Chemotherapy risk assessment method for lung cancer patient

Abstract

The invention provides a chemotherapy risk assessment method for a lung cancer patient, and belongs to the technical field of risk prediction, and the method comprises the steps: S1, collecting lung cancer chemotherapy peripheral neuropathy influence factor data, constructing an input data set of a chemotherapy risk prediction model through the influence factor data, and carrying out the preprocessing of the input data set; s2, improving a smell optimization algorithm; s3, a lung cancer patient chemotherapy risk prediction model is constructed, an input data set is organized into input features and target output, the input features and the target output are divided into a training set and a prediction set, and a WSAO-BP prediction model method comprises the steps that the number of hidden layers and learning rate factors of a BP prediction model are optimized through an improved odor optimization algorithm, and a prediction result is obtained; training the optimized BP prediction model by using the training set to obtain a WSAO-BP prediction model; and S4, inputting the test set data into the lung cancer patient chemotherapy risk prediction model, and outputting the risk rate of the peripheral neurotoxicity symptom caused by the lung cancer patient chemotherapy. And the chemotherapy risk assessment precision of the lung cancer patient is improved.

Description

Technical Field

[0001] This invention relates to the field of risk prediction technology, and in particular to a method for assessing the risk of chemotherapy in lung cancer patients. Background Technology

[0002] Chemotherapy risk prediction methods for lung cancer patients mainly assess the efficacy and potential risks of chemotherapy through a comprehensive analysis of multidimensional information such as the patient's lifestyle and physical condition, negative emotions, and chemotherapy drugs. Predictive models are established by collecting basic patient information, such as hypertension, depression levels, chemotherapy drug dosages, and chemotherapy cycles. Chemotherapy is an important treatment for lung cancer, but not all patients benefit from it; some experience side effects, leading to peripheral neuropathy symptoms (CIPN). Effective risk prediction allows for personalized treatment plans to be developed for each patient, avoiding unnecessary treatment and improving treatment precision.

[0003] Smell Agent Optimization (SAO) is a novel heuristic optimization algorithm inspired by the process by which organisms perceive and track odor sources through smell. Unlike many other optimization algorithms, SAO performs global and local searches by simulating the interaction between organisms and odor molecules to find the optimal solution. Its working principle combines the olfactory mechanism in biology with search strategies in optimization problems. However, the algorithm's performance is highly dependent on some key parameters, such as olfactory parameters and step size. If these parameters are not chosen properly, the algorithm's convergence speed will be slowed down, or it may even get stuck in a local optimum.

[0004] Traditional BP neural network training typically relies on gradient descent algorithms, which are prone to getting stuck in local optima, especially in large datasets. Odor optimization algorithms, on the other hand, simulate the diffusion and tracking of odor molecules, enabling them to explore the search space extensively and effectively avoid getting stuck in local optima, thus possessing powerful global search capabilities. By utilizing SAO to optimize the training process of BP neural networks, the accuracy of chemotherapy risk prediction models for lung cancer patients can be significantly improved, facilitating early intervention in treatment and rehabilitation guidance. Summary of the Invention

[0005] Based on the current problems in predicting chemotherapy risks for lung cancer patients and the shortcomings of odor optimization algorithms, this invention improves the odor optimization algorithm by using an improved odor optimization algorithm (WSAO) to optimize the BP prediction model. By constructing a WSAO-BP-based chemotherapy risk prediction model for lung cancer patients, this invention predicts peripheral neuropathy symptoms (CIPN) after chemotherapy in lung cancer patients, thereby improving the accuracy of chemotherapy risk assessment for lung cancer patients and enabling early intervention and rehabilitation guidance.

[0006] This invention proposes a method for assessing the risk of chemotherapy in lung cancer patients, the specific steps of which are as follows: S1. Collect data on factors affecting peripheral neuropathy during lung cancer chemotherapy, use the data on these factors to construct an input dataset for a chemotherapy risk prediction model, and preprocess the input dataset. S2. Improvements to the odor optimization algorithm, including: S21. By improving olfactory parameters from the perspective of environmental perception and nonlinear dynamic systems, an adaptive olfactory factor based on odor attenuation and environmental feedback is proposed. S22. A mutation strategy is proposed. When trapped in a local optimum, the agent uses historical optimal path information to select a high-quality position for jumping. The direction is corrected by combining historical information so that it moves along the potential global optimum direction and the position of the agent is updated. S3. Construct a chemotherapy risk prediction model for lung cancer patients, WSAO-BP. Organize the input dataset into input features and target output, and divide it into a training set and a prediction set. The WSAO-BP prediction model method is as follows: optimize the number of hidden layers and learning rate factor of the BP prediction model using an improved odor optimization algorithm, and train the optimized BP prediction model using the training set to obtain the WSAO-BP prediction model. S4. Input the test set data into the WSAO-BP model for predicting the risk of chemotherapy in lung cancer patients, and output the risk rate of chemotherapy-induced peripheral neuropathy symptoms (CIPN) in lung cancer patients.

