Urine control management system and method based on intelligent catheter

Abstract

The invention relates to the technical field of medical care informatics, in particular to a urine control management system and method based on an intelligent catheter, and the method comprises the steps: obtaining the data of a historical patient in a catheter insertion process, thereby obtaining a scalar value sequence and a weight vector sequence of nodes in a hidden layer in a neural network architecture, and screening out a target hidden layer; dividing each target hidden layer into a plurality of node groups, acquiring an updated node group, thereby acquiring a neural network after architecture updating, performing training by using the training data set, acquiring a trained neural network, outputting a pain point in the catheter insertion process of the current patient, and performing local anesthesia on the pain point. According to the invention, the pain point of the patient in the catheter insertion process can be accurately identified, and then local anesthesia is performed on the pain point, so that the urine control management effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical care informatics, and in particular to a urine control management system and method based on an intelligent urinary catheter. Background Art

[0002] In urological surgery, a urinary catheter is an indispensable medical device used for urination management, bladder decompression, and urine volume monitoring. However, existing non-intelligent urinary catheters have many problems in practical applications. For example, traditional urinary catheters are usually made of hard materials (such as rubber or PVC), which can easily cause mechanical stimulation to the urethral mucosa during insertion. Especially in male patients, due to the long and curved urethra, the process of inserting a urinary catheter is often accompanied by severe pain and discomfort, and it is impossible to effectively judge the patient's physical changes during the catheterization process.

[0003] Existing problems: During the urinary control management process of the intelligent catheter, the microelectrodes on the surface of the catheter are used to record the urethral electrical signals during the insertion of the catheter, and then the urethral electrical signals are analyzed through the neural network model to locate the local discomfort (pain) points in the urethra, so as to perform local anesthesia on the pain points. However, when the coordinates of the pain points in the urethra are identified through the neural network, if there is pain at a certain position in the urethra during the insertion of the catheter, it means that the urethra or surrounding tissues in this area are irritated or damaged, and these damaged areas will affect the adjacent areas, and then the electrical signals in the adjacent areas will also produce signal changes. Therefore, each local discomfort point may affect the electrical signals in the adjacent areas, causing the neural network to learn more repeated information, making it difficult to accurately locate the pain point area. Summary of the invention

[0004] The present invention provides a urine control management system and method based on an intelligent urinary catheter to solve the existing problems.

[0005] The urine control management system and method based on the intelligent urinary catheter of the present invention adopts the following technical solutions:

[0006] An embodiment of the present invention provides a urine control management method based on an intelligent urinary catheter, the method comprising the following steps:

[0007] Obtain a historical data set consisting of all pieces of data of all historical patients during the catheter insertion process, and divide the historical data set into a test data set and a training data set; input all pieces of data in the test data set into the neural network one by one, and output a scalar value sequence and a weight vector sequence of each node in each hidden layer in the neural network architecture;

[0008] According to the size of the modulus of the weight vector in the weight vector sequence of the nodes in each hidden layer, a number of target hidden layers are selected; in each target hidden layer, according to the similarity between the scalar value sequences of the nodes, all the nodes are divided into a number of node groups;

[0009] Delete the nodes in the node group to obtain an updated node group;

[0010] In the neural network architecture, each updated node group of each target hidden layer is taken as a node to obtain the neural network with updated architecture; the neural network with updated structure is trained using the training data set to obtain the trained neural network; the pain points of the current patient during the catheter insertion process are obtained using the trained neural network, and local anesthesia is performed on the pain points.

[0011] Furthermore, the screening out of a plurality of target hidden layers includes the following specific steps:

[0012] Obtain the modulus of each weight vector in the weight vector sequence of any node in any hidden layer to form a weight vector modulus sequence;

[0013] Filter out the target node according to the size of the weight vector modulus in the weight vector modulus sequence of each node in the hidden layer;

[0014] According to the difference of adjacent weight vector moduli in the weight vector modulus sequence of each target node, duplicate information nodes are screened out;

[0015] According to the number of repeated information nodes in each hidden layer, the target hidden layer is selected.

