Feature extraction method, system and equipment for medical clinical data and medium

Abstract

The invention discloses a medical clinical data feature extraction method, system and device and a medium, and relates to the technical field of data processing. Comprising the steps of obtaining a medical clinical data set; a moss breeding mechanism of a moss optimization algorithm is replaced by a biogeography-based learning strategy, the implicit effect of the moss optimization algorithm is removed, and dual-mode propagation search is introduced to obtain a biological learning moss optimization algorithm; a clinical feature subset most related to sample classification is searched in the feature extraction process of the medical clinical data set through the biological learning moss optimization algorithm; and carrying out data classification on the clinical feature subset most related to sample classification by adopting a K-nearest neighbor classifier to obtain an optimal clinical feature subset of the medical clinical data. According to the method, the defect that the moss optimization algorithm is insufficient in local development capability of the optimal solution is overcome, and the algorithm convergence precision is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, system, device and medium for extracting features from medical clinical data. Background Art

[0002] Lung cancer is the leading cause of malignant tumor-related death worldwide, with an overall five-year survival rate of less than 20%. Based on histopathological characteristics, lung cancer is primarily divided into non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC), with NSCLC accounting for approximately 80%-85% of lung cancer patients. NSCLC includes three major subtypes: squamous cell carcinoma, adenocarcinoma, and large cell carcinoma. Current clinical treatment strategies include surgical resection, radiotherapy, chemotherapy, molecularly targeted therapy, and immunotherapy.

[0003] In recent years, breakthroughs in molecularly targeted drugs, immune checkpoint inhibitors (ICIs), and anti-angiogenic agents have significantly improved the prognosis of NSCLC patients. For patients with advanced NSCLC, the choice of first-line treatment is primarily based on driver gene mutation status and programmed death-ligand 1 (PD-L1) expression levels.

[0004] For patients without targetable driver gene mutations, ICIs, represented by PD-1 / PD-L1 inhibitors, have become a core treatment option. To identify the optimal biomarker combination that accurately predicts the efficacy and prognosis of immunochemotherapy in patients with non-small cell lung cancer (NSCLC), the Moss Growth Optimization (MGO) algorithm can be used. This technique operates as follows: clinical data from NSCLC patients is used as input, with each feature considered a dimension in the search space. The algorithm encodes each possible biomarker combination (i.e., feature subset) as a "moss patch," forming an initial population. The performance of each "moss patch" is then evaluated using a pre-defined fitness function. This function typically combines the classification accuracy (or AUC) of the prediction model constructed using the current feature subset with a penalty for the number of selected features, aiming to achieve a balance between high accuracy and model simplicity. During the iterative optimization process, the moss optimization algorithm simulates the natural growth and reproduction strategies of mosses, primarily through mechanisms such as spore diffusion, conducting large-scale global exploration in an effort to rapidly locate "suitable zones" with high predictive potential within the vast space of feature combinations. As the iterations proceed, the entire population migrates and concentrates toward the region containing the current optimal solution based on fitness, gradually converging. This process continues until a preset termination condition is reached, ultimately outputting the feature subset corresponding to the "moss patch" with the highest fitness, which serves as the biological basis for constructing the final predictive model.

[0005] However, the existing moss optimization algorithm has insufficient ability to locally develop the optimal solution in the later stage of the search, and its convergence accuracy is limited, which can easily lead to the results deviating from the global optimal or the accuracy of the final solution being low. Summary of the Invention

[0006] Based on this, it is necessary to provide a feature extraction method, system, equipment and medium for medical clinical data to address the above technical problems.

[0007] An embodiment of the present invention provides a method for extracting features from medical clinical data, comprising: Acquire medical clinical data sets, including blood test data and clinical data sets; By replacing the moss reproduction mechanism of the moss optimization algorithm with a biogeographically based learning strategy, the cryptobiotic effect of the moss optimization algorithm was removed, and a two-mode propagation search was introduced to obtain the biological learning moss optimization algorithm. The biological learning moss optimization algorithm was used to find the clinical feature subset most relevant to sample classification during feature extraction from medical clinical datasets. The K-nearest neighbor classifier is used to classify the clinical feature subset that is most relevant to the sample classification, and the optimal clinical feature subset of medical clinical data is obtained; Among them, the biological learning moss optimization algorithm is used to find the clinical feature subset most relevant to sample classification during the feature extraction process of medical clinical datasets, specifically including: Initialize the moss population, where each moss individual represents the candidate medical clinical data and has an independent eigenvalue, thereby obtaining a eigenvalue set; perform feature selection on the eigenvalue set to obtain the optimal feature subset; A biogeography-based learning strategy was used to construct a set of example vectors corresponding to the optimal feature subset. The particles in the example vector set were sorted according to the moss individual's own historical optimal solution to obtain the ranking value of each particle. The in-migration rate and out-migration rate were assigned to the moss individual corresponding to the optimal feature subset based on the ranking value of each particle. A two-mode propagation search was performed based on the in-migration rate and out-migration rate to dynamically update the position of the moss individual to obtain the global optimal solution, which was the clinical feature subset most relevant to the sample classification.