[0007] Preferably, the influencing factors of peripheral neuropathy caused by chemotherapy for lung cancer mainly include three aspects: lifestyle and physical condition, negative emotions, and chemotherapy drugs. This invention selects four influencing factor data for these three aspects: blood pressure, depression level, chemotherapy drug dosage, and chemotherapy cycle. These influencing factor data are denoted as input features, and the target output is the risk rate of peripheral neuropathy symptoms (CIPN). The dataset of influencing factors of peripheral neuropathy caused by chemotherapy for lung cancer is preprocessed. The preprocessing includes data organization and data feature encoding and transformation. Specifically, it checks for missing values ​​in the input dataset. For each input feature, a weighted average imputation strategy based on its similar samples is adopted. Specifically, the similarity between samples is calculated using Euclidean algorithm, and weights are assigned to each neighbor sample. Using these neighbor values, a weighted average is calculated based on the weights to imput missing values. Each input feature includes n data samples. The input data features are encoded and transformed. Blood pressure data is a continuous numerical value. Depression level is a scale score, which is also a continuous numerical value. This invention standardizes the blood pressure and depression level data. Chemotherapy drug dosage and chemotherapy cycle are numerical data, and normalization is performed to compress the values ​​to [0, ...]. Within the range of 1], to make it have a similar scale in the chemotherapy risk prediction model for lung cancer patients, the CIPN risk rate is a continuous variable and is directly used as the target output of the chemotherapy risk prediction model for lung cancer patients.

[0008] Preferably, olfactory parameters This invention controls the global exploration and local exploitation capabilities of agent individuals during the odor optimization algorithm search process. By improving olfactory parameters from the perspectives of environmental perception and nonlinear dynamic systems, it proposes an adaptive olfactory factor based on odor decay and environmental feedback. The core design idea is to treat the olfactory factor as the concentration of odor molecules, which gradually decays with time and spatial distance. The agent individual dynamically adjusts its behavior by perceiving the gradient information of the "odor field." Secondly, the odor optimization algorithm obtains clues from the fitness change rate and search space complexity feedback of the lung cancer patient chemotherapy risk prediction model problem, and corrects the dynamic update behavior of the olfactory factor in real time. The fitness function of the lung cancer patient chemotherapy risk prediction model problem... for: ; In the formula, This represents the number of samples in the training set. Let be the actual risk rate of the true CIPN for the i-th sample. Let be the predicted CIPN risk rate for the i-th sample. This represents the number of hidden layers and the learning rate factor corresponding to the agent individual.

[0009] Preferably, olfactory parameters The specific steps for improvement are as follows: S201, Simulates the diffusion of odor molecules from the odor source into the search space [lb, ub], odor concentration. With the number of iterations and distance Exponential decay, where odor molecules act as surrogate individuals, is modeled mathematically as follows: ; In the formula, This represents the current iteration number. This represents the initial odor concentration. The diffusion coefficient is related to the search space and reflects the speed of odor propagation. Let i be the position of the i-th agent in the t-th iteration. This represents the initial iteration position of the i-th agent individual; S202, Based on odor concentration Adjust olfactory parameters The mathematical model is: ; In the formula, Let be the olfactory parameter value in the t-th iteration. Odor concentration The gradient guides the movement direction of the agent. This is the time decay coefficient; S203. Design an environment feedback-driven adjustment mechanism, which measures the state of the search space by the convergence trend of search space complexity and fitness, and dynamically corrects the olfactory parameters. The mathematical model is as follows: ; In the formula, Let t be the olfactor factor value improved in the t-th iteration. Let V be the variance of the location distribution of the agent individuals in the t-th iteration. Let be the mean of the location distribution of agent individuals in the t-th iteration, and γ be a local minimum value of 0.001. For the scale of individual agents, Let be the fitness value of the position of the i-th agent in the t-th iteration. The fitness value of the position of the surrogate individual at the population mean position in the t-th iteration.

[0010] Preferably, the time decay coefficient in the model and spatial coefficient By controlling the propagation speed and attenuation process of odors—that is, the speed and process of changes in the agent's location—this dynamic adjustment allows the agent to adaptively adjust its behavior according to changes in the search environment. In the early stages of the search, odor molecules diffuse rapidly, bringing more information; however, as the target approaches, the odor attenuates more significantly, and the agent's behavior gradually focuses on a local area. It can flexibly adjust the intensity of olfactory perception according to changes in time and space, avoiding overexploration of already familiar areas during the search process. Simultaneously, This represents the distance between the current location and the odor source, making the agent's olfactory ability dependent on its relative distance to the odor source. It adaptively adjusts its search behavior based on environmental feedback, thus avoiding getting trapped in local optima and improving overall optimization efficiency. It also addresses the problem of unstable model prediction accuracy caused by the odor optimization algorithm in the process of optimizing the chemotherapy risk prediction model for lung cancer patients, thereby improving the accuracy of chemotherapy risk prediction for lung cancer patients.