[0016] Furthermore, the target node is screened out according to the size of the weight vector modulus in the weight vector modulus sequence of each node in the hidden layer, and the specific steps include the following:

[0017] Select any node A in any hidden layer, and take the median in the weight vector modulus sequence of A as the benchmark weight value; obtain the number S of weight vector moduli in the weight vector modulus sequence of A that are less than the benchmark weight value, and when the ratio of the number S to the length of the weight vector modulus sequence of A is greater than a preset first proportion threshold, record A as the target node.

[0018] Furthermore, the method of screening out duplicate information nodes according to the difference between adjacent weight vector modules in the weight vector module sequence of each target node includes the following specific steps:

[0019] In the weight vector modulus sequence of any target node B, select any two adjacent weight vector moduli C and D, and take the ratio of the absolute value of the difference between C and D to the maximum value between C and D as the difference value of C and D. When the maximum value of the difference values ​​of all two adjacent weight vector moduli is less than the preset judgment threshold, B is recorded as a duplicate information node.

[0020] Furthermore, the method of selecting a target hidden layer according to the number of repeated information nodes in each hidden layer includes the following specific steps:

[0021] In any hidden layer, when the ratio of the number of repeated information nodes to the number of nodes in the any hidden layer is less than a preset second proportion threshold, the any hidden layer is recorded as a target hidden layer.

[0022] Furthermore, in each target hidden layer, all nodes are divided into a number of node groups according to the similarity between the scalar value sequences of the nodes, including the following specific steps:

[0023] In any target hidden layer, an inverse proportional value of the cosine similarity between the scalar value sequences of any two nodes is obtained as the clustering distance of the any two nodes, and a clustering operation is performed on all nodes to obtain a plurality of node groups.

[0024] Furthermore, the process of deleting nodes in the node group to obtain an updated node group includes the following specific steps:

[0025] Obtain the modulus of each weight vector in the weight vector sequence of any node in any hidden layer to form a weight vector modulus sequence;

[0026] In the weight vector modulus sequence of any node in any node group, the mean of all weight vector moduli is obtained, and the ordinal values ​​of all weight vector moduli greater than the mean are counted to form a weight ordinal value set;

[0027] Obtain the input sequence number of each data item in the test data set and input it into the neural network one by one, and count the input sequence number set of all data items in the test data set for each historical patient;

[0028] According to the weight sequence value set and the input sequence number set of each historical patient, the initial reserved nodes are selected from any node group, and the historical patients corresponding to each initial reserved node are obtained;

[0029] According to the historical patients corresponding to the initial reserved nodes, the final reserved nodes are screened out from the initial reserved nodes;

[0030] All the final retained nodes in any node group form an update node group.

[0031] Furthermore, the method of selecting the initial reserved nodes from any node group according to the weight sequence value set and the input sequence number set of each historical patient, and obtaining the historical patients corresponding to each initial reserved node, includes the following specific steps:

[0032] If the weight sequence value set of any node in any node group is a subset of the input sequence number set of any historical patient, the arbitrary node is recorded as the initial reserved node, and the arbitrary historical patient is used as the historical patient corresponding to the initial reserved node.

[0033] Furthermore, the method of selecting the final reserved nodes from the initial reserved nodes according to the historical patients corresponding to the initial reserved nodes includes the following specific steps:

[0034] In any node group, the historical patients corresponding to all the initial reserved nodes constitute a historical patient set, and the initial reserved node corresponding to the historical patient with the largest number of the same historical patients in the historical patient set is recorded as the final reserved node.

[0035] The present invention also proposes a urine control management system based on a smart urinary catheter, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program stored in the memory to implement the steps of the aforementioned urine control management method based on a smart urinary catheter.

