A digital early warning system for breast cancer postoperative lymphedema

Abstract

The application relates to the technical field of health prediction, and discloses a breast cancer postoperative lymphedema digital early warning system, which comprises a data acquisition module, a data application module and a model acquisition module.The data acquisition module is used for acquiring lymphedema monitoring data after breast cancer surgery.The data application module is used for training a neural network correction model for correcting the lymphedema monitoring data based on the acquired lymphedema monitoring data.The model acquisition module is used for acquiring to-be-decided data from the lymphedema monitoring data and applying the to-be-decided data to the neural network-based correction model.Compared with a traditional method, the lymphedema diagnosis accuracy is improved by reducing misjudgment caused by various interference factors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of health prediction, in particular to a digital early warning system for breast cancer postoperative lymphedema. BACKGROUND

[0002] Breast cancer surgery, especially axillary lymph node dissection, can destroy the normal pathway of lymphatic return, causing lymphatic fluid to accumulate locally and increasing the risk of lymphedema. Postoperative lymphedema not only affects the patient's limb function, but also can cause complications such as pain and infection, which seriously reduces the patient's quality of life. Therefore, early warning is crucial to improve patient outcomes.

[0003] In the detection of breast cancer postoperative lymphedema, bioelectrical impedance analysis (BIA) is an important basis for assessing body composition and physiological status by measuring the electrical impedance characteristics of human tissues to weak alternating current. In human tissues, water, electrolytes, and other factors are key factors affecting electrical impedance. When lymphedema occurs, the interstitial fluid increases, causing changes in the electrical impedance of the local tissue. By placing electrodes on the body surface and injecting weak alternating current into the body, the voltage and current can be measured to calculate the electrical impedance value and determine the changes in tissue water content to indicate the possibility of lymphedema.

[0004] Currently, patients are prone to electrolyte imbalance after surgery, such as dietary changes, drug side effects, and disease itself, which can cause electrolyte imbalance and changes in electrical impedance values. If not corrected, it will affect subsequent electrical impedance measurement results. Therefore, a digital early warning system for breast cancer postoperative lymphedema is proposed. SUMMARY

[0005] The main purpose of the present application is to provide a digital early warning system for breast cancer postoperative lymphedema to solve the problems raised in the background.

[0006] A digital early warning system for breast cancer postoperative lymphedema, comprising:

[0007] A data acquisition module for acquiring lymphedema monitoring data after breast cancer surgery;

[0008] A data application module for training a neural network correction model for correcting lymphedema monitoring data based on the acquired lymphedema monitoring data;

[0009] A model acquisition module for obtaining decision data from the lymphedema monitoring data and applying the decision data to the neural network-based correction model.

[0010] The lymphedema monitoring data after breast cancer surgery includes skin humidity value data and original electrical impedance value data.

[0011] The process of training a neural network correction model for correcting lymphedema monitoring data is as follows:

[0012] Correction and normalization were performed on the lymphedema monitoring data;

[0013] Skin moisture data is expressed as The original electrical impedance data are expressed as Experimental data was read from the local database, outlier data points were removed, and for each skin moisture value... The electrical impedance value below Obtain its average value and standard deviation ,when If so, then delete the data point;

[0014] Skin moisture value data and original electrical impedance data The normalization operation is performed, and the formula is expressed as: ,in, These are normalized skin moisture values.

[0015] For the original electrical impedance data Perform normalization: ,in, These are the normalized original electrical impedance values.

[0016] The corrected and normalized lymphedema monitoring data were divided into training set, validation set and test set, with the ratio of training data to validation set.

[0017] Construct a feedforward neural network with an input layer, hidden layers, and an output layer;

[0018] The input layer nodes are the corrected and normalized humidity values. The output layer nodes are corrected and normalized impedance values. Hidden layer settings A neuron, through a Activation function, used to initialize the weights of the neural network. and bias Weight in arrive Random initialization is performed between intervals, and the bias is initialized to 0.