[0011] Preferably, the odor optimization algorithm mainly searches through three behavioral modes: sniffing mode, tracking mode, and random mode. This invention proposes a mutation strategy, which includes: each agent individual retains a set of historical optimal paths, that is, records the local optimal solution of the agent individual in each iteration, where the local optimal solution is the local optimal agent individual position. When the fitness value of the agent individual changes little in multiple iterations, it is determined that it is trapped in a local optimum. When trapped in a local optimum, the agent individual uses the historical optimal path information to select a high-quality position for jumping, and combines historical information to make directional correction, so that it moves along the potential global optimum direction.

[0012] Preferably, the mathematical model of the agent's location is updated using a mutation strategy as follows: S41. Record the local optimal solution of the agent in each iteration, and construct a set of historical optimal paths. The mathematical model is as follows: ; In the formula, Let i be the set of historical best paths for the i-th agent. Let be the position of the local optimal solution recorded by the i-th agent in the k-th instance; S42. When the fitness of the agent individual changes... Less than the threshold And it continues for more than a certain number of iterations. When this happens, it is determined that the system is trapped in a local optimum: ; Among them, threshold Adjustments were made based on the predicted results, with the initial value set at 0.15. The value is 5; Let be the fitness value of the position of the i-th agent in the t-th iteration. This is the current optimal fitness value; S43. If it is determined that the path is trapped in a local optimum, the agent selects k historical optimal solutions from the set of historical optimal paths as reference positions for jumping; the mathematical model is: ; In the formula, Let be the position of the i-th agent after jumping. Let i be the position of the i-th agent in the t-th iteration. Let be the position of the local optimal solution for the i-th agent in the y-th record, where y = 1, 2, ..., k; Let be the mutation weight of the i-th agent individual. The mathematical model is as follows: .

[0013] Preferably, the mutation strategy fully utilizes past exploration results, reduces invalid searches, and improves search efficiency. Through jump optimization, the agent can quickly escape the predicament of being trapped in local optima. At the same time, by combining memory paths and real-time environmental feedback, it adaptively adjusts the search direction and jump behavior, enabling the agent to find a balance between global and local optimization. This avoids local optimum traps and allows for rapid adjustment of direction for global search. It solves the problem of low prediction accuracy caused by being trapped in local optima during the optimization of the odor optimization algorithm for predicting chemotherapy risks in lung cancer patients. The mutation strategy can quickly jump out of local optima and seek better solutions, even if the optimization process is trapped in local optima. Especially with large amounts of data, it can significantly improve prediction accuracy and computational efficiency.

[0014] Preferably, the input dataset of the WSAO-BP model for predicting the chemotherapy risk of lung cancer patients is divided into a training set and a prediction set in an 8:2 ratio. An improved odor optimization algorithm is used to optimize the number of hidden layers and the learning rate factor of the BP prediction model. During the optimization process, the input dataset of the WSAO-BP model serves as the training set. Specifically, when optimizing the number of hidden layers and the learning rate factor of the BP prediction model using the improved odor optimization algorithm, a mapping relationship needs to be established between the position of the surrogate individual in the improved odor optimization algorithm and the hidden layer number and learning rate factor value of the BP prediction model. This is achieved by encoding the number of hidden layers h and the learning rate factor q into a spatial vector, with the surrogate individual size N denoted as the number of spatial vectors. The position of the i-th surrogate individual is mapped to the i-th spatial vector. Since the spatial vector is two-dimensional, the surrogate individual position is also two-dimensional. The mathematical model is as follows: .

[0015] Preferably, the number of hidden layers and the learning rate factor of the BP prediction model are optimized using an improved odor optimization algorithm, and the specific steps are as follows: S301. Initialize the relevant parameters of the improved odor optimization algorithm, including: maximum number of iterations T, agent size N, upper bound ub and lower bound lb of agent location, and olfactory parameters. ; S302. A group of surrogate individuals is generated randomly, with each surrogate individual's position representing a candidate solution. The candidate solution includes the values ​​of the hidden layer number h and the learning rate factor q. The mathematical model is as follows: ; In the formula, Let i be the position of the i-th agent in the t-th iteration. Let j be the lower bound of the position of the proxy individual in the j-th dimension. Let j be the upper bound of the position of the proxy individual in the j-th dimension. A random number between 0 and 1; S303. Calculate the fitness value of each agent's position using the fitness function. The fitness values ​​are arranged in ascending order, and the position of the agent with the smallest fitness value in this iteration is retained. S304. In sniffing mode, the agent updates its position through Brownian motion, and the mathematical model is as follows: ; In the formula, Let i be the position of the i-th agent in the (t+1)-th iteration. To assign an incremental value to the agent during each iteration of the optimization process, i.e., increment by one with each iteration, Let be the update rate of the odor molecules in the (t+1)th iteration, i.e., the update rate of the surrogate individual. The mathematical model is as follows: , Let v be the update rate of odor molecules in the t-th iteration, and v be the rate update parameter. S305. In tracking mode, an improved olfactory factor value is introduced, and the agent updates its position by comparing the odor concentration at the current position with the odor concentration at the worst position. ; In the formula, and A random number between 0 and 1. Let t be the olfactor factor value improved in the t-th iteration. Let this be the optimal position of the agent in the t-th iteration. Let be the worst position of the agent in the t-th iteration; S306. In random mode, use a mutation strategy to update the position of the agent individual; S307. Calculate the fitness value of the current agent's position, select the agent with the smallest fitness as the current optimal solution, update the position of the optimal solution, and update the values ​​of the hidden layer number and learning rate factor. S308. Execute t=t+1 for the current iteration number. If the current iteration number is less than the maximum iteration number T, return to execute 304; otherwise, output the current optimal solution.