[0036] The beneficial effects of the technical solution of the present invention are:

[0037] In an embodiment of the present invention, a historical data set consisting of historical patient data during the catheter insertion process is obtained to output the scalar value sequence and weight vector sequence of each node in each hidden layer in the neural network architecture, thereby screening out several target hidden layers, dividing each target hidden layer into several node groups, and deleting the nodes in the node group to obtain an updated node group, with each updated node group as a node, thereby reducing the impact of electrical signals in adjacent areas of the pain point in the urethra, thereby preventing the neural network from learning more repeated information, so that the pain point area can be accurately located, and the neural network after the updated architecture is obtained, and the training data set is used for training to obtain the trained neural network. It is used to output the pain point of the current patient during the catheter insertion process and perform local anesthesia on the pain point. So far, the present invention can accurately identify the pain point of the patient during the catheter insertion process, and then perform local anesthesia on the pain point to improve the urine control management effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0039] Figure 1 This is a flowchart of the steps of a urine control management method based on an intelligent urinary catheter of the present invention;

[0040] Figure 2 Schematic diagram of urinary catheter insertion;

[0041] Figure 3 This is a schematic diagram of the neural network architecture;

[0042] Figure 4 Schematic diagram of the neural network after updating the architecture. DETAILED DESCRIPTION

[0043] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation, structure, features and effects of a urine control management system and method based on a smart urinary catheter proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0044] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0045] The specific scheme of a urine control management system and method based on an intelligent urinary catheter provided by the present invention is described in detail below with reference to the accompanying drawings.

[0046] See also Figure 1 , which shows a flowchart of a urine control management method based on an intelligent urinary catheter provided by an embodiment of the present invention, the method comprising the following steps:

[0047] Step S001: Obtain a historical data set consisting of all data items of all historical patients during the catheter insertion process, and divide the historical data set into a test data set and a training data set; input all data items in the test data set into the neural network one by one, and output a scalar value sequence and a weight vector sequence for each node in each hidden layer in the neural network architecture.

[0048] It should be noted that: in this embodiment, the catheter adopts a double-layer design. The inner layer is used for urine diversion. It is made of biocompatible materials to ensure that urine can flow out of the body smoothly without causing damage to the urethra or bladder, and the inner layer material has high flexibility and pressure resistance to adapt to the physiological curve of the urethra. The interlayer is configured with anesthetic ingredients such as Licardoine. Licardoine is slowly released through the penetration port. The function of the interlayer is to provide anesthesia to the local area in the urethra, especially to relieve discomfort or pain during the insertion of the catheter. The surface layer is the part of the catheter with microelectrodes. The microelectrodes contact the inner wall of the urethra through the surface to collect electrophysiological signals of the urethra. Schematic diagram of catheter insertion, as shown Figure 2 Thus, the urethral electrical signals during the insertion of the catheter are recorded by using the surface microelectrodes, and then the urethral electrical signals are analyzed by the neural network model to locate the local discomfort (pain) points in the urethra, and finally the local anesthesia in the urethra is performed by local drug delivery using the surface penetration port.

[0049] For several historical patients, the voltage and three-dimensional coordinates of each microelectrode on the surface of the catheter of each historical patient at each moment during the catheter insertion process are collected.

[0050] It should be noted that in this embodiment, the data acquisition frequency is once per second. The microelectrode configured on the surface of the catheter uses nanometer-level electrode materials, which are highly sensitive and can detect weak electrophysiological signals in the urethra. The electrode contacts the urethral tissue through the tiny holes on the surface to collect electrical signals in the urethra. During the insertion of the catheter, the microelectrode collects electrical signals from the urethra in real time. These signals mainly come from the physiological activities of the urethral wall, such as urethral muscle contraction, potential changes, etc. During the insertion process, due to the stress response of the urethral tissue, local potential changes or pain signals may be generated. These signals can be captured by the microelectrode sensor. The three-dimensional coordinates of the microelectrode are then measured by a coordinate measuring machine in a contact or non-contact manner. This is a high-precision measurement device.

