Urinary incontinence risk assessment prediction model construction method based on neural network

Abstract

The invention belongs to the technical field of risk assessment and prediction, and discloses a method for constructing a urinary incontinence risk assessment and prediction model based on a neural network, and the method comprises the following steps: collecting data information about a urinary incontinence patient; performing data preprocessing on the collected data information, and establishing a risk assessment data set; based on the risk assessment data set, constructing a urinary incontinence risk assessment prediction model; carrying out optimization training on the constructed urinary incontinence risk assessment prediction model; and based on the optimized urinary incontinence risk assessment model, performing risk assessment prediction on the test set. According to the construction method of the urinary incontinence risk assessment prediction model based on the neural network, complex data can be better processed, the accuracy and reliability of the urinary incontinence risk assessment model are improved, and the life quality and convenience of a user are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of risk assessment and prediction, and in particular to a method for constructing a urinary incontinence risk assessment and prediction model based on a neural network. Background Art

[0002] Machine learning is a part of AI that is increasingly used for diagnosis, complications, prognosis, and recurrence prediction. Compared with traditional statistical models, machine learning can actively learn complex relationships between data, overcome nonlinear limitations, maintain the stability of high-dimensional data sets, and has unique advantages in analyzing unstructured data (such as images and other types of data). It is very suitable for assisting clinical prediction of treatment outcomes or disease progression, but this potential remains largely untapped.

[0003] The pathogenesis of female urinary incontinence involves multiple factors, such as age, weight, reproductive history, pelvic floor muscle function, etc., and a large amount of data is needed to establish an accurate prediction model. At the same time, the quality of the data will also affect the accuracy and reliability of the prediction model. Therefore, it is particularly necessary to seek a more powerful algorithm model. At present, a variety of prediction models for female urinary incontinence have been established, but these models have poor discrimination and calibration in the Chinese population.

[0004] Therefore, in view of the above technical defects, the present invention proposes a method for constructing a urinary incontinence risk assessment prediction model based on a neural network. Summary of the invention

[0005] The purpose of the present invention is to provide a method for constructing a urinary incontinence risk assessment prediction model based on a neural network, so as to solve the technical problems in the related art that the female urinary incontinence incidence prediction model has difficulty in processing unbalanced data, poor generalization ability, inability to effectively process high-dimensional sparse features of medical data, poor interpretability and other factors leading to poor prediction effect.

[0006] To achieve the above object, the present invention provides a method for constructing a urinary incontinence risk assessment prediction model based on a neural network, comprising the following steps:

[0007] Step S1, collecting data information about patients with urinary incontinence;

[0008] Step S2, preprocessing the data information collected in step S1, and establishing a risk assessment data set;

[0009] Step S3: constructing a urinary incontinence risk assessment prediction model based on the risk assessment data set;

[0010] Step S4, optimizing and training the constructed urinary incontinence risk assessment prediction model;

[0011] Step S5: Based on the optimized urinary incontinence risk assessment model, risk assessment prediction is performed on the test set.

[0012] Preferably, in step S1, the data information of the urinary incontinence patients is derived from the hospital's electronic medical record system, patient questionnaires, and clinical research;

[0013] The data information of patients with urinary incontinence includes: basic information of patients, relevant environmental risk factors, gynecological diseases and other medical history.

[0014] Preferably, in step S2, data preprocessing includes: data cleaning, data conversion and data normalization;

[0015] Among them, data cleaning includes filling missing values, correcting erroneous data, and deleting duplicate records; data conversion is to convert non-numeric data into numerical form using unique hot encoding; data normalization is to standardize or normalize numerical data to eliminate the impact of dimension;

[0016] Based on the preprocessed data information, a risk assessment dataset is constructed.

[0017] Preferably, in step S3, a urinary incontinence risk assessment prediction model is constructed based on the risk assessment data set, and the specific process is as follows:

[0018] Step S31, dividing the risk assessment data set into a training set and a test set, and setting the ratio of the training set to the test set to 8:2, to construct a urinary incontinence risk assessment prediction model;

[0019] Among them, the urinary incontinence risk assessment prediction model is a neural network model, which specifically includes an input layer, a hidden layer and an output layer; the input layer is responsible for receiving data in the risk assessment data set, the hidden layer performs data processing and feature extraction, and the output layer outputs an assessment report;

[0020] Step S32: Update the neural network model parameters.