[0008] Optionally, a medical clinical data set is obtained through electronic medical records, and the medical clinical data set specifically includes: demographic characteristics, disease characteristics, pre-treatment hematological indicators, immune characteristics, and efficacy outcome indicators; Efficacy outcome indicators include: efficacy response and survival endpoints; efficacy response, including: disease remission group and disease non-remission group; survival endpoints, including: long survival group and short survival group.

[0009] Optionally, feature selection is performed on the eigenvalue set to obtain an optimal feature subset, specifically including: Binary encoding is performed on each moss individual and the eigenvalue set is encoded through a conversion function, so as to perform feature selection according to the position of the eigenvalue of each moss individual in the eigenvalue set, thereby obtaining a plurality of feature subsets; For each feature subset, the weight of the feature subset in the feature extraction process is determined, the fitness value of the feature subset is obtained, and the feature subset with a fitness value greater than a set threshold is taken as the optimal feature subset.

[0010] Optionally, the eigenvalue set is encoded using a conversion function based on the following formula to obtain a binary encoding value of the initial moss population: ; ; in, For the The first The individual's binary coded values, For the The first The individual's eigenvalues, is a random number in the range [0, 1], To negate the sign, is the conversion function, is the initial moss population; The weight of the feature subset in the feature extraction process is determined based on the following formula to obtain the fitness value of the feature subset: ; ; in, For the The fitness value of moss individuals, is the classification error rate, is the length of the feature subset, is the characteristic total length, is the weight of the classification error rate, is the weight for selecting the number of feature subsets, is the number of correct classifications in the sample, is the number of misclassified samples.

[0011] Optionally, a biogeography-based learning strategy is employed to construct a set of example vectors corresponding to the optimal feature subset: = ]; ; Based on the following formula, the particles in the example vector set are sorted according to the moss individual's own historical optimal solution to obtain the ranking value of each particle: = N- 1 , = N- 2 , … = 0 ; Based on the following formula, the in-migration rate and out-migration rate are assigned to the moss individuals corresponding to the optimal feature subset according to the ranking value of each particle: ; in, is the particle’s own historical optimal solution, sN is the particle subscript, is the immigration rate, is the emigration rate, N is the sort value of the particle, rank is the sorting function, is an example vector, is a set of example vectors, i is the index.

[0012] Optionally, a two-mode propagation search is performed based on the immigration rate and the emigration rate to dynamically update the position of the moss individuals, specifically including: The moss individual with the best self-historical optimal solution has the lowest immigration rate and tends to update its position toward itself; The moss individual with the worst historical optimal solution has the highest immigration rate and tends to update its position toward other moss individuals; The particle with the best historical optimal solution has the highest migration rate and is used as the update direction by other moss individuals; The particle with the worst historical optimal solution has the lowest migration rate and is not used as an update direction by other moss individuals.

[0013] An embodiment of the present invention provides a feature extraction system for medical clinical data, comprising: A data acquisition module is used to acquire medical clinical data sets, which include blood test data and clinical data sets; A processing module is used to replace the moss reproduction mechanism of the moss optimization algorithm with a biogeographic-based learning strategy, remove the cryptobiotic effect of the moss optimization algorithm, and introduce a two-mode propagation search to obtain a biological learning moss optimization algorithm. The biological learning moss optimization algorithm is used to find the clinical feature subset most relevant to sample classification during the feature extraction process of medical clinical datasets. The classification module is used to classify the clinical feature subset that is most relevant to the sample classification using a K-nearest neighbor classifier to obtain the optimal clinical feature subset of the medical clinical data; Among them, the processing module includes: feature selection module and optimization module; A feature selection module is used to initialize the moss population. Each moss individual represents the candidate medical clinical data and has an independent eigenvalue, thereby obtaining a set of eigenvalues. Feature selection is performed on the set of eigenvalues ​​to obtain the optimal feature subset. The optimization module is used to construct a set of example vectors corresponding to the optimal feature subset using a biogeographical learning strategy. The particles in the example vector set are sorted according to the moss individual's own historical optimal solution to obtain the ranking value of each particle. The in-migration rate and out-migration rate are assigned to the moss individual corresponding to the optimal feature subset based on the ranking value of each particle. A two-mode propagation search is performed based on the in-migration rate and out-migration rate to dynamically update the position of the moss individual to obtain the global optimal solution, which is the clinical feature subset most relevant to the sample classification.