[0051] The neural network in this implementation adopts the CNN (Convolutional Neural Network) deep network model and selects the ResNet (Residual Network) architecture, which consists of an input layer, an output layer, and several hidden layers.

[0052] The input of the neural network is: for each microelectrode on the surface of the catheter of each historical patient during the catheter insertion process, three voltages within every three seconds and the three-dimensional coordinates of the last second constitute a piece of data (such as: the first piece of data includes the voltages of the 1st second, the 2nd second and the 3rd second and the three-dimensional coordinates of the 3rd second, the second piece of data includes the voltages of the 4th second, the 5th second and the 6th second and the three-dimensional coordinates of the 6th second. If the last time does not meet 3 seconds, it is not considered as a piece of data). This is used as an example for description. In other implementations, it can be set as the acquisition method of each other piece of data, which is not limited in this embodiment. All pieces of data of all historical patients constitute a historical data set as the input of the neural network.

[0053] It should be noted that this ensures that the input length of the network at different times is consistent, avoiding the length of the electrical signal sequence becoming longer and longer over time, resulting in different lengths of the electrical signal sequences at different times, and thus being unable to be input into the same neural network.

[0054] Among them, each piece of data contains 6 features, namely the voltage for 3 consecutive seconds and the coordinate values ​​of three dimensions, so the number of input nodes of the neural network is set to 6.

[0055] The output of the neural network is: the three-dimensional coordinates of the pain point.

[0056] Schematic diagram of neural network architecture, such as Figure 3 As shown, Figure 3 Each circle in the figure represents a node in the neural network. From left to right, the first column of nodes is the input layer, the last column of nodes is the output layer, and the other columns of nodes are each hidden layer.

[0057] In this embodiment, in the historical data set, based on the moment when each historical patient reported pain, each piece of data at the moment of pain for each historical patient is obtained as the electrical signal data of pain generated by the historical patient when the catheter was inserted. Doctors or professional medical personnel annotate the data based on the patient's symptoms, physiological signals and clinical diagnosis, and mark which data corresponds to the pain signal to guide the neural network model to learn how to distinguish between pain and non-pain signals.

[0058] It should be noted that: due to the diffusion of nerve conduction, pain signals will cause corresponding signal changes in adjacent areas, that is, the pain signal when inserting a catheter will diffuse between different areas through nerve conduction, especially for more extensive or deep pain stimulation, which will cause the neural network to contain more repetitive information, making it difficult for the neural network to accurately distinguish pain in a specific location. In this regard, in this embodiment, the hidden layer nodes containing more repetitive information are grouped together through the method of group convolution, and then the hidden nodes containing more repetitive information are given a smaller proportion of allowable error in each hidden layer.

[0059] In the historical data set, any The data set is composed of the experimental data, among which, is the number of all data in the historical data set, In order to round down, all data outside the test data set constitute the training data set, and this is used as an example for description. Other implementations may be set to other allocation methods, which are not limited in this embodiment. Thus, the neuron nodes in some hidden layers of the neural network are grouped by the data in the training process of the test data set and the data obtained after the training.

[0060] During the neural network training process for the test data set, each time the first data is input into the network, each node in each hidden layer will get a scalar value, and the weight of each edge in the neural network is obtained by feedback of the loss value calculated by the loss function in the neural network. The weights of all edges corresponding to each node in each hidden layer constitute a weight vector. Then the second data is input into the network, and each node in the hidden layer gets an updated scalar value. Then the updated weight of each edge in the neural network is obtained by feedback of the loss value calculated by the loss function in the neural network. The update is iterated in sequence until all the data in the test data set are input. Among them, the loss value calculated by the loss function in the neural network measures the difference between the model prediction output and the true label. When there is a difference, the edge weight in the network is updated by the difference.

[0061] After all the data in the test data set are input into the neural network one by one, the scalar value sequence and weight vector sequence of each node in each hidden layer in the neural network architecture will be obtained.