[0021] Preferably, in step S32, the neural network model parameters are updated, and the specific process is as follows:

[0022] Step S321, calculate the first-order moment and the second-order moment, as shown below:

[0023]

[0024] m t =β1m t-1 +(1-β1)g t ;

[0025]

[0026] Among them, g tis the gradient of the loss function to the neural network parameters at the tth update; is the gradient on the parameter θ; J is the loss function; θ t-1 is the neural network parameter at the t-1th update; m t and m t-1 are the first-order moments at the t-th and t-1-th updates respectively; v t and v t-1 are the second-order moments at the t-th and t-1-th updates, respectively; β1 and β2 are the first-order moment decay rates and the second-order moment decay rates; Yes t The square of , t is the update index of the neural network parameters;

[0027] Step S322, offset correction is as follows:

[0028]

[0029] in, and are the first-order moment and second-order moment after bias correction at the tth update; and are the t-th power of β1 and β2 respectively;

[0030] Step S323, gradient pruning, designing an adaptive scaling factor, and pruning the gradient of the neural network parameters based on the adaptive scaling factor;

[0031] The total gradient norm is calculated as follows:

[0032]

[0033] Among them, sf t is the total gradient norm at the tth update; d is the parameter index, D is the total number of neural network parameters, is the gradient of the dth parameter at the jth update, is the square of the L2 norm;

[0034] Calculate the historical gradient benchmark, which includes the historical gradient mean and historical gradient variance, as shown below:

[0035]

[0036] Among them, lz t and lc t are the historical gradient mean and historical gradient variance at the tth update; g j is the gradient of the loss function with respect to the neural network parameters at the jth update;

[0037] Design the adaptive scaling factor as follows:

[0038]

[0039] Among them, zs t is the adaptive scaling factor at the tth update; gn max is the maximum gradient norm; γ is the control factor;

[0040] Apply trimming as follows:

[0041]

[0042] in, is the gradient of the dth parameter after pruning at the jth update;

[0043] Step S325: Design an adaptive learning rate, dynamically adjust the learning rate according to time and set parameters, and the learning rate changes periodically during the training process, as shown below:

[0044]

[0045] Among them, xl t is the adaptive learning rate at the tth update; xl0, xl fd and xl zq They are the initial value of the learning rate, the cycle length of the learning rate change, and the amplitude of the learning rate change;

[0046] Step S325, neural network model parameter update, neural network parameters are updated based on the adaptive learning rate and interpolation factor, as shown below:

[0047]

[0048] Among them, θ t is the neural network parameter at the tth update; σ is the interpolation factor, and ∈ is the smoothing term.

[0049] Preferably, in step S4, the constructed urinary incontinence risk assessment prediction model is optimized and trained, and the specific process is as follows:

[0050] Initialize the individual position, and adjust the first-order moment decay rate β1, second-order moment decay rate β2, and maximum gradient norm gn in the urinary incontinence risk assessment model max , control factor γ, initial value of learning rate xl0, cycle length of learning rate change xl fd , the magnitude of the learning rate change xl zq A search space is established with the interpolation factor σ, A individual positions are randomly initialized in the search space, and the individual positions are used as representatives of the parameters of the urinary incontinence risk assessment model. The loss function of the urinary incontinence risk assessment model established based on the model parameters is used as the individual fitness value.