[0016] Compared to existing methods, the beneficial effects and innovations of the method proposed in this invention are as follows: This invention improves the odor optimization algorithm by proposing an adaptive olfactory factor based on odor attenuation and environmental feedback, thereby enhancing the search capability of the odor optimization algorithm. By introducing an adaptive adjustment mechanism, WSAO can correct its behavior in real time according to changes in fitness and the complexity of the search space, thus improving global search capability and avoiding the risk of getting trapped in local optima. This invention designs a mutation strategy that allows the agent individual to jump using historical optimal path information when trapped in a local optimum, helping the model escape the local optimum predicament and quickly move towards the global optimum. By utilizing WSAO to optimize the number of hidden layers and the learning rate factor of the BP neural network, this invention significantly improves the prediction accuracy of the chemotherapy risk prediction model for lung cancer patients, enabling early intervention in treatment and rehabilitation guidance, and improving the precision of treatment. This invention optimizes the BP neural network model through an improved odor optimization algorithm, which can efficiently predict the risk of peripheral neuropathy after chemotherapy in lung cancer patients while also possessing good scalability, adapting to more medical data and complex clinical application scenarios. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for assessing the chemotherapy risk of lung cancer patients according to the present invention.

[0018] Figure 2 This is a flowchart of the improved odor optimization algorithm for optimizing the number of hidden layers and the learning rate factor in a BP prediction model.

[0019] Figure 3 This is a comparison chart of the fitness values ​​of solutions during the optimization process of the improved odor optimization algorithm and the standard odor optimization algorithm.

[0020] Figure 4 This is a flowchart of the hidden layer optimization process for the BP prediction model of the chemotherapy risk prediction model.

[0021] Figure 5 This is a flowchart of the learning rate optimization process for the BP prediction model in the chemotherapy risk prediction model.

[0022] Figure 6 This is a comparison chart of prediction errors for the risk rate of peripheral neurotoxicity symptoms in the training set.

[0023] Figure 7This is a comparison chart of prediction errors for the risk rate of peripheral neurotoxicity symptoms in the test set.

[0024] Figure 8 This is a comparison chart of the predicted risk rates of peripheral neurotoxicity symptoms in the test set. Detailed Implementation

[0025] This invention improves the odor optimization algorithm by using an improved odor optimization algorithm (WSAO) to optimize the BP prediction model. By constructing a WSAO-BP-based chemotherapy risk prediction model for lung cancer patients, it predicts peripheral neuropathy symptoms (CIPN) after chemotherapy, thereby improving the accuracy of chemotherapy risk prediction and enabling earlier intervention and rehabilitation guidance. To make the objectives, technical solutions, and advantages of the embodiments of this invention clearer, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0026] Example 1, such as Figure 1 As shown, the present invention provides a method for assessing the risk of chemotherapy in lung cancer patients, with specific steps S1 to S4.

[0027] S1. Collect data on factors affecting peripheral neuropathy during lung cancer chemotherapy, use the data on these factors to construct an input dataset for a chemotherapy risk prediction model, and preprocess the input dataset.

[0028] Specifically, this invention selects four types of data on influencing factors of peripheral neuropathy caused by chemotherapy for lung cancer in three aspects: blood pressure, depression level, dosage of chemotherapy drugs, and chemotherapy cycle. The data on these influencing factors are denoted as input features, i.e., the number of input features is set to 4. The target output is the risk rate of peripheral neuropathy symptoms (CIPN). The dataset of influencing factors of peripheral neuropathy caused by chemotherapy for lung cancer is preprocessed, including data sorting and data feature encoding and conversion.

[0029] Specifically, during the data processing, the input dataset is checked for missing values. For each input feature, a weighted average imputation strategy based on its similar samples is adopted. Specifically, the similarity between samples is calculated using Euclidean algorithm, and weights are assigned to each neighboring sample. Using these neighboring values, a weighted average is calculated based on the weights to impute missing values. Each input feature includes n data samples. In the experimental implementation of this invention, 150 sets of data on factors influencing peripheral neuropathy caused by chemotherapy for lung cancer were collected; denoted as n=150. The mathematical model is as follows: S11. Calculate the similarity between samples; ; In the formula, and Let c and u be the values ​​of the i-th feature; S12. Select G nearest neighbor samples for weighting, assigning a weight to each neighbor sample. The weight is proportional to the similarity. The weight calculation formula is as follows: ; In the formula, Let G be the weight between the c-th sample and the u-th sample. In the experimental process of this invention, the nearest neighbor samples are selected as 5, denoted as G=5. This is a hyperparameter, set to 1; S13. For each missing feature, calculate the weighted average as the imputation value: ; In the formula, The imputation value is the value used to fill in the missing values ​​in the sample. S14. Perform similar processing on all missing values ​​in sequence. For each missing feature, calculate its similar samples, calculate the weighted average and impute them.