[0062] It should be noted that: each scalar value in the scalar value sequence corresponds to a piece of data in the test data set, each weight vector in the weight vector sequence corresponds to a piece of data in the test data set, the scalar value sequence represents the change process of the value of each node in the hidden layer during the neural network training process, and the weight vector sequence represents the process of iterative update of the weight of each node in the hidden layer during the neural network training process. The above explanations of the neural network are all well-known technologies.

[0063] Step S002: According to the size of the modulus of the weight vector in the weight vector sequence of the nodes in each hidden layer, a number of target hidden layers are screened out; in each target hidden layer, all nodes are divided into a number of node groups according to the similarity between the scalar value sequences of the nodes.

[0064] It should be noted that: if the scalar value sequences of several nodes in a hidden layer are similar, it means that the changes of these nodes when pain occurs in different patients are similar. These nodes may represent more repeated information that is transmitted from the urethral nerve where the pain point is located to the adjacent nerve, causing pain stimulation at this location. For all hidden layers, not every hidden layer can concentrate repeated information into several nodes. It is possible that each node in a hidden layer contains some repeated information. In this case, it is difficult to process the repeated information by grouping. Therefore, it is necessary to first find a hidden layer where the repeated information is concentrated into several nodes. If a node in a hidden layer contains more repeated information, then since the node cannot learn more useful information, the change in the weight value of the node during the iterative update of the weight is relatively small compared to other nodes.

[0065] Preferably, in one embodiment of the present invention, the method for obtaining the node group in the target hidden layer includes:

[0066] The first ratio threshold is preset to 0.7, the second ratio threshold is preset to 0.3, and the judgment threshold is preset to 0.3, which is used as an example for description.

[0067] Taking any node A in any hidden layer in the neural network architecture as an example, obtain the modulus of each weight vector in the weight vector sequence of A to form a weight vector modulus sequence, use the median in the weight vector modulus sequence as the benchmark weight value, obtain the number S of weight vector moduli that are less than the benchmark weight value in the weight vector modulus sequence, and then obtain the ratio of the number S to the length of the weight vector modulus sequence. When the ratio is greater than the preset first proportion threshold, A is recorded as the target node.

[0068] It should be noted that the modulus of a vector and the acquisition of the median in a sequence are both well-known technologies, and the median of a data sequence refers to the value in the middle position after a set of data is arranged in order of size. Thus, several target nodes in the hidden layer are obtained.

[0069] For any target node B, in the weight vector modulus sequence of B, select any two adjacent weight vector moduli C and D, and take the ratio of the absolute value of the difference between C and D to the maximum value between C and D as the difference value of C and D. When the maximum value of the difference values ​​of all two adjacent weight vector moduli is less than the preset judgment threshold, B is recorded as a duplicate information node.

[0070] In any hidden layer, when the ratio of the number of repeated information nodes to the number of nodes in the any hidden layer is less than a preset second proportion threshold, the any hidden layer is recorded as a target hidden layer.

[0071] In any target hidden layer, the inverse proportional value of the cosine similarity between the scalar value sequences of any two nodes is obtained as the clustering distance of the any two nodes, and a density clustering algorithm is used to cluster all nodes to obtain several clusters, with each cluster as each node group.

[0072] It should be noted that the density clustering algorithm and cosine similarity are both well-known technologies. The value range of cosine similarity is between -1 and 1. The larger the value, the more similar the two data sequences are. Therefore, in this embodiment, the difference between the cosine similarity of the scalar value sequences of any two nodes is taken as the inverse proportional value of the cosine similarity of the scalar value sequences of any two nodes. It can be seen that the scalar value sequences of the nodes in each cluster are similar.

[0073] Step S003: Delete the nodes in the node group to obtain an updated node group.