[0051] Preferably, the design reference position includes the following contents:

[0052] Step S41: Calculate the local search ratio value as follows:

[0053]

[0054] Where Δh(u) is the local search ratio value at the uth iteration; x best (u) is the position of the global optimal individual at the uth iteration; is the average position of all individuals at the uth iteration; u is the current iteration number; i is the individual index; rand(·) is the random number generation function; x i (u) is the position of the i-th individual at the u-th iteration;

[0055] Step S42: Calculate the local position correction value as follows:

[0056]

[0057] Among them, w i (u) is the local position correction value of the ith individual at the uth iteration; x worst (u) is the position of the global worst individual at the u-th iteration;

[0058] Step S43: Calculate the query value, and generate two different query values ​​for each individual based on the local position correction value and the local search ratio value, as shown below:

[0059]

[0060] in, and are the first query value and the second query value of the i-th individual at the u-th iteration respectively;

[0061] Step S44: Calculate two reference positions, generate two different reference positions for each individual based on the query value, and guide the individual to move in a more promising direction, as shown below:

[0062]

[0063] in, and are the first reference individual and the second reference position of the i-th individual at the u-th iteration; x r1 (u) is the position of the first random individual at the u-th iteration;

[0064] Step S45, calculate the disturbance position, and generate a disturbance position for each individual based on the Levy flight strategy and the random individual, as shown below:

[0065]

[0066] Among them, i (u) is the perturbed position of the ith individual at the uth iteration; a is a random number generated based on the Levy flight strategy; b and p are taken from the normal distribution; b~N(0,λ 2 ), p~N(0,λ 2 ), λ is the shape parameter, Γ(·) is the gamma function, ε is the standard deviation of the normal distribution, x r2 (u) and x r3 (u) are the positions of the second and third random individuals at the u-th iteration; is the ceiling function;

[0067] Step S46: Individual position update: all individual positions are updated based on the reference position and the disturbance position, as shown below:

[0068]

[0069] Among them, x i (u+1) is the position of the i-th individual at the u+1-th iteration; s1 and s2 are the first and second random values ​​in the range (0,1), respectively;

[0070] Step S47, optimal parameters are determined, a fitness threshold is pre-set, the individual fitness value and the global optimal individual are updated, when the fitness value corresponding to the global optimal individual is lower than the fitness threshold, the model parameters corresponding to the global optimal individual position are the optimal parameters, and a urinary incontinence risk assessment model is constructed based on the optimal parameters; otherwise, if the maximum number of iterations is reached, the individual position is reinitialized; otherwise, the number of iterations is increased by 1 and the reference position and disturbance position are recalculated, and the individual position is updated.

[0071] Preferably, in step S5, based on the optimized urinary incontinence risk assessment model, risk assessment prediction is performed on the test set, and the specific process is as follows:

[0072] First, the test set data is input into the urinary incontinence risk assessment model; second, the risk assessment threshold is set in advance. When the risk assessment level output by the model is higher than the risk assessment threshold, the smart chip sends a warning message; finally, the risk assessment report is output for subsequent diagnostic assistance.

[0073] A urinary incontinence risk assessment prediction model based on neural network: comprising a data acquisition module, a data preprocessing module, a model building module, a model optimization module and a model prediction module;

[0074] Among them, the data acquisition module is used to collect data information of urinary incontinence patients; the data preprocessing module is used to preprocess the collected data information and establish a risk assessment data set; the model construction module is used to construct a urinary incontinence risk assessment prediction model; the model optimization module is used to optimize the training of the urinary incontinence risk assessment prediction model; the model prediction module is used to conduct urinary incontinence risk assessment on the tester and output the risk assessment results.

[0075] Therefore, the present invention adopts the above-mentioned method for constructing a urinary incontinence risk assessment prediction model based on a neural network, and the beneficial effects are as follows:

[0076] (1) Based on the total gradient norm and historical gradient benchmark, an adaptive scaling factor is designed. The gradient of the neural network parameters is pruned using the adaptive scaling factor. An adaptive learning rate is designed. The neural network parameters are updated in combination with the interpolation factor. This avoids training anomalies caused by gradient instability and gradient explosion, and enables the learning rate to better adapt to different training stages, which helps to better process complex data, improve the accuracy and reliability of the urinary incontinence risk assessment model, and improve the quality of life and convenience of users.