[0030] Furthermore, the missing values ​​are completed in Matlab, as shown in the following code: function filled_data = fill_missing_values_with_weighted_average(data) n=150; M=4; [n, M] = size(data); G = 5; sigma = 1; filled_data = data; for c = 1:n for u = 1:M if isnan(filled_data(c, u)) distances = zeros(n, 1); for t = 1:n if ~isnan(filled_data(t, u)) distances(t) = sqrt(sum((data(c, :) - data(t, :)).^2)); else distances(t) = Inf; end end weights = exp(-distances.^2 / (2 * sigma^2)); [~, idx] = sort(weights, 'descend'); idx = idx(1:G); weighted_sum = sum(weights(idx) .* filled_data(idx, u)); total_weight = sum(weights(idx)); filled_data(c, u) = weighted_sum / total_weight; end end end end.

[0031] Furthermore, the input data features are encoded and transformed. Blood pressure data is a continuous numerical value, and depression level is a scale score, which is also a continuous numerical value. This invention standardizes the blood pressure data and depression level data. The dosage and chemotherapy cycle of chemotherapy drugs are numerical data, which are normalized to compress the values ​​to the range of [0, 1], so that they have a similar scale in the chemotherapy risk prediction model for lung cancer patients. CIPN risk rate is a continuous variable and is directly used as the target output of the chemotherapy risk prediction model for lung cancer patients.

[0032] S2. Improvements to the odor optimization algorithm, including: S21. By improving olfactory parameters from the perspective of environmental perception and nonlinear dynamic systems, an adaptive olfactory factor based on odor attenuation and environmental feedback is proposed. S22. A mutation strategy is proposed. When trapped in a local optimum, the agent uses historical optimal path information to select a high-quality position for jumping. The agent then uses historical information to correct its direction, moving it along a potential global optimum direction and updating its position.

[0033] Specifically, olfactory factors are considered as the concentration of odor molecules, which gradually decreases with time and spatial distance. The agent dynamically adjusts its behavior by perceiving the gradient information of the "odor field." Secondly, the odor optimization algorithm obtains clues from the fitness rate and search space complexity feedback of the lung cancer patient chemotherapy risk prediction model problem, and corrects the dynamic update behavior of olfactory factors in real time. The fitness function of the lung cancer patient chemotherapy risk prediction model problem is: ; In the formula, The number of samples in the training set is set to 150. Let be the actual risk rate of the true CIPN for the i-th sample. Let be the predicted CIPN risk rate for the i-th sample. This represents the number of hidden layers and the learning rate factor corresponding to the agent individual.

[0034] Specifically, olfactory parameters The specific steps for improvement are as follows: S201, Simulates the diffusion of odor molecules from the odor source into the search space [lb, ub], odor concentration. With the number of iterations and distance Exponential decay, where odor molecules act as surrogate individuals, is modeled mathematically as follows: ; In the formula, This represents the current iteration number. The initial odor concentration was set to 0.1 during the experiments conducted in this invention. The diffusion coefficient, which is related to the search space and reflects the speed of odor propagation, was set to 0.5 during the experimental implementation of this invention. Let i be the position of the i-th agent in the t-th iteration. This represents the initial iteration position of the i-th agent individual; S202, Based on odor concentration Adjust olfactory parameters The mathematical model is: ; In the formula, Let be the olfactory parameter value in the t-th iteration. Odor concentration The gradient guides the movement direction of the agent. The time decay coefficient was set to 0.3 during the experimental implementation of this invention. S203. Design an environment feedback-driven adjustment mechanism, which measures the state of the search space by the convergence trend of search space complexity and fitness, and dynamically corrects the olfactory parameters. The mathematical model is as follows: ; In the formula, Let t be the olfactor factor value improved in the t-th iteration. Let V be the variance of the location distribution of the agent individuals in the t-th iteration. Let be the mean of the location distribution of agent individuals in the t-th iteration, and γ be a local minimum value of 0.001. For the scale of individual agents, Let be the fitness value of the position of the i-th agent in the t-th iteration. The fitness value of the position of the surrogate individual at the population mean position in the t-th iteration.

[0035] Specifically, the mathematical model for updating the location of the agent individual using the mutation strategy is as follows: S41. Record the local optimal solution of the agent in each iteration, and construct a set of historical optimal paths. The mathematical model is as follows: ; In the formula, Let i be the set of historical best paths for the i-th agent. Let k be the location of the local optimal solution recorded by the i-th agent individual in the k-th instance. In the experimental implementation of this invention, k is set to 5. S42. When the fitness of the agent individual changes... Less than the threshold And it continues for more than a certain number of iterations. When this happens, it is determined that the system is trapped in a local optimum: ; Among them, threshold Adjustments were made based on the predicted results, with the initial value set at 0.15. The value is 5; Let be the fitness value of the position of the i-th agent in the t-th iteration. This is the current optimal fitness value; S43. If it is determined that the path is trapped in a local optimum, the agent selects k historical optimal solutions from the set of historical optimal paths as reference positions for jumping; the mathematical model is: ; In the formula, Let be the position of the i-th agent after jumping. Let i be the position of the i-th agent in the t-th iteration. Let be the position of the local optimal solution for the i-th agent in the y-th record, where y = 1, 2, ..., k; Let be the mutation weight of the i-th agent individual. The mathematical model is as follows: .