[0074] It should be noted that: if the nodes belonging to the same category in each hidden layer are directly divided into a group, and then group convolution is performed, since the pain data of different times also have similar information, it is necessary to remove this repeated information, and then obtain similar information that can represent the influence of the electrical signals at different parts of the urethra caused by the spread of pain, that is, repeated information. In view of this situation, in this embodiment, by obtaining the corresponding weight vector sequence for each node in each node group in the hidden layer that needs to be grouped, and then through the larger values ​​in the weight vector sequence of those nodes, it is obtained whether the input that contributes more to these larger values ​​belongs to the same patient. If they belong to the same patient, the obtained node group does not contain other pain information (similar information existing in pain data of different times).

[0075] Preferably, in one embodiment of the present invention, the method for acquiring the updated node group includes:

[0076] For any node F in any node group of any target hidden layer, in the weight vector modulus sequence of F, obtain the mean of all weight vector moduli, and count the ordinal values ​​of all weight vector moduli greater than the mean to form a weight ordinal value set.

[0077] It should be noted that the set of weight ordinal values ​​corresponds to the larger weight of F in the weight iteration process, which can better represent the weight change of the node. Each weight ordinal value corresponds to a piece of data in the test data set, that is, it corresponds to a historical patient. If there is a node in the node group to which the node belongs that is the same as the corresponding patient of the node, then the repeated information carried by these identical nodes does not contain other pain information (similar information existing in pain data of different times). This is what this embodiment wants to represent, which can represent the similar information caused by the spread of pain that affects the electrical signals at different parts of the urethra, that is, repeated information.

[0078] The input sequence number of each data in the test data set is obtained and input into the neural network one by one, and the set of input sequence numbers of all data in the test data set of each historical patient is counted as the input sequence number set of each historical patient.

[0079] It should be noted that: in this embodiment, all data in the test data set are input into the neural network one by one, and the input sequence number of each data is {1, 2, 3, ...} in sequence.

[0080] If the weight sequence value set of F is a subset of the input sequence number set of any historical patient, then F is recorded as the initial reserved node, and the arbitrary historical patient is used as the historical patient corresponding to the initial reserved node F.

[0081] It should be noted that all the input data corresponding to the larger weight of each initial retained node are from the same historical patient.

[0082] In any node group, the historical patients corresponding to all the initial reserved nodes constitute a historical patient set, and the initial reserved node corresponding to the historical patient with the largest number of the same historical patients in the historical patient set is recorded as the final reserved node.

[0083] In any node group, all final reserved nodes constitute an update node group.

[0084] This completes the node deletion operation in the node group.

[0085] It should be noted that: if the historical patient set is {historical patient 1, historical patient 1, historical patient 2, historical patient 3}, the three initial reserved nodes corresponding to historical patient 1 are used as the final reserved nodes. If there are multiple historical patients with the largest number of the same historical patients in the historical patient set, any one is taken as an example for mechanical energy analysis. At this time, the larger weights corresponding to all nodes in each updated node group during the neural network iteration process belong to the data of the same historical patient, thereby avoiding the node group from containing other pain information, such as: similar information in the pain data of different patients.

[0086] Step S004: In the neural network architecture, each updated node group of each target hidden layer is taken as a node to obtain the neural network after the updated architecture; the training data set is used to train the neural network after the updated structure to obtain the trained neural network; the trained neural network is used to obtain the pain points of the current patient during the catheter insertion process, and local anesthesia is performed on the pain points.

[0087] Preferably, in one embodiment of the present invention, the method for obtaining pain points includes:

[0088] In the neural network architecture, each updated node group of each target hidden layer is regarded as a node to obtain the neural network after the updated architecture.

[0089] It should be noted that in neural networks, the practice of treating multiple nodes in the same hidden layer as one node is usually called "node aggregation" or "feature fusion". This is a well-known technology and the specific method will not be introduced here. Figure 4 shown. Figure 4 The third column from left to right is the target hidden layer, where the three selected nodes are an update node group. All nodes in each update node group share a weight vector sequence, that is, the final update architecture is obtained. The network structure at this time is conducive to removing similar information caused by different pain data, and then obtaining similar information that can represent the influence of electrical signals at different parts of the urethra caused by the spread of pain, that is, repeated information. This repeated information is processed in a unified manner to avoid the influence of redundant information caused by repeated information on the network recognition function.