[0077] (2) The query value and reference position are generated based on the local position correction value and the local search ratio value, the perturbation position is generated based on the Levy flight strategy and random individuals, and the positions of all individuals are updated based on the reference position and the perturbation position to determine the optimal parameters. This improves the efficiency and accuracy of model parameter optimization, enhances the adaptability and flexibility of the model, better adapts to the individual differences of different patients, improves the generalization ability of the model, and improves the performance and reliability of the urinary incontinence risk assessment model, which helps to more accurately warn of urinary incontinence risks.

[0078] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 It is a flow chart of a method for constructing a urinary incontinence risk assessment prediction model based on a neural network according to the present invention;

[0080] Figure 2 It is a system block diagram of the neural network-based urinary incontinence risk assessment prediction model of the present invention. DETAILED DESCRIPTION

[0081] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0082] like Figure 1 As shown, the present invention provides a method for constructing a urinary incontinence risk assessment prediction model based on a neural network, comprising the following steps:

[0083] Step S1, collecting data information about patients with urinary incontinence; wherein the data information about patients with urinary incontinence comes from the hospital's electronic medical record system, patient questionnaires, clinical research, etc.;

[0084] Step S2, preprocessing the data information collected in step S1, and establishing a risk assessment data set;

[0085] Step S3: constructing a urinary incontinence risk assessment prediction model based on the risk assessment data set;

[0086] Step S4, optimizing and training the constructed urinary incontinence risk assessment prediction model;

[0087] Step S5: Based on the optimized urinary incontinence risk assessment model, risk assessment prediction is performed on the test set.

[0088] Example 1

[0089] Step S1, collecting data information about patients with urinary incontinence; wherein the data information about patients with urinary incontinence comes from the hospital's electronic medical record system, patient questionnaires, clinical research, etc.

[0090] The data information collected from patients with urinary incontinence includes: basic information of patients (such as age, height, weight, occupation, ethnicity, education level, marital status, etc.); relevant environmental risk factors (such as smoking, drinking, current menstrual status, constipation, chronic cough, number of births, mode of delivery, fetal weight, etc.); gynecological diseases (such as gynecological inflammation, chronic pelvic pain, endometriosis, uterine fibroids, etc.); history of other diseases (such as hypertension, diabetes, stroke, chronic bronchitis, cancer, depression, fecal incontinence, urinary incontinence, etc.).

[0091] Step S2: preprocess the data information collected in step S1 and establish a risk assessment data set.

[0092] Data preprocessing includes: data cleaning, data conversion and data normalization. Data cleaning includes filling missing values, correcting erroneous data, and deleting duplicate records; data conversion is to convert non-numeric data (such as categorical variables) into numerical form using One-Hot Encoding; data normalization is to standardize or normalize numerical data to eliminate the impact of dimension.

[0093] Based on the preprocessed data information, a risk assessment dataset is constructed.

[0094] Step S3: constructing a urinary incontinence risk assessment prediction model based on the risk assessment data set.

[0095] Step S31, divide the risk assessment data set into a training set and a test set, and set the ratio of the training set to the test set to 8:2, and build a urinary incontinence risk assessment prediction model. Among them, the urinary incontinence risk assessment prediction model is a neural network model, which specifically includes an input layer, a hidden layer and an output layer; the input layer is responsible for receiving data in the risk assessment data set, the hidden layer performs data processing and feature extraction, and the output layer outputs the assessment label.

[0096] Step S32: Update the neural network model parameters.

[0097] Step S321, calculate the first-order moment and the second-order moment, as shown below:

[0098]

[0099] m t =β1m t-1 +(1-β1)g t ;

[0100]

[0101] Among them, g t is the gradient of the loss function to the neural network parameters at the tth update; is the gradient on the parameter θ; J is the loss function; θ t-1 is the neural network parameter at the t-1th update; m t and m t-1 are the first-order moments at the t-th and t-1-th updates respectively; v t and v t-1 are the second-order moments at the t-th and t-1-th updates, respectively; β1 and β2 are the first-order moment decay rates and the second-order moment decay rates; Yes t The square of , t is the update index of the neural network parameters.