[0019] The feedforward neural network is trained using the training set data, and the corrected and normalized humidity values ​​in the dataset are then processed. and electrical impedance value Propagate forward;

[0020] First, from the input layer to the hidden layer, let the weights of the input layer be... , the input of the neuron in the hidden layer is: where, is the input of the neuron in the hidden layer, when p = 1, t = 0, = 1;

[0021] Further, by weighting and summing the corrected and normalized humidity values of the input layer with weights, and adding a bias, the input signal of the hidden layer neuron is obtained, determines the degree of influence of the input humidity value on the hidden layer neuron, different values can make the model pay attention to different features or trends of the humidity value, provides an additional learnable parameter for the neuron, allowing the neuron to have a certain output even when the input is zero, increasing the flexibility of the model;

[0022] The output of the neuron in the hidden layer is: , is the activation function;

[0023] By activating the input signal, the output of the hidden layer is obtained, which will be used as the input of the next layer (hidden layer to output layer) to continue passing and processing information;

[0024] Then the hidden layer to the output layer, let the weight of the input layer be , the bias is , the output of the neuron in the hidden layer is: where, is the number of neurons in the hidden layer, is the output value of the output layer to obtain a feedforward neural network;

[0025] Further, the output of the hidden layer is weighted and summed with the weight of the hidden layer to the output layer , and a bias is added to obtain the final prediction output. By adjusting, the model can learn how to convert the features extracted by the hidden layer into accurate electrical impedance value predictions, providing an adjustable parameter that helps optimize the prediction results;

[0026] Train a loss function, the formula is: where, is the number of samples in the training set;

[0027] ​Further, the loss function measures the difference between the model's predicted value and the true value. For each lymphedema monitoring data sample in the training set, the square of the difference between the predicted value and the true value is calculated, then the sum of the squared differences of all samples is summed and averaged, the smaller the value of the loss function, the better the prediction effect of the model;

[0028] Finally, the trained neural network model is evaluated using the validation set and test set to obtain the coefficient of determination , wherein, is the average of the actual normalized impedance value, and the best performing model is selected based on the evaluation results;

[0029] Further, The coefficient of determination reflects the fitting degree of the model to the data, with a value range of 0 to 1, the closer to 1, the better the model fits the data; on the contrary, the closer to 0, the poorer the fitting effect of the model, the numerator part represents the total squared error between the model's predicted value and the true value, and the denominator part represents the total squared error between the true value and the average of the true value, which is 0 if the model cannot predict the data at all, and 1 if the model can perfectly predict the data.

[0030] The data to be decided includes normalized and corrected skin humidity value data and original impedance value data ;

[0031] Further, by using the trained neural network correction model to process new lymphedema monitoring data, early warning of lymphedema can be achieved. After inputting the normalized and corrected skin humidity value into the model, the model can predict and correct the impedance value based on the relationship between humidity and impedance learned during the training phase. By passing the original impedance value, the normalized and corrected impedance value output by the neural network model can be converted into the actual corrected impedance value in the model application submodule. This actual corrected impedance value, combined with the original impedance value and other clinical information, can more accurately determine whether the patient is at risk of lymphedema, providing more valuable diagnostic basis for doctors;

[0032] Further, the early warning module compares the predicted impedance value with the patient's previous measurement value and the normal reference value of the same type of patient based on the output decision result of the neural network correction model. If the predicted impedance value exceeds the normal range or shows an abnormal change trend (such as the measured value gradually deviating from the normal interval for several consecutive times), the system triggers the lymphedema warning mechanism.

[0033] The process of applying the data to be decided to the neural network-based correction model is:

[0034] The normalized corrected skin moisture value data is represented Input into the neural network correction model, and obtain a predicted corrected electrical impedance value based on the neural network correction model ;

[0035] Further, the normalized corrected electrical impedance value is calculated through forward propagation, and the lymphedema monitoring data is sequentially passed through each hidden layer from the input layer to the output layer, in the process, the weighted sum of each layer of neurons and the activation function are passed and transformed, and finally the predicted corrected electrical impedance value is obtained ;

[0036] Based on the neural network correction model, the predicted corrected electrical impedance value is inversely normalized

[0037] Then, the system compares the corrected electrical impedance value with the previous measurement value of the patient and the normal reference value of the same type of patient, if the predicted electrical impedance value exceeds the normal range or shows an abnormal change trend (such as the measured value gradually deviates from the normal interval for several times in succession), the system triggers the lymphedema early warning mechanism, provides personalized early warning scheme according to the specific situation of the patient, analyzes the monitoring data and early warning results of the patient, and the system can formulate targeted rehabilitation training plan, pressure treatment scheme and lifestyle adjustment suggestion for the patient, so as to help the patient reduce the risk of lymphedema and improve the postoperative rehabilitation effect.

[0038] A digital early warning method for lymphedema after breast cancer surgery, comprising:

[0039] S1: used for collecting lymphedema monitoring data after breast cancer surgery;

[0040] S2: based on the obtained lymphedema monitoring data, a neural network correction model for correcting the lymphedema monitoring data is trained;

[0041] S3: obtaining decision-making data from the lymphedema monitoring data, and applying the decision-making data to the neural network correction model;

[0042] S4: the decision-making result output by the neural network correction model is compared with the previous measurement value of the patient.