[0014] An embodiment of the present invention provides a computer device including a memory and a processor. The memory stores a computer program, and the processor implements the steps of a feature extraction method for medical clinical data when executing the computer program.

[0015] An embodiment of the present invention provides a storage medium storing a computer program. When the computer program is executed by a processor, the computer program implements the steps of a method for extracting features from medical clinical data.

[0016] The above-mentioned method, system, device, and medium for extracting features from medical clinical data provided by the embodiments of the present invention have the following beneficial effects compared with the prior art: The present invention enhances the diversity and flexibility of the moss optimization algorithm in the search process through a biogeographic-based learning strategy. The use of dual-mode propagation search in the later stage of the search can more carefully explore the solution space, screen and extract the clinical features most relevant to sample classification, and can make up for the defect of the existing moss optimization algorithm in its insufficient ability to locally develop the optimal solution, effectively improving the algorithm convergence accuracy.

[0017] In addition, the present invention uses the K-nearest neighbor classifier to classify the clinical feature subset that is most relevant to the sample classification, further avoiding the results from deviating from the global optimum, thereby improving the accuracy of the optimal clinical feature subset and providing a more efficient and accurate method for the analysis and processing of medical clinical data. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a method for extracting features from medical clinical data provided in one embodiment; Figure 2 A function result diagram of different parameters BLMGO of a feature extraction method for medical clinical data provided in one embodiment; Figure 3 This is a graph showing the number of feature selection results of bBLMGO, a medical clinical data feature extraction method provided in one embodiment, on a dataset of 241 patients (binary classification) with laboratory data of PFS less than 12 months and greater than 12 months; Figure 4 This is a graph showing the number of feature selection results of bBLMGO, a medical clinical data feature extraction method provided in one embodiment, on a dataset of 289 patients (binary classification) with efficacy PD, SD, PR, and CR test data; Figure 5 A flowchart of a biological learning moss optimization algorithm for a feature extraction method of medical clinical data provided in one embodiment. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0020] Immune checkpoint inhibitors (ICIs), such as those targeting programmed death 1 (PD-1) / programmed death ligand 1 (PD-L1) or cytotoxic T lymphocyte-associated antigen 4 (CTLA-4), can induce durable responses in subsets of patients with advanced cancer. Combining ICIs with platinum-based chemotherapy has demonstrated a clear survival benefit in multiple randomized controlled trials (RCTs), providing a new first-line treatment strategy for metastatic NSCLC. Notably, this survival benefit was observed across patients with varying PD-L1 expression levels, with an overall manageable safety profile.

[0021] However, it should be noted that while ICI combined with chemotherapy significantly improves objective response rate and median survival, this treatment only provides long-term benefits to a small subset of patients, and drug resistance and disease progression remain major clinical challenges. The vast majority of patients do not experience a response or experience disease progression during treatment. Furthermore, treatment with immune checkpoint inhibitors can lead to immune-related adverse events.

[0022] The significant heterogeneity of immunotherapy responses highlights the importance of developing predictive biomarkers, which can help screen potential beneficiaries and achieve precision treatment. The current biomarker system can be divided into four categories: (1) tumor-intrinsic markers: including PD-L1 expression, tumor mutational burden (TMB) and specific driver gene mutations; (2) tumor microenvironment markers: covering the composition of tumor-infiltrating immune cells (TAICs) and T cell receptor (TCR) diversity; (3) liquid biopsy indicators: such as circulating tumor DNA (ctDNA) and peripheral blood immune cell subsets; (4) host-related factors: involving human leukocyte antigen class I (HLA-I) polymorphisms, clinical characteristics and metabolomics characteristics.