[0102] Step S322, offset correction is as follows:

[0103]

[0104]

[0105] in, and are the first-order moment and second-order moment after bias correction at the tth update; and are the t-th power of β1 and β2 respectively.

[0106] Step S323, gradient pruning, designing an adaptive scaling factor, and pruning the gradient of the neural network parameters based on the adaptive scaling factor.

[0107] The total gradient norm is calculated as follows:

[0108]

[0109] Among them, sf t is the total gradient norm at the tth update; d is the parameter index, D is the total number of neural network parameters, is the gradient of the dth parameter at the jth update, is the square of the L2 norm.

[0110] Calculate the historical gradient benchmark, which includes the historical gradient mean and historical gradient variance, as shown below:

[0111]

[0112] Among them, lz t and lc t are the historical gradient mean and historical gradient variance at the tth update; g j is the gradient of the loss function with respect to the neural network parameters at the jth update.

[0113] Design the adaptive scaling factor as follows:

[0114]

[0115] Among them, zs t is the adaptive scaling factor at the tth update; gn max is the maximum gradient norm; γ is the control factor.

[0116] Apply trimming as follows:

[0117]

[0118] in, is the gradient of the dth parameter after clipping at the jth update.

[0119] Step S325: Design an adaptive learning rate, dynamically adjust the learning rate according to time and set parameters, and the learning rate changes periodically during the training process, as shown below:

[0120]

[0121] Among them, xl t is the adaptive learning rate at the tth update; xl0, xl fd and xl zq They are the initial value of the learning rate, the cycle length of the learning rate change, and the amplitude of the learning rate change.

[0122] Step S325, neural network model parameter update, neural network parameters are updated based on the adaptive learning rate and interpolation factor, as shown below:

[0123]

[0124] Among them, θ t is the neural network parameter at the tth update; σ is the interpolation factor, and ∈ is the smoothing term.

[0125] Step S4: Optimize and train the constructed urinary incontinence risk assessment prediction model.

[0126] Initialize the individual position, and adjust the first-order moment decay rate β1, second-order moment decay rate β2, and maximum gradient norm gn in the urinary incontinence risk assessment model max , control factor γ, initial value of learning rate xl0, cycle length of learning rate change xl fd , the magnitude of the learning rate change xl zq A search space is established with the interpolation factor σ, A individual positions are randomly initialized in the search space, and the individual positions are used as representatives of the parameters of the urinary incontinence risk assessment model. The loss function of the urinary incontinence risk assessment model established based on the model parameters is used as the individual fitness value.

[0127] Design reference locations, including the following:

[0128] Step S41: Calculate the local search ratio value as follows:

[0129]

[0130] Where Δh(u) is the local search ratio value at the uth iteration; x best (u) is the position of the global optimal individual at the uth iteration; is the average position of all individuals at the uth iteration; u is the current iteration number; i is the individual index; rand(·) is the random number generation function; x i (u) is the position of the i-th individual at the u-th iteration.

[0131] Step S42: Calculate the local position correction value as follows:

[0132]

[0133] Among them, w i (u) is the local position correction value of the ith individual at the uth iteration; x worst (u) is the position of the global worst individual at the u-th iteration.

[0134] Step S43: Calculate the query value, and generate two different query values ​​for each individual based on the local position correction value and the local search ratio value, as shown below:

[0135]

[0136] in, and are the first query value and the second query value of the i-th individual at the u-th iteration, respectively.

[0137] Step S44: Calculate two reference positions, generate two different reference positions for each individual based on the query value, and guide the individual to move in a more promising direction, as shown below:

[0138]

[0139] in, and are the first reference individual and the second reference position of the i-th individual at the u-th iteration; x r1 (u) is the position of the first random individual at the u-th iteration.

[0140] Step S45, calculate the disturbance position, and generate a disturbance position for each individual based on the Levy flight strategy and the random individual, as shown below:

[0141]

[0142] Among them, i (u) is the perturbed position of the ith individual at the uth iteration; a is a random number generated based on the Levy flight strategy; b and p are taken from the normal distribution; b~N(0,λ 2 ), p~N(0,λ 2 ), λ is the shape parameter, Γ(·) is the gamma function, ε is the standard deviation of the normal distribution, x r2 (u) and x r3 (u) are the positions of the second and third random individuals at the u-th iteration; is the ceiling function.