[0043] The present application has the following beneficial effects:

[0044] 0、In the present application, the input layer receives the normalized humidity value, the output layer outputs the normalized electrical impedance value, and the hidden layer is provided with a plurality of neurons and adopts The activation function can effectively learn the complex nonlinear relationship between skin humidity and electrical impedance value, and in the training process, through the learning of a large amount of experimental data, the model can accurately capture the change rule of the electrical impedance value under different humidity conditions, so that in actual application, the change trend of the electrical impedance value can be more accurately predicted according to the input skin humidity value.

[0045] 1、In the present application, the system can collect the lymphedema monitoring data of the patient in real time, including the skin humidity value and the original electrical impedance value, and quickly input the normalized corrected skin humidity value into the trained neural network correction model, and the model quickly calculates the predicted corrected electrical impedance value based on the learned humidity and electrical impedance relationship.

[0046] 2、In the present application, in the model application process, by transmitting the original electrical impedance value, the system can perform the inverse normalization operation in the model application submodule to convert the normalized corrected electrical impedance value output by the neural network model into the actual corrected electrical impedance value, which can more comprehensively and accurately judge the numerical accuracy of the electrical impedance value, and reduce the misjudgment caused by various interference factors (such as skin humidity change, electrolyte imbalance, etc.). BRIEF DESCRIPTION OF DRAWINGS

[0047] Fig. 1 The system block diagram of the breast cancer postoperative lymphedema digital early warning system of the present application;

[0048] Fig. 2 The method step diagram of the breast cancer postoperative lymphedema digital early warning system of the present application. DETAILED DESCRIPTION

[0049] In order to make the technical means, creative features, purposes and effects achieved by the present application easy to understand, the present application will be further described below in conjunction with specific embodiments.

[0050] Embodiment one

[0051] Please refer to Figs. 1-2 The breast cancer postoperative lymphedema digital early warning system shown in the figure comprises:

[0052] Data acquisition module: used for acquiring lymphedema monitoring data after breast cancer surgery;

[0053] Data application module: based on the acquired lymphedema monitoring data, a neural network correction model for correcting lymphedema monitoring data is trained;

[0054] Model acquisition module: acquires the data to be decided from the lymphedema monitoring data, and applies the data to be decided to the correction model based on the neural network.

[0055] Post-mastectomy lymphedema monitoring data includes skin moisture values ​​and raw electrical impedance values.

[0056] Furthermore, the data acquisition module is fundamental to the entire early warning process. It is responsible for collecting lymphedema monitoring data from post-operative breast cancer patients, including skin moisture values ​​and raw electrical impedance values. This data provides raw material for subsequent model training and early warning judgments.

[0057] The process of training a neural network correction model for correcting lymphedema monitoring data is as follows:

[0058] Correction and normalization were performed on the lymphedema monitoring data;

[0059] Skin moisture data is expressed as The original electrical impedance data are expressed as Experimental data was read from the local database, outlier data points were removed, and for each skin moisture value... The electrical impedance value below Obtain its average value and standard deviation ,when If so, then delete the data point;

[0060] Skin moisture value data and original electrical impedance data The normalization operation is performed, and the formula is expressed as: ,in, These are normalized skin moisture values.

[0061] For the original electrical impedance data Perform normalization: ,in, These are the normalized original electrical impedance values.

[0062] The corrected and normalized lymphedema monitoring data were divided into training set, validation set and test set, with the ratio of training data to validation set.

[0063] Construct a feedforward neural network with an input layer, hidden layers, and an output layer;

[0064] The input layer nodes are the corrected and normalized humidity values. The output layer nodes are corrected and normalized impedance values. Hidden layer settings A neuron, through a Activation function, used to initialize the weights of the neural network. and bias Weight in arrive Randomly initialized between, bias initialized to 0;

[0065] The forward propagation is performed using the training set data to train the feedforward neural network, the corrected and normalized humidity values in the dataset And the electrical impedance values The forward propagation is performed using the training set data to train the feedforward neural network, the corrected and normalized humidity values in the dataset

[0066] First, the input layer to the hidden layer, the input layer weight is set to , the bias is , the input of the hidden layer neuron is: , wherein is the input of the hidden layer neuron, when p = 1, t = 0, = 1;