[0036] S3. Construct a chemotherapy risk prediction model for lung cancer patients, WSAO-BP. Organize the input dataset into input features and target output, and divide it into a training set and a prediction set. The WSAO-BP prediction model method is as follows: optimize the number of hidden layers and learning rate factor of the BP prediction model using an improved odor optimization algorithm, and train the optimized BP prediction model using the training set to obtain the WSAO-BP prediction model.

[0037] Specifically, the MATLAB code for the BP prediction model is as follows: Establish a network; net = newff(p_train, t_train, best_hd); Set training parameters; net.trainParam.epochs = 1000; % Number of training iterations; net.trainParam.goal = 0.005; % Target error; net.trainParam.lr = best_q; % Learning rate; The initial value of the learning rate code value best_q is 0.1.

[0038] Specifically, the input dataset of the WSAO-BP model for predicting chemotherapy risks in lung cancer patients is divided into training and prediction sets in an 8:2 ratio. An improved odor optimization algorithm is used to optimize the number of hidden layers and the learning rate factor of the BP prediction model. During the optimization process, the input dataset of the WSAO-BP model serves as the training set. Specifically, when optimizing the number of hidden layers and the learning rate factor of the BP prediction model using the improved odor optimization algorithm, a mapping relationship needs to be established between the position of the surrogate individual in the improved odor optimization algorithm and the hidden layer number and learning rate factor value of the BP prediction model. This is achieved by encoding the number of hidden layers h and the learning rate factor q into a spatial vector, with the surrogate individual size N representing the number of spatial vectors. The position of the i-th surrogate individual is mapped to the i-th spatial vector. Since the spatial vector is two-dimensional, the surrogate individual position is also two-dimensional. The mathematical model is as follows: .

[0039] Specifically, the improved odor optimization algorithm is used to optimize the number of hidden layers and the learning rate factor of the BP prediction model. The specific steps are as follows: S301. Initialize the relevant parameters of the improved odor optimization algorithm, including: maximum number of iterations T, agent size N, upper bound ub and lower bound lb of agent location, and olfactory parameters. ; S302. A group of surrogate individuals is generated randomly, with each surrogate individual's position representing a candidate solution. The candidate solution includes the values ​​of the hidden layer number h and the learning rate factor q. The mathematical model is as follows: ; In the formula, Let i be the position of the i-th agent in the t-th iteration. Let j be the lower bound of the position of the proxy individual in the j-th dimension. Let j be the upper bound of the position of the proxy individual in the j-th dimension. A random number between 0 and 1; S303. Calculate the fitness value of each agent's position using the fitness function. The fitness values ​​are arranged in ascending order, and the position of the agent with the smallest fitness value in this iteration is retained. S304. In sniffing mode, the agent updates its position through Brownian motion, and the mathematical model is as follows: ; In the formula, Let i be the position of the i-th agent in the (t+1)-th iteration. To assign an incremental value to the agent during each iteration of the optimization process, i.e., increment by one with each iteration, Let be the update rate of the odor molecules in the (t+1)th iteration, i.e., the update rate of the surrogate individual. The mathematical model is as follows: , Let v be the update rate of odor molecules in the t-th iteration, and v be the rate update parameter. S305. In tracking mode, an improved olfactory factor value is introduced, and the agent updates its position by comparing the odor concentration at the current position with the odor concentration at the worst position. ; In the formula, and A random number between 0 and 1. Let t be the olfactor factor value improved in the t-th iteration. Let this be the optimal position of the agent in the t-th iteration. Let be the worst position of the agent in the t-th iteration; S306. In random mode, use a mutation strategy to update the position of the agent individual; S307. Calculate the fitness value of the current agent's position, select the agent with the smallest fitness as the current optimal solution, update the position of the optimal solution, and update the values ​​of the hidden layer number and learning rate factor. S308. Execute t=t+1 for the current iteration number. If the current iteration number is less than the maximum iteration number T, return to execute 304; otherwise, output the current optimal solution.

[0040] S4. Input the test set data into the WSAO-BP model for predicting the risk of chemotherapy in lung cancer patients, and output the risk rate of chemotherapy-induced peripheral neuropathy symptoms (CIPN) in lung cancer patients.