[0143] Step S46: Individual position update: all individual positions are updated based on the reference position and the disturbance position, as shown below:

[0144]

[0145] Among them, x i (u+1) is the position of the i-th individual at the u+1-th iteration; s1 and s2 are the first and second random values ​​in the range of (0,1), respectively.

[0146] Step S47, optimal parameters are determined, a fitness threshold is pre-set, the individual fitness value and the global optimal individual are updated, when the fitness value corresponding to the global optimal individual is lower than the fitness threshold, the model parameters corresponding to the global optimal individual position are the optimal parameters, and a urinary incontinence risk assessment model is constructed based on the optimal parameters; otherwise, if the maximum number of iterations is reached, the individual position is reinitialized; otherwise, the number of iterations is increased by 1 and the reference position and disturbance position are recalculated, and the individual position is updated.

[0147] Step S5: Based on the optimized urinary incontinence risk assessment model, risk assessment prediction is performed on the test set.

[0148] First, the test set data is input into the urinary incontinence risk assessment model; second, the risk assessment threshold is set in advance. When the risk assessment level output by the model is higher than the risk assessment threshold, the smart chip sends a warning message; finally, the risk assessment report is output for subsequent diagnostic assistance.

[0149] Example 2

[0150] Based on the above, if Figure 2 As shown, a neural network-based urinary incontinence risk assessment prediction model of the present invention includes: a data acquisition module, a data preprocessing module, a model building module, a model optimization module and a model prediction module.

[0151] Among them, the data acquisition module is used to collect data information of urinary incontinence patients; the data preprocessing module is used to preprocess the collected data information and establish a risk assessment data set; the model construction module is used to construct a urinary incontinence risk assessment prediction model; the model optimization module is used to optimize the training of the urinary incontinence risk assessment prediction model; the model prediction module is used to conduct urinary incontinence risk assessment on the tester and output the risk assessment results.

[0152] Therefore, the present invention adopts the above-mentioned method for constructing a urinary incontinence risk assessment and prediction model based on a neural network, designs an adaptive scaling factor based on the total gradient norm and the historical gradient benchmark, uses the adaptive scaling factor to prune the gradient of the neural network parameters, designs an adaptive learning rate, and updates the neural network parameters in combination with the interpolation factor, thereby avoiding training anomalies caused by gradient instability and gradient explosion, so that the learning rate can better adapt to different training stages, which helps to better process complex data, improves the accuracy and reliability of the urinary incontinence risk assessment model, and improves the quality of life and convenience of users; generates query values ​​and reference positions based on local position correction values ​​and local search ratio values, generates perturbation positions based on Levy flight strategies and random individuals, and updates the positions of all individuals based on the reference positions and perturbation positions, determines the optimal parameters, improves the efficiency and accuracy of model parameter optimization, enhances the adaptability and flexibility of the model, better adapts to the individual differences of different patients, improves the generalization ability of the model, improves the performance and reliability of the urinary incontinence risk assessment model, and helps to more accurately warn of urinary incontinence risks.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for constructing a urinary incontinence risk assessment prediction model based on a neural network, characterized in that: The following steps are involved: Step S1, collecting data information about patients with urinary incontinence; Step S2, preprocessing the data information collected in step S1, and establishing a risk assessment data set; Step S3: constructing a urinary incontinence risk assessment prediction model based on the risk assessment data set; Step S4, optimizing and training the constructed urinary incontinence risk assessment prediction model; Step S5: Based on the optimized urinary incontinence risk assessment model, risk assessment prediction is performed on the test set.

2. The method for constructing a neural network-based urinary incontinence risk assessment prediction model according to claim 1, characterized in that: In step S1, the data information of patients with urinary incontinence is obtained from the hospital's electronic medical record system, patient questionnaires, and clinical research; The data information of patients with urinary incontinence includes: basic information of patients, relevant environmental risk factors, gynecological diseases and other medical history.