[0023] Although the FDA has approved PD-L1 expression levels and TMB as predictive markers for ICI efficacy, their clinical application still has significant limitations and cannot fully reflect the complex interactions between tumors and the immune system. PD-L1 lacks standardized staining interpretation criteria, and expression varies significantly between primary and metastatic lesions. TMB requires sufficient tumor tissue and resources for DNA sequencing. More importantly, these indicators alone cannot accurately predict durable responses and survival.

[0024] In this context, peripheral blood biomarkers have attracted considerable attention due to their accessibility and dynamic monitoring advantages. Several indicators and ratios of the complete blood count (CBC) have been suggested as predictors of ICI efficacy and patient outcomes. Among blood markers, a low neutrophil-lymphocyte ratio (NLR <5), a low systemic immune inflammatory index (SII), and decreased C-reactive protein (CRP) are associated with a better prognosis, while a high lactate dehydrogenase (LDH) and elevated baseline CRP predict a poor outcome. For example, in a study of patients with advanced NSCLC, Diem et al. and Ren et al. reported that high pretreatment neutrophil-lymphocyte ratio (NLR) and platelet-lymphocyte ratio (PLR) were prognostic markers significantly associated with poor survival and lower response rates in patients receiving nivolumab monotherapy.

[0025] The current limitations of single biomarkers in predicting disease progression highlight the urgent need to develop multidimensional prognostic models. These integrated models can improve clinical decision-making accuracy, enabling physicians to identify patients with high ICI response rates and those at risk of rapid progression. Artificial intelligence (AI) technology offers a breakthrough path toward this goal. AI / ML models can accurately predict immunotherapy response, progression-free survival, and overall survival in NSCLC patients, enabling tailored treatment decisions based on their unique molecular, clinical, and imaging profiles.

[0026] Defects include but are not limited to: (1) Limited convergence speed: In high-dimensional complex or multi-peak optimization problems, although the MGO group has strong global expansion capabilities, its convergence speed is slow, resulting in low efficiency in solving practical engineering problems.

[0027] (2) Limited convergence accuracy: The existing MGO is not capable of locally developing the optimal solution in the later stages of the search, which can easily lead to the results deviating from the global optimum or the accuracy of the final solution being low.

[0028] (3) The robustness of escaping local extremes needs to be improved: Although a dormant mechanism is set up, in a more complex search space or one with significant dynamic changes, individuals may still fall into a new local optimum, making it difficult to ensure the acquisition of the global optimum.

[0029] The embodiment of the present invention provides a method for extracting features from medical clinical data. Figure 1 The method includes: Obtain medical clinical data sets, which include blood test data and clinical data sets.

[0030] By replacing the moss reproduction mechanism of the moss optimization algorithm with a biogeographically based learning strategy, removing the cryptobiotic effect of the moss optimization algorithm and introducing a bimodal propagation search, the resulting biological learning moss optimization algorithm was developed. This algorithm was used to find the subset of clinical features most relevant to sample classification during feature extraction in medical clinical datasets.

[0031] The K-nearest neighbor classifier is used to classify the clinical feature subset that is most relevant to the sample classification, and the optimal clinical feature subset of medical clinical data is obtained.

[0032] Among them, the biological learning moss optimization algorithm is used to find the clinical feature subset most relevant to sample classification in the feature extraction process of medical clinical data sets, specifically including: initializing the moss population, each moss individual represents the candidate medical clinical data and has an independent eigenvalue, and obtaining a eigenvalue set; feature selection is performed on the eigenvalue set to obtain the optimal feature subset.

[0033] A biogeographically informed learning strategy was used to construct a set of example vectors corresponding to the optimal feature subset. The particles in the example vector set were sorted according to the moss individual's historical optimal solution, yielding a ranking value for each particle. Based on the ranking value of each particle, in-migration and out-migration rates were assigned to the moss individuals corresponding to the optimal feature subset. A two-mode propagation search was performed based on the in-migration and out-migration rates to dynamically update the positions of the moss individuals and obtain the global optimal solution, which was the clinical feature subset most relevant to the sample classification.