[0041] Furthermore, during the experimental implementation of this invention, the maximum number of iterations T=60, the size of the surrogate individuals N=30, the upper bound of the surrogate individual position ub=[20,20x10^-6] and the lower bound lb=[0,1x10^-6], and the olfactory parameter... The initial value is 0.5, and the training set is trained 1000 times. The optimal solution output after reaching the maximum number of iterations, i.e., the optimal hidden layer number h and learning rate factor q, is used to reconstruct the WSAO-BP chemotherapy risk prediction model for lung cancer patients. The improved odor optimization algorithm is programmed in MATLAB, and the experimental design of this invention is completed. The prediction results of peripheral neurotoxicity symptoms after chemotherapy in lung cancer patients are output. The hidden layer number and learning rate values ​​of the BP prediction model of the chemotherapy risk prediction model optimized by the improved odor optimization algorithm and the standard odor optimization algorithm are as follows: Figure 4 and Figure 5 As shown, the number of hidden layers is an integer. The improved odor optimization algorithm has 11 hidden layers and a learning rate of 8.2 x 10^-6, while the standard odor optimization algorithm has 8 hidden layers and a learning rate of 4.4 x 10^-6.

[0042] Furthermore, such as Figure 3 As shown, the fitness value of the improved odor optimization algorithm decreases rapidly in the early stages of optimization and gradually stabilizes after about 20 iterations, eventually converging to a relatively low fitness value of about 7.5. The fitness value of the standard odor optimization algorithm decreases more slowly, and its final fitness value is significantly higher than that of the improved odor optimization algorithm. The improved odor optimization algorithm basically converges after about 20 iterations, indicating that its global search ability is stronger and it can quickly find a better solution. The standard odor optimization algorithm is relatively slow in terms of convergence speed, and it only begins to stabilize after 40 iterations. The final fitness value of the improved odor optimization algorithm is significantly lower than that of the standard algorithm, indicating that the improved odor optimization algorithm achieves better results in solving the problem of predicting the risk rate of peripheral neurotoxicity symptoms. The standard odor optimization algorithm, due to the lack of mutation strategy and adaptive olfactory factor, is prone to getting trapped in local optima, resulting in a higher final fitness value than the improved odor optimization algorithm.

[0043] Furthermore, such as Figure 6As shown, the training results of the training set indicate that the prediction error of the risk rate of peripheral neurotoxicity symptoms fluctuates greatly, showing high instability. On multiple samples, the prediction error deviates significantly from the true value, indicating that the existing method has a weak fitting ability to the training set samples and is easily affected by data characteristics and noise. The prediction error distribution of the method of the present invention is more stable, with a significantly lower fluctuation range than the existing methods. The prediction results are closer to the true risk rate, showing a higher fitting ability to the training set data.

[0044] Furthermore, such as Figure 7 As shown, the two methods, namely the standard odor optimization algorithm-optimized BP prediction model SAO-BP and the method of this invention, namely the chemotherapy risk prediction model WSAO-BP for lung cancer patients, are compared in predicting the risk rate of chemotherapy-induced peripheral neuropathy (CIPN) in lung cancer patients. Based on the trained prediction models, the existing methods show large fluctuations in prediction error, exhibiting high instability and overfitting or underfitting, resulting in inaccurate predictions. The method of this invention has a more stable prediction error, with a much smaller fluctuation range than the existing methods, and the error is less and close to 0. This indicates that the method of this invention can fit the peripheral neuropathy risk rate prediction test set data better.

[0045] Furthermore, the prediction results of the risk rate of peripheral neurotoxicity symptoms on the test set were compared between existing methods and the method of the present invention, such as... Figure 8 As shown, the predicted values ​​of existing methods fluctuate greatly, and many predicted values ​​deviate significantly from the true risk values, resulting in large errors. On some samples, the predicted values ​​of existing methods are close to or exceed the fluctuation range of the true values, demonstrating low prediction accuracy. The predicted values ​​of the method of this invention are closer to the true risk values, and the error is significantly smaller. The prediction curve matches the fluctuation of the true risk values.

Claims

1. A method for assessing the risk of chemotherapy in lung cancer patients, characterized in that, The specific steps for predicting peripheral neurotoxicity symptoms after chemotherapy in lung cancer patients are as follows: S1. Collect data on factors affecting peripheral neuropathy during lung cancer chemotherapy, use the data on these factors to construct an input dataset for a chemotherapy risk prediction model, and preprocess the input dataset. S2. Improvements to the odor optimization algorithm, including: S21. By improving olfactory parameters from the perspective of environmental perception and nonlinear dynamic systems, an adaptive olfactory factor based on odor attenuation and environmental feedback is proposed. S22. A mutation strategy is proposed. When trapped in a local optimum, the agent uses historical optimal path information to select a high-quality position for jumping. The direction is corrected by combining historical information so that it moves along the potential global optimum direction and the position of the agent is updated. S3. Construct a chemotherapy risk prediction model for lung cancer patients, WSAO-BP. Organize the input dataset into input features and target output, and divide it into a training set and a prediction set. The WSAO-BP prediction model method is as follows: optimize the number of hidden layers and learning rate factor of the BP prediction model using an improved odor optimization algorithm, and train the optimized BP prediction model using the training set to obtain the WSAO-BP prediction model. S4. Input the test set data into the WSAO-BP model for predicting the risk of chemotherapy in lung cancer patients, and output the risk rate of peripheral neurotoxicity symptoms caused by chemotherapy in lung cancer patients.