3. The method for constructing a neural network-based urinary incontinence risk assessment prediction model according to claim 1, characterized in that: In step S2, data preprocessing includes: data cleaning, data conversion and data normalization; Among them, data cleaning includes filling missing values, correcting erroneous data, and deleting duplicate records; data conversion is to convert non-numeric data into numerical form using unique hot encoding; data normalization is to standardize or normalize numerical data to eliminate the impact of dimension; Based on the preprocessed data information, a risk assessment dataset is constructed.

4. The method for constructing a neural network-based urinary incontinence risk assessment prediction model according to claim 1, characterized in that: In step S3, a urinary incontinence risk assessment prediction model is constructed based on the risk assessment data set. The specific process is as follows: Step S31, dividing the risk assessment data set into a training set and a test set, and setting the ratio of the training set to the test set to 8:2, to construct a urinary incontinence risk assessment prediction model; Among them, the urinary incontinence risk assessment prediction model is a neural network model, which specifically includes an input layer, a hidden layer and an output layer; the input layer is responsible for receiving data in the risk assessment data set, the hidden layer performs data processing and feature extraction, and the output layer outputs an assessment report; Step S32: Update the neural network model parameters.

5. The method for constructing a neural network-based urinary incontinence risk assessment prediction model according to claim 4, characterized in that: In step S32, the neural network model parameters are updated, and the specific process is as follows: Step S321, calculate the first-order moment and the second-order moment, as shown below: m t =β1m t-1 +(1-β1)g t ; Among them, g t is the gradient of the loss function to the neural network parameters at the tth update; is the gradient on the parameter θ; J is the loss function; θ t-1 is the neural network parameter at the t-1th update; m t and m t-1 are the first-order moments at the t-th and t-1-th updates respectively; v t and v t-1 are the second-order moments at the t-th and t-1-th updates, respectively; β1 and β2 are the first-order moment decay rates and the second-order moment decay rates; Yes t The square of , t is the update index of the neural network parameters; Step S322, offset correction is as follows: in, and are the first-order moment and second-order moment after bias correction at the tth update; and are the t-th power of β1 and β2 respectively; Step S323, gradient pruning, designing an adaptive scaling factor, and pruning the gradient of the neural network parameters based on the adaptive scaling factor; The total gradient norm is calculated as follows: Among them, sf t is the total gradient norm at the tth update; d is the parameter index, D is the total number of neural network parameters, is the gradient of the dth parameter at the jth update, is the square of the L2 norm; Calculate the historical gradient benchmark, which includes the historical gradient mean and historical gradient variance, as shown below: Among them, lz t and lc t are the historical gradient mean and historical gradient variance at the tth update; g j is the gradient of the loss function with respect to the neural network parameters at the jth update; Design the adaptive scaling factor as follows: Among them, zs t is the adaptive scaling factor at the tth update; gn max is the maximum gradient norm; γ is the control factor; Apply trimming as follows: in, is the gradient of the dth parameter after pruning at the jth update; Step S325: Design an adaptive learning rate, dynamically adjust the learning rate according to time and set parameters, and the learning rate changes periodically during the training process, as shown below: Among them, xl t is the adaptive learning rate at the tth update; xl0, xl fd and xl zq They are the initial value of the learning rate, the cycle length of the learning rate change, and the amplitude of the learning rate change; Step S325, neural network model parameter update, neural network parameters are updated based on the adaptive learning rate and interpolation factor, as shown below: Among them, θ t is the neural network parameter at the tth update; σ is the interpolation factor, and ∈ is the smoothing term.

6. The method for constructing a neural network-based urinary incontinence risk assessment prediction model according to claim 5, characterized in that: In step S4, the constructed urinary incontinence risk assessment prediction model is optimized and trained, and the specific process is as follows: Initialize the individual position, and adjust the first-order moment decay rate β1, second-order moment decay rate β2, and maximum gradient norm gn in the urinary incontinence risk assessment model max , control factor γ, initial value of learning rate xl0, cycle length of learning rate change xl fd , the magnitude of the learning rate change xl zq A search space is established with the interpolation factor σ, A individual positions are randomly initialized in the search space, and the individual positions are used as representatives of the parameters of the urinary incontinence risk assessment model. The loss function of the urinary incontinence risk assessment model established based on the model parameters is used as the individual fitness value.