[0034] According to another preferred embodiment of the present invention, for the clinical data of patients with non-small cell lung cancer, the present invention adopts the moss optimization method improved by biological learning strategy to extract features. These feature extraction methods can capture key features from clinical data, providing important features for subsequent analysis and classification. When processing routine blood tests and clinical data of patients with non-small cell lung cancer, challenges such as limited sample size, high feature dimension, and correlation between data are often encountered. In order to overcome these problems, the reliability and stability of their analysis results are verified by adopting a 10-fold cross-validation method. Figure 5 As shown, the dataset is divided into 10 equal parts, with one used as the test set and the other nine as the training set. This cycle repeats 10 times, selecting different training and test sets each time. The average of all results is then used as the final performance evaluation metric. This ensures consistent and consistent performance during each training and test phase, reducing errors caused by improper data partitioning and improving the accuracy and reliability of model evaluation.

[0035] Step S1: Extract a training sample set based on the clinical data set of non-small cell lung cancer patients, and set the initial training sample population ,in, is the number of training sample individuals, is the number of dimensions for each training sample, Indicates in The population obtained in the iterations is Indicates The next iteration The jth eigenvalue of an individual, Indicates the current number of iterations, ranging from [0, 1000].

[0036] Step S2: designing a classifier implementation method and optimization technology based on K-Nearest Neighbor (KNN) according to the training sample set, and performing classification.

[0037] The basic idea of ​​KNN is that it is a simple, parameter-free and efficient learning method. It has shown excellent results in problems such as function classification and approximation, with high completion rate and classification accuracy. In recent studies, KNN has also shown excellent training speed and classification accuracy, so K-nearest neighbor (KNN) was used as a classifier for experimental evaluation. KNN is an instance-based learning model that predicts the category of a new sample by performing a majority vote on the k nearest neighbor categories of the test sample. When making classification decisions, the category of the new sample is determined based on the minimum distance (based on a similarity measure) between the new sample and each sample in the training set. The most widely used similarity measure is the Euclidean distance. For two D-dimensional points and , the calculation process of its Euclidean distance is as follows: ;

[0038] in, is a D-dimensional point and distance. Represents D-dimensional points and The coordinate value of .

[0039] Specifically include: Step S1: Collect blood test data and clinical data sets of patients with non-small cell lung cancer and extract them through electronic medical records.

[0040] Step S2: Divide it into a training set and a test set, initialize the population, perform binary encoding on each moss individual in the current population, and use a conversion function to encode the features.

[0041] Step S3: Calculate the fitness value of the current population using the moss optimization algorithm.

[0042] Step S4: Improve the moss reproduction mechanism, remove the cryptobiotic (dormant) effect, introduce a dual-mode propagation search, and update the individual positions of the population.

[0043] Step S5: testing the relevant parameters of the biological learning strategy, selecting the optimal biological learning moss optimization algorithm, and dynamically adjusting the moss individuals.

[0044] Step S6: Optimize the moss algorithm using the optimal biological learning strategy and build a prediction model with the KNN classifier.

[0045] Step S7: Circulate and update to the maximum number of iterations, and output the optimal feature subsets of routine blood data and clinical data.

[0046] The implementation is as follows: Step S1: Collect a dataset on the efficacy of immunotherapy for patients with non-small cell lung cancer. A preliminary screening of 1,799 patients with non-small cell lung cancer (NSCLC) admitted through a hospital's electronic medical record system until July 2024 was conducted. After strict inclusion and exclusion criteria, 289 patients with locally advanced or metastatic NSCLC who received first-line PD-L1 immune checkpoint inhibitors (ICIs) combined with platinum-based chemotherapy were ultimately included. Eligibility criteria included age ≥18 years, histologically / cytologically confirmed diagnosis, baseline hematological testing within 30 days before treatment, and at least one RECIST 1.1 efficacy assessment. Exclusion criteria included prior treatment with targeted therapy or radiotherapy, the exclusion of patients receiving other anti-tumor regimens during treatment (e.g., surgery, radiotherapy), missing key clinical data (e.g., staging, efficacy records), or loss to follow-up where complete follow-up data could not be obtained.

[0047] Electronic medical records were used to extract the following: ① demographic characteristics (sex, age, BMI, smoking history, and comorbidities); ② disease characteristics (TNM stage, pathological type, and PD-L1 TPS%); ③ hematological indicators 30 days before treatment (including complete blood count, liver and kidney function, inflammatory markers, coagulation function, and tumor markers); ④ immune characteristics (lymphocyte subsets, immunoglobulins, and complement); and ⑤ efficacy outcome indicators. The study established dual outcome indicators: efficacy response was divided into disease remission group (complete remission [CR] + partial remission [PR]) and disease non-remission group (stable disease [SD] + progressive disease [PD]) based on the RECIST 1.1 criteria; survival endpoint was divided into long / short survival group (12 months) based on progression-free survival (PFS, from the start of treatment to progression / death).