2. The method for assessing chemotherapy risk in lung cancer patients according to claim 1, characterized in that, The peripheral neuropathy influencing factor data includes: blood pressure, depression level, chemotherapy drug dosage, and chemotherapy cycle. These influencing factor data are denoted as input features, and the target output is the risk rate of peripheral neuropathy symptoms. The dataset of influencing factors for peripheral neuropathy caused by lung cancer chemotherapy is preprocessed. During preprocessing, for each input feature, a weighted average imputation strategy based on its similar samples is adopted. Specifically, the similarity between samples is calculated using Euclidean algorithm, and weights are assigned to each neighbor sample. Using these neighbor values, a weighted average is calculated based on the weights to imput missing values. Each input feature includes n data samples. The input data features are encoded and transformed. Blood pressure data is a continuous numerical value, and depression level is a scale score, which is also a continuous numerical value. This invention standardizes the blood pressure and depression level data. Chemotherapy drug dosage and chemotherapy cycle are numerical data, which are normalized to compress the values ​​to the range [0, 1], making them have a similar scale in the lung cancer patient chemotherapy risk prediction model. The CIPN risk rate is a continuous variable and is directly used as the target output of the lung cancer patient chemotherapy risk prediction model.

3. The method for assessing chemotherapy risk in lung cancer patients according to claim 2, characterized in that, The specific steps for improving olfactory parameters from the perspective of environmental perception and nonlinear dynamic systems are as follows: S201, Simulates the diffusion of odor molecules from the odor source into the search space [lb, ub], odor concentration. With the number of iterations and distance Exponential decay, in which odor molecules act as surrogate entities; S202, Based on odor concentration Adjust olfactory parameters The mathematical model is: ； In the formula, Let be the olfactory parameter value in the t-th iteration. Odor concentration The gradient guides the movement direction of the agent. This is the time decay coefficient; S203. Design an environment feedback-driven adjustment mechanism, which measures the state of the search space by the convergence trend of search space complexity and fitness, and dynamically corrects the olfactory parameters. The mathematical model is as follows: ； In the formula, Let be the olfactor factor value improved in the t-th iteration. Let V be the variance of the location distribution of the agent individuals in the t-th iteration. Let be the mean of the location distribution of agent individuals in the t-th iteration, and γ be a local minimum value of 0.

001. For the scale of individual agents, Let be the fitness value of the position of the i-th agent in the t-th iteration. The fitness value of the position of the surrogate individual at the population mean position in the t-th iteration.

4. The method for assessing chemotherapy risk in lung cancer patients according to claim 3, characterized in that, The specific method for updating the location of the agent individual using the mutation strategy is as follows: S41. Record the local optimal solution of the agent in each iteration, and construct a set of historical optimal paths. The mathematical model is as follows: ； In the formula, Let i be the set of historical best paths for the i-th agent. Let be the position of the local optimal solution recorded by the i-th agent in the k-th instance; S42. When the fitness of the agent individual changes... Less than the threshold And it continues for more than a certain number of iterations. When this happens, it is determined that the system is trapped in a local optimum: S43. If it is determined that the path is trapped in a local optimum, the agent selects k historical optimal solutions from the set of historical optimal paths as reference positions for jumping; the mathematical model is: ； In the formula, Let i be the position of the i-th agent after jumping. Let i be the position of the i-th agent in the t-th iteration. Let be the position of the local optimal solution for the i-th agent in the y-th record, where y = 1, 2, ..., k; Let be the mutation weight of the i-th agent individual. The mathematical model is as follows: .

5. The method for assessing chemotherapy risk in lung cancer patients according to claim 4, characterized in that, The specific steps for optimizing the number of hidden layers and learning rate factor of the BP prediction model using the improved odor optimization algorithm are as follows: S301. Initialize the relevant parameters of the improved odor optimization algorithm, including: maximum number of iterations T, agent size N, upper bound ub and lower bound lb of agent location, and olfactory parameters. ; S302. A group of agent individuals is generated randomly, and each agent individual's position represents a candidate solution. The candidate solution includes the value of the number of hidden layers h and the learning rate factor q. S303. Calculate the fitness value of each agent's position using the fitness function. The fitness values ​​are arranged in ascending order, and the position of the agent with the smallest fitness value in this iteration is retained. S304. In sniffing mode, the agent updates its position through Brownian motion; S305. In tracking mode, an improved olfactory factor value is introduced, and the agent updates its position by comparing the odor concentration at the current position with the odor concentration at the worst position. ； In the formula, and A random number between 0 and 1. Let be the olfactor factor value improved in the t-th iteration. The optimal position for the agent in the t-th iteration. Let be the worst position of the agent in the t-th iteration; S306. In random mode, use a mutation strategy to update the position of the agent individual; S307. Calculate the fitness value of the current agent's position, select the agent with the smallest fitness as the current optimal solution, update the position of the optimal solution, and update the values ​​of the hidden layer number and learning rate factor. S308. Execute t=t+1 for the current iteration number. If the current iteration number is less than the maximum iteration number T, return to execute 304; otherwise, output the current optimal solution.

Images