to or greater than the maximum number of training times. If so, re-obtain the optimal model parameter coding based on the model parameter coding after the quick search, and use the optimal model parameter coding as the final parameter of the platelet separation concentration prediction model to obtain the platelet separation concentration prediction model after training. Otherwise, return to the step of obtaining the error function value corresponding to the model parameter coding.

2. The method for controlling the concentration of platelet-rich plasma by suspension centrifugation according to claim 1, characterized in that: For the model parameter encoding, a random matching sinusoidal information interaction strategy is used to perform neighborhood exploration on the model parameter encoding, and the model parameter encoding after the neighborhood exploration is determined, including: For any model parameter code, randomly match the model parameter code with another model parameter code to obtain a first target model parameter code; The model parameter code is obtained to perform information interaction with the corresponding first target model parameter code, and the first information interaction item is determined as: ;in, Indicates the tth training process i Model parameter encoding, Indicates i Model parameter encoding The corresponding first target model parameter encoding, i =1,2,...,L, L represents the total number of model parameter encodings; The first information interaction item is processed by using a sine function, and the first information interaction item after processing is obtained as follows: ;in, represents the first random number in (0,1), represents pi, and sin represents the sine function; The model parameter encoding is subjected to neighborhood exploration according to the first information interaction item, and the model parameter encoding after the neighborhood exploration is determined to be: ;in, Represents the model parameter encoding after neighborhood exploration .

3. The method for controlling the concentration of platelet-rich plasma by suspension centrifugation according to claim 2, characterized in that: For the model parameter coding after neighborhood exploration, according to the optimal model parameter coding, an information sharing strategy combined with random jumps is used to perform adaptive jump exploration on the model parameter coding, and the model parameter coding after adaptive jump exploration is obtained, including: Get the random jump coefficient as: ;in, represents the random jump coefficient, Indicates compliance N (0, ) a second random number from a normal distribution, Indicates compliance N (0,1) The third random number from normal distribution, represents the intermediate parameter, represents a constant term, and , represents the gamma function, and π represents the circumference of a circle; Based on the current training times, the random jump intensity is determined as: ;in, represents the random jump intensity, represents the fourth random number in (0,1), and T represents the preset maximum number of training times; For any model parameter encoding after neighborhood exploration, randomly match the model parameter encoding with another model parameter encoding to obtain a second target model parameter encoding; For the model parameter encoding after any neighborhood exploration, the random jump term is determined according to the second target model parameter encoding, the random jump coefficient and the random jump strength: ;in, Indicates n The model parameter encoding after neighborhood exploration, n =1,2,...,L, L represents the total number of model parameter encodings, Indicates n Model parameter encoding after neighborhood exploration The corresponding second target model parameter encoding; According to the random jump term and the optimal model parameter encoding, the model parameter encoding is adaptively jumped and explored, and the model parameter encoding after the adaptive jump exploration is obtained as follows: ;in, Represents the model parameter encoding after adaptive jump exploration , represents the fifth random number in (0,1), represents the sixth random number in (0,1), Represents the optimal model parameter encoding.

4. The method for controlling the concentration of platelet-rich plasma separated by suspension centrifugation according to claim 3, characterized in that: For the model parameter encoding after the adaptive jump exploration, the information interaction strategy affected by the boundary is used to explore the unfamiliar area for the model parameter encoding, and the model parameter encoding after the unfamiliar area exploration is obtained, including: Based on the current number of training times, the boundary influence factor is obtained as: ;in, represents the boundary impact factor, t represents the current number of training times, T represents the maximum number of training times, and L represents the total number of model parameter encodings; According to the boundary impact factor, the boundary impact code is obtained as: ;in, Indicates that the boundary affects the encoding, ub The upper limit encoding representing the upper limit composition of all dimensional model parameters, lb The lower bound encoding representing the lower bound composition of all dimensional model parameters; For any model parameter encoding after the adaptive jump exploration, randomly match another model parameter encoding for the model parameter encoding to obtain a third target model parameter encoding; According to the third target model parameter coding and the boundary influence coding, the model parameter coding is explored in the unfamiliar area, and the model parameter coding after the unfamiliar area exploration is obtained as follows: ;in, Indicates t The model parameter encoding after the mth adaptive jump exploration in the training process, represents the sixth random number in (0,1), Represents the model parameter encoding after exploring unfamiliar areas , represents the seventh random number in (0,1), Represents model parameter encoding The corresponding third target model parameter encoding, Represents the eighth random number in (0,1).

5. The method for controlling the concentration of platelet-rich plasma separated by suspension centrifugation according to claim 4, characterized in that: For the model parameter encoding after exploring the unfamiliar area, according to the optimal model parameter encoding, an adaptive optimal guidance strategy is used to quickly search the model parameter encoding in the optimal direction, and the model parameter encoding after the rapid search is obtained, including: Based on the current number of training times, the adaptive inertia weight factor is determined as: ;in, Indicates t The adaptive inertia weight factor during the training process, represents the maximum value corresponding to the adaptive inertia weight factor, represents the minimum value corresponding to the adaptive inertia weight factor, cos represents the cosine function, represents pi, T represents the maximum number of training times; According to the optimal model parameter encoding and the adaptive inertia weight factor, the adaptive update speed is determined as: ;in, Indicates t During the training process k The update speed of the model parameter encoding after exploring a strange area, Indicates t +1 training session k The update speed of the model parameter encoding after exploring a strange area, represents the first pilot coefficient, represents the second guidance coefficient, c represents the guidance range control factor, Indicates t During the training process k The model parameter encoding after exploring unfamiliar areas, Represents model parameter encoding The corresponding historical optimal value is dsit kp Represents model parameter encoding The best historical value The Euclidean distance between them, e represents a natural constant, represents the optimal model parameter encoding, dsit kbest Represents model parameter encoding Encoding with optimal model parameters The Euclidean distance between According to the update speed , perform a quick search on the model parameter encoding in the optimal direction, and obtain the model parameter encoding after the quick search as: ;in, Indicates the model parameter encoding after fast search , represents the balance factor, and , sin represents the sine function.

6. The method for controlling the concentration of platelet-rich plasma separated by suspension centrifugation according to any one of claims 1 to 5, characterized in that: Obtaining the target platelet-rich plasma concentration input by the medical staff, and encoding the platelet-rich plasma separation system parameters multiple times to determine multiple system parameter codes, including: Obtain target platelet-rich plasma concentration entered by healthcare provider; The platelet-rich plasma separation system parameters are randomly generated within the upper and lower limits of the platelet-rich plasma separation system parameters, and the generated platelet-rich plasma separation system parameters are combined into a system parameter code, and a plurality of different system parameter codes are repeatedly obtained.

7. The method for controlling the concentration of platelet-rich plasma separated by suspension centrifugation according to claim 6, characterized in that: With the goal of minimizing the loss function value, an intelligent optimization algorithm is used to optimize multiple system parameter encodings to determine the global optimal system parameter encoding, including: According to the loss function value, determine the system parameter encoding with the smallest loss function value as the optimal system parameter encoding; Based on the optimal system parameter encoding, a particle swarm optimization algorithm is used to optimize the system parameter encoding; The system parameter encoding is optimized repeatedly until the optimization end condition is met, and the global optimal system parameter encoding is determined.

Images