[0048] Continuous variables that conformed to normal distribution were analyzed using the Kolmogorov-Smirnov test and expressed as mean ± standard deviation (X ± S). Inter-group comparisons were performed using the u test. Non-normally distributed continuous variables were analyzed using nonparametric tests, described as median and interquartile range (IQR). Count data were expressed as percentages (%). Inter-group comparisons were performed using the chi-square test or Fisher's exact test. Statistical analysis results are shown in Tables 1 and 2. Assume that the population initialized with training samples ,in, , is the number of training sample individuals, is the number of dimensions of the training samples, For the The population obtained in the iterations is Indicates The next iteration The individual's j eigenvalues, Indicates the current number of iterations, with a value range of [0, 1000].

[0049] Table 1 Baseline characteristics of patients with locally advanced / metastatic NSCLC receiving first-line immunotherapy combined with chemotherapy: comparison of disease response group (CR+PR) and non-response group (SD+PD)

[0050] Table 2 Baseline characteristics of patients with locally advanced / metastatic NSCLC receiving first-line immunotherapy combined with chemotherapy: comparison of PFS < 12 months and PFS > 12 months

[0051] Figure 3 This is a graph showing the number of feature selection results for bBLMGO, a medical clinical data feature extraction method provided in one embodiment, on a dataset of 241 patients (binary classification) with laboratory data of PFS less than 12 months and greater than 12 months.

[0052] Figure 4 This is a graph showing the number of feature selection results of bBLMGO, a medical clinical data feature extraction method provided in one embodiment, on a dataset of 289 patients (binary classification) with efficacy PD, SD, PR, and CR test data.

[0053] Step S2: Divide it into training set and test set, initialize the population, the population Each characteristic value of each individual in is simulated into a binary coded value through formula (1) and formula (2).

[0054] In step S2, the population Each characteristic value of each individual in is simulated into a binary coded value through formula (1) and formula (2): (1) (2) in, It is The first The individual's binary coded values, For the The first The individual's eigenvalues, represents a random number in the range [0, 1], To negate the sign, is the conversion function, For moss populations .

[0055] Step S3: Calculate the moss population using formula (3) and formula (4) The fitness value of: (3) (4) in, For the The fitness value of moss individuals, is the classification error rate, represents the length of the feature subset, is the characteristic total length, is the weight of the classification error rate, is the weight for selecting the number of feature subsets, is the number of correct classifications in the sample, is the number of misclassified samples.

[0056] The following description is made for step S2 and step S3: (1) The role and status of binary coded values.

[0057] Binary encoding is used to convert the natural feature space (i.e., continuous feature values) into a discrete representation of feature selection. Through the encoding in step S2, the position of each individual (i.e., moss individual) in the original feature space is represented as a binary string of 0 / 1, where each 0 or 1 corresponds to whether a feature is selected. This encoding method conveniently represents the feature subset selection problem and lays the foundation for subsequent evolutionary operations (such as crossover and mutation) and fitness assessment.

[0058] In short, binary coded values ​​are the basis of feature selection schemes (i.e., individual representations), which endow each individual with a clear feature subset representation.

[0059] (2) The role and status of fitness value.

[0060] The fitness value evaluates each individual given a given feature subset. The fitness function typically takes into account factors such as the classification accuracy and number of features in the selected feature subset, reflecting the quality of that subset and serving as a key indicator for driving population evolution. In step S3, the fitness value, combined with the classification error rate and feature subset length, serves as the basis for selection and update operations in the evolutionary algorithm.

[0061] In short, the fitness value is a performance evaluation of binary coding (feature selection) and is the goal and driving force of algorithm evolutionary optimization.

[0062] (3) Logical connection between the two A. Each moss individual is first mapped to a binary coded value by its feature value through formula (1) and formula (2), and the individual representation (i.e., a feature subset) representing the feature selection scheme is obtained; B. After obtaining the binary code of each individual, based on these selected feature subsets, calculate its performance in the classification task according to formula (3) and formula (4) to obtain the fitness value; C. Binary-coded values ​​(feature selection representation) are essential for searching for optimal feature subsets using the Moss Optimization Algorithm. Without them, effective feature subset search and combination cannot be performed. Fitness values ​​are the core for evaluating and driving iterative evolution of a population. Without them, it is impossible to judge and optimize feature selection results. Therefore, binary-coded values ​​are a prerequisite for fitness calculation.

[0063] Step S4: Improve the moss reproduction mechanism, remove the cryptobiotic (dormant) mechanism, introduce a dual-mode propagation search, and update the individual positions of the population.

[0064] Step S5: Test the relevant parameters of the biological learning strategy, select the optimal biological learning moss optimization algorithm, and dynamically adjust the moss individuals. The core idea of ​​the biogeography-based learning strategy (BLS) is to generate example vector indexes through the biogeography-based optimization algorithm (BBO). = ]. According to the optimal solution of the moss individual's own history ( ) to sort all particles by their mass. Taking the minimization problem as an example, the steps for generating examples in BLS are described as follows: ; in, s 1 is to have the best The subscript of the moss individual, s 2 is to have a suboptimal The subscript of the moss individual, sN Is to have the worst The subscript of the particle.

[0065] Then, the particles' sorting values ​​are assigned as follows: = N- 1 , = N- 2 , … = 0 (5) According to formula 5, s 1Best The highest ranking value of the moss individual, Have the worst The particle with the lowest ranking value gets .

[0066] Secondly, all particles are assigned in-migration rate and out-migration rate. Here, taking the linear migration model as an example, the in-migration rate and out-migration rate of all particles can be calculated according to the following formula (6): (6) According to formula 6, Have the best The solution with the lowest immigration rate and the highest emigration rate ; And have the worst solution will have the highest immigration rate and the lowest emigration rate .

[0067] Finally, biologically based particles i An example of the generation method can be seen: (1) Having the best moss individuals , due to its immigration rate lowest, and therefore more inclined to learn in their own direction; (2) Having the worst moss individuals , due to its immigration rate The highest, so it tends to move towards other particles study; (3) Have the best Particles , which has the highest emigration rate, so its Easier to be learned by other particles; (4) Having the worst The particles with the lowest migration rate have the lowest migration rate, so their It is not easy to be learned by other particles.

[0068] First, we compared the performance of BLMGO under six different migration models. These six versions of BLMGO are marked as BLMGOi. i From 1 to 6, corresponding to i In general, BLMGO2 performs the best among the six BLMGO versions, as it achieves the best results on 12 of the 30 CEC 2017 test functions. The results are shown in Tables 3 and 4, and the function curves are shown in Table 3. Figure 2 shown.

[0069] Figure 2 In the figure, the blue dotted line is MGO, the yellow solid line is BLMGO1, the red solid line is BLMGO2, the green dotted line is BLMGO3, the blue solid line is BLMGO4, the purple dotted line is BLMGO5, and the pink dotted line is BLMGO6.

[0070] Table 3. Test results of BLMGO and MGO with different parameters on the CEC2017 function

[0071] Table 4 Ranking results

[0072] Step S6: Optimizing the moss algorithm using the optimal biological learning strategy and constructing a prediction model with the KNN classifier; designing a k-nearest neighbor (KNN) based classifier implementation method and its optimization technology based on the training sample set, and performing classification; For example, a KNN classifier is designed based on the sample set and classified. By calculating the distance between the sample to be classified and all samples in the training set, the k samples with the closest distance are selected, and the category of the sample to be classified is decided by voting based on the category of these k samples. Each characteristic value of each individual in is simulated into a binary code value through formula (1) and formula (2), and the fitness value of the current population individual is obtained through formula (3) and (4), and the individuals are sorted from small to large according to the fitness value.

[0073] The basic concept behind KNN is a simple, parameter-free, and efficient learning method. It has demonstrated excellent results in problems such as function classification and approximation, with high completion rates and classification accuracy. In recent studies, KNN, as a classifier in models, has also demonstrated excellent training speed and classification accuracy. Feature selection is the process of achieving optimal classification accuracy using the fewest subsets, essentially improving classification accuracy while reducing the number of features. This problem is formulated as a combinatorial optimization problem. To meet each objective, a fitness function is defined using a linear combination of the error rate and the number of features. Candidate solutions are comprehensively evaluated to accurately predict the efficacy of immunochemotherapy in patients with non-small cell lung cancer.

[0