[0067] Further, by weighting the corrected and normalized humidity values of the input layer and the weights and adding the bias, the input signal of the hidden layer neuron is obtained, The influence degree of the input humidity value on the hidden layer neuron is determined, and different values can make the model pay attention to different features or trends of the humidity value, An additional learnable parameter is provided for the neuron, which enables the neuron to have a certain output even when the input is zero, increasing the flexibility of the model;

[0068] The output of the hidden layer neuron is: , The activation function is

[0069] The input signal is nonlinearly transformed by the activation function to obtain the output of the hidden layer, which will be used as the input of the next layer (hidden layer to output layer) to continue passing and processing information;

[0070] Then the hidden layer to the output layer, the input layer weight is set to , the bias is , the output of the hidden layer neuron is: , wherein is the number of hidden layer neurons, is the output value of the output layer, and a feedforward neural network is obtained;

[0071] Further, the output of the hidden layer is weighted and summed with the weight from the hidden layer to the output layer , and the bias is added to obtain the final prediction output. By adjusting, the model can learn how to convert the features extracted by the hidden layer into accurate electrical impedance value predictions, providing an adjustable parameter that helps optimize the prediction results;

[0072] A loss function is trained, and the formula is: wherein, is the number of samples in the training set;

[0073] Further, the loss function measures the difference between the model prediction value and the true value. For each lymphedema monitoring data sample in the training set, the square of the difference between the predicted value and the true value is calculated, and then the sum of the squared differences of all samples is summed and averaged. The smaller the value of the loss function, the better the prediction effect of the model;

[0074] Finally, the trained neural network model is evaluated using the validation set and the test set to obtain the determination coefficient wherein, is the average value of the actual normalized impedance value, and the model with the best performance is selected based on the evaluation results;

[0075] Further, The determination coefficient reflects the fitting degree of the model to the data, and the value range is between 0 and 1. The closer to 1, the better the model fits the data; on the contrary, the closer to 0, the poorer the fitting effect of the model. The numerator part represents the total squared error between the model prediction value and the true value, and the denominator part represents the total squared error between the true value and the average value of the true value. If the model cannot predict the data at all, it is 0; if the model can perfectly predict the data, it is 1.

[0076] The data to be decided includes normalized corrected skin humidity value data and original impedance value data ;

[0077] Further, by using the trained neural network correction model to process new lymphedema monitoring data, early warning of lymphedema can be realized. After inputting the normalized corrected skin humidity value into the model, the model can predict and correct the impedance value according to the relationship between humidity and impedance learned in the training phase. By passing the original impedance value, the inverse normalization operation can be performed in the model application submodule to convert the normalized corrected impedance value output by the neural network model into the actual corrected impedance value. This actual corrected impedance value, combined with the original impedance value and other clinical information, can more accurately determine whether the patient has a risk of lymphedema, providing more valuable diagnostic basis for doctors;

[0078] Further, the early warning module compares the predicted electrical impedance value with the previous measurement value of the patient and the normal reference value of the same type of patient based on the output decision result of the neural network correction model. If the predicted electrical impedance value exceeds the normal range or presents an abnormal change trend (such as the gradual deviation of the continuous multiple measurement values from the normal interval), the system triggers the lymphedema early warning mechanism.

[0079] The process of applying the data to be decided to the neural network-based correction model is:

[0080] The normalized corrected skin humidity value data is represented Input into the neural network correction model, and obtain the predicted corrected electrical impedance value based on the neural network correction model ;

[0081] Further, the normalized corrected electrical impedance value is calculated through forward propagation. Through the lymphedema monitoring data, the input layer is sequentially passed through each hidden layer, and finally reaches the output layer. In this process, the weighted sum and activation function of each layer of neurons are used for transmission and transformation, and finally the predicted corrected electrical impedance value is obtained ;

[0082] The neural network correction model is based on the predicted corrected electrical impedance value Reverse normalization processing is performed;

[0083] Then, the system compares the corrected electrical impedance value with the previous measurement value of the patient and the normal reference value of the same type of patient. If the predicted electrical impedance value exceeds the normal range or presents an abnormal change trend (such as the gradual deviation of the continuous multiple measurement values from the normal interval), the system triggers the lymphedema early warning mechanism, and provides personalized recommendation programs according to the specific situation of the patient. Through the analysis of the monitoring data and the early warning result of the patient, the system can formulate targeted rehabilitation training plan, pressure treatment scheme and lifestyle adjustment suggestion for the patient, so as to help the patient reduce the risk of lymphedema and improve the postoperative rehabilitation effect.