[0090] The training data set is input into the neural network after the architecture is updated. The loss function adopts the mean square error loss function. Through iterative training, the trained neural network is obtained. This is a well-known operation.

[0091] According to the above method of acquiring each piece of data of the historical patient, each piece of data of the current patient during the catheter insertion process is acquired, all pieces of data of the current patient during the catheter insertion process are input into the trained neural network, and the pain point of the current patient during the catheter insertion process is output. Local anesthesia is then performed on the pain point.

[0092] It should be noted that after the location of the pain point is determined, the infiltration port designed in the catheter will release Licardoine or other anesthetics as needed. The drug release process is carried out through the micropores or infiltration ports on the surface, and Licardoine will directly penetrate into the local area where the pain point is located, thereby achieving the effect of local anesthesia. This process can not only relieve pain in the urethra, but also reduce the physiological discomfort of the patient during the insertion of the catheter.

[0093] The present invention also provides a urine control management system based on an intelligent urinary catheter, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program stored in the memory to implement the steps of the aforementioned urine control management method based on an intelligent urinary catheter.

[0094] So far, the present invention is completed.

[0095] In summary, in an embodiment of the present invention, a historical data set consisting of all data of all historical patients during the catheter insertion process is obtained, and the historical data set is divided into a test data set and a training data set; all data in the test data set are input into the neural network one by one, and the scalar value sequence and weight vector sequence of each node in each hidden layer in the neural network architecture are output; according to the size of the modulus of the weight vector in the weight vector sequence of the node in each hidden layer, several target hidden layers are screened out; in each target hidden layer, according to the similarity between the scalar value sequences of the nodes, all nodes are divided into several node groups; the nodes in the node group are deleted to obtain an updated node group; in the neural network architecture, each updated node group of each target hidden layer is used as a node to obtain a neural network after the updated architecture; the training data set is used to train the neural network after the updated structure to obtain the trained neural network; the trained neural network is used to obtain the pain points during the current patient's catheter insertion process, and the pain points are locally anesthetized. The present invention can accurately identify the pain points during the patient's catheter insertion process, and then locally anesthetize the pain points to improve the urine control management effect.

[0096] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A urine control management method based on an intelligent urinary catheter, characterized in that: The method comprises the following steps: Obtain a historical data set consisting of all pieces of data of all historical patients during the catheter insertion process, and divide the historical data set into a test data set and a training data set; input all pieces of data in the test data set into the neural network one by one, and output a scalar value sequence and a weight vector sequence of each node in each hidden layer in the neural network architecture; According to the size of the modulus of the weight vector in the weight vector sequence of the nodes in each hidden layer, a number of target hidden layers are selected; in each target hidden layer, according to the similarity between the scalar value sequences of the nodes, all the nodes are divided into a number of node groups; Delete the nodes in the node group to obtain an updated node group; In the neural network architecture, each updated node group of each target hidden layer is taken as a node to obtain the neural network with updated architecture; the neural network with updated structure is trained using the training data set to obtain the trained neural network; the pain points of the current patient during the catheter insertion process are obtained using the trained neural network, and local anesthesia is performed on the pain points.

2. A urine control management method based on an intelligent urinary catheter according to claim 1, characterized in that: The specific steps of screening out a number of target hidden layers are as follows: Obtain the modulus of each weight vector in the weight vector sequence of any node in any hidden layer to form a weight vector modulus sequence; Filter out the target node according to the size of the weight vector modulus in the weight vector modulus sequence of each node in the hidden layer; According to the difference of adjacent weight vector moduli in the weight vector modulus sequence of each target node, duplicate information nodes are screened out; According to the number of repeated information nodes in each hidden layer, the target hidden layer is selected.