7. The method for constructing a neural network-based urinary incontinence risk assessment prediction model according to claim 6, characterized in that: Design reference positions, including the following: Step S41: Calculate the local search ratio value as follows: Where Δh(u) is the local search ratio value at the uth iteration; x best (u) is the position of the global optimal individual at the uth iteration; is the average position of all individuals at the uth iteration; u is the current iteration number; i is the individual index; rand(·) is the random number generation function; x i (u) is the position of the i-th individual at the u-th iteration; Step S42: Calculate the local position correction value as follows: Among them, w i (u) is the local position correction value of the ith individual at the uth iteration; x worst (u) is the position of the global worst individual at the u-th iteration; Step S43: Calculate the query value, and generate two different query values ​​for each individual based on the local position correction value and the local search ratio value, as shown below: in, and are the first query value and the second query value of the i-th individual at the u-th iteration respectively; Step S44: Calculate two reference positions, generate two different reference positions for each individual based on the query value, and guide the individual to move in a more promising direction, as shown below: in, and are the first reference individual and the second reference position of the i-th individual at the u-th iteration; x r1 (u) is the position of the first random individual at the u-th iteration; Step S45, calculate the disturbance position, and generate a disturbance position for each individual based on the Levy flight strategy and the random individual, as shown below: Among them, i (u) is the perturbed position of the ith individual at the uth iteration; a is a random number generated based on the Levy flight strategy; b and p are taken from the normal distribution; b~N(0,λ 2 ), p~N(0,λ 2 ), λ is the shape parameter, Γ(·) is the gamma function, ε is the standard deviation of the normal distribution, x r2 (u) and x r3 (u) are the positions of the second and third random individuals at the u-th iteration; is the ceiling function; Step S46: Individual position update: all individual positions are updated based on the reference position and the disturbance position, as shown below: Among them, x i (u+1) is the position of the i-th individual at the u+1-th iteration; s1 and s2 are the first and second random values ​​in the range (0,1), respectively; Step S47, optimal parameters are determined, a fitness threshold is pre-set, the individual fitness value and the global optimal individual are updated, when the fitness value corresponding to the global optimal individual is lower than the fitness threshold, the model parameters corresponding to the global optimal individual position are the optimal parameters, and a urinary incontinence risk assessment model is constructed based on the optimal parameters; otherwise, if the maximum number of iterations is reached, the individual position is reinitialized; otherwise, the number of iterations is increased by 1 and the reference position and disturbance position are recalculated, and the individual position is updated.

8. The method for constructing a neural network-based urinary incontinence risk assessment prediction model according to claim 1, characterized in that: In step S5, based on the optimized urinary incontinence risk assessment model, risk assessment prediction is performed on the test set. The specific process is as follows: First, the test set data is input into the urinary incontinence risk assessment model; second, the risk assessment threshold is pre-set, and when the risk assessment level output by the model is higher than the risk assessment threshold, the smart chip sends a warning message; Finally, the risk assessment report is output for subsequent diagnostic assistance.

9. A urinary incontinence risk assessment prediction model constructed by the method for constructing a urinary incontinence risk assessment prediction model based on a neural network according to any one of claims 1 to 8, characterized in that: It includes data acquisition module, data preprocessing module, model building module, model optimization module and model prediction module; Among them, the data acquisition module is used to collect data information of patients with urinary incontinence; the data preprocessing module is used to preprocess the collected data information and establish a risk assessment data set; the model building module is used to build a urinary incontinence risk assessment prediction model; The model optimization module is used to optimize the training of the urinary incontinence risk assessment prediction model; the model prediction module is used to conduct urinary incontinence risk assessment on the test subject and output the risk assessment results.

Images