[0084] Please refer to Figs. 1-2 The digital early warning method for lymphedema after breast cancer surgery is as follows:

[0085] S1: used for collecting lymphedema monitoring data after breast cancer surgery;

[0086] S2: based on the obtained lymphedema monitoring data, a neural network correction model for correcting the lymphedema monitoring data is trained;

[0087] S3: obtain the data to be decided from the lymphedema monitoring data, and apply the data to be decided to the neural network-based correction model;

[0088] S4: outputting a decision result based on the correction model of the neural network, and comparing the result with previous measurement values of the patient.

[0089] The above shows and describes the basic principles and main features of the present application and the advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only to illustrate the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1. A digital early warning system for breast cancer postoperative lymphedema, characterized in that, The application relates to a lymphedema monitoring method for postoperative breast cancer, which comprises the following steps: a data collection module is used to collect lymphedema monitoring data after breast cancer surgery; a data application module is used to train a neural network correction model for correcting lymphedema monitoring data based on the obtained lymphedema monitoring data; Model obtaining module: obtaining to-be-decided data from lymphedema monitoring data, applying the to-be-decided data to a neural network-based correction model to obtain a predicted correction electrical impedance value the training process of the neural network correction model is as follows: The skin moisture value data is denoted as R, the original electrical impedance value data is denoted as Z, the lymphedema monitoring data is corrected and normalized to obtain the corrected and normalized skin moisture value data The corrected and normalized electrical impedance value data is denoted as the corrected and normalized lymphedema monitoring data is divided into a training set, a verification set and a test set; a feedforward neural network with an input layer, a hidden layer and an output layer is constructed; a feedforward neural network training loss function is used to evaluate the trained neural network model by using the verification set and the test set, and the model with the best performance is selected according to the coefficient of determination evaluation result; the process of constructing the feedforward neural network with the input layer, the hidden layer and the output layer is as follows: corrected and normalized humidity values as input layer nodes, corrected and normalized electrical impedance values as output layer nodes, the hidden layer is set with q neurons, a ReLU activation function is used, the weights W and biases b of the neural network are initialized, the weights are randomly initialized between -0.5 and 0.5, and the biases are initialized as 0; The forward neural network is trained using the training set data, and the normalized humidity values in the data set and electrical impedance values are forward propagated to ultimately obtain a forward neural network; The corrected and normalized humidity values in the data set and electrical impedance values The process of forward propagation is: First input layer to hidden layer, set the input layer weight W1, bias b1, then the hidden layer neuron input is: Wherein, h1 is the hidden layer neuron input, when p = 1, t = 0, the neuron output of the hidden layer is h2=ReLU (h1), and ReLU is an activation function; Then the hidden layer to the output layer, set the input layer weight W2, bias b2, then the output layer of neurons output is: Wherein, q is the number of hidden layer neurons, Output value of the output layer; the process of selecting the model with the best performance according to the coefficient of determination evaluation result is as follows: Training a loss function: where N is the number of samples in the training set; The trained neural network model is evaluated using the validation set and the test set to obtain a coefficient of determination wherein, is an average value of the actual normalized electrical impedance value, and the model with the best performance is selected based on the evaluation result.

2. The breast cancer postoperative lymphedema digital early warning system according to claim 1, characterized in that: the lymphedema monitoring data after the breast cancer surgery comprises skin humidity value data and original electrical impedance value data. 3.The breast cancer postoperative lymphedema digital early warning system according to claim 1, characterized in that: the process of correcting and normalizing the lymphedema monitoring data is as follows: Reading experimental data from local database, removing abnormal data points, for each impedance value Z at a skin humidity value R, obtaining its average value and standard deviation σ, when then the data point is removed; The skin humidity value data R and the original electrical impedance value data Z are normalized, and the formula is expressed as: wherein, is the normalized skin humidity value data; The raw electrical impedance value data Z is normalized: wherein, is the normalized electrical impedance value data.

4. The breast cancer postoperative lymphedema digital early warning system according to claim 1, characterized in that: the to-be-decided data comprises normalized and corrected skin humidity value data and original electrical impedance value data.

5. The breast cancer postoperative lymphedema digital early warning system according to claim 1, characterized in that: the process of applying the to-be-decided data to the neural network-based correction model is as follows: representing the normalized corrected skin moisture value data inputting into the neural network correction model, obtaining the predicted corrected electrical impedance value based on the neural network correction model Correcting predicted resistance and impedance values based on a neural network correction model Reverse normalization is performed.

Images