3. A urine control management method based on an intelligent urinary catheter according to claim 2, characterized in that: The specific steps of screening out the target node according to the size of the weight vector modulus in the weight vector modulus sequence of each node in the hidden layer include the following: Select any node A in any hidden layer, and take the median in the weight vector modulus sequence of A as the benchmark weight value; obtain the number S of weight vector moduli in the weight vector modulus sequence of A that are less than the benchmark weight value, and when the ratio of the number S to the length of the weight vector modulus sequence of A is greater than a preset first proportion threshold, record A as the target node.

4. The urine control management method based on the intelligent urinary catheter according to claim 2, characterized in that: The specific steps of screening out duplicate information nodes according to the difference of adjacent weight vector modules in the weight vector module sequence of each target node include the following: In the weight vector modulus sequence of any target node B, select any two adjacent weight vector moduli C and D, and take the ratio of the absolute value of the difference between C and D to the maximum value between C and D as the difference value of C and D. When the maximum value of the difference values ​​of all two adjacent weight vector moduli is less than the preset judgment threshold, B is recorded as a duplicate information node.

5. The urine control management method based on the intelligent urinary catheter according to claim 2, characterized in that: The method of selecting a target hidden layer according to the number of repeated information nodes in each hidden layer includes the following specific steps: In any hidden layer, when the ratio of the number of repeated information nodes to the number of nodes in the any hidden layer is less than a preset second proportion threshold, the any hidden layer is recorded as a target hidden layer.

6. The urine control management method based on the intelligent urinary catheter according to claim 1, characterized in that: In each target hidden layer, all nodes are divided into a number of node groups according to the similarity between the scalar value sequences of the nodes, and the specific steps included are as follows: In any target hidden layer, an inverse proportional value of the cosine similarity between the scalar value sequences of any two nodes is obtained as the clustering distance of the any two nodes, and a clustering operation is performed on all nodes to obtain a plurality of node groups.

7. The urine control management method based on the intelligent urinary catheter according to claim 1, characterized in that: The specific steps of deleting the nodes in the node group to obtain the updated node group are as follows: Obtain the modulus of each weight vector in the weight vector sequence of any node in any hidden layer to form a weight vector modulus sequence; In the weight vector modulus sequence of any node in any node group, the mean of all weight vector moduli is obtained, and the ordinal values ​​of all weight vector moduli greater than the mean are counted to form a weight ordinal value set; Obtain the input sequence number of each data item in the test data set and input it into the neural network one by one, and count the input sequence number set of all data items in the test data set for each historical patient; According to the weight sequence value set and the input sequence number set of each historical patient, the initial reserved nodes are selected from any node group, and the historical patients corresponding to each initial reserved node are obtained; According to the historical patients corresponding to the initial reserved nodes, the final reserved nodes are screened out from the initial reserved nodes; All the final retained nodes in any node group form an update node group.

8. A urine control management method based on an intelligent urinary catheter according to claim 7, characterized in that: The specific steps of selecting the initial reserved nodes from any node group according to the weight sequence value set and the input sequence number set of each historical patient and obtaining the historical patients corresponding to each initial reserved node are as follows: If the weight sequence value set of any node in any node group is a subset of the input sequence number set of any historical patient, the arbitrary node is recorded as the initial reserved node, and the arbitrary historical patient is used as the historical patient corresponding to the initial reserved node.

9. The urine control management method based on the intelligent urinary catheter according to claim 7, characterized in that: The method of selecting the final reserved nodes from the initial reserved nodes according to the historical patients corresponding to the initial reserved nodes includes the following specific steps: In any node group, the historical patients corresponding to all the initial reserved nodes constitute a historical patient set, and the initial reserved node corresponding to the historical patient with the largest number of the same historical patients in the historical patient set is recorded as the final reserved node.

10. A urine control management system based on an intelligent urinary catheter, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is executed by a processor, the steps of a urine control management method based on an intelligent urinary catheter are implemented as described in any one of claims 1 to 9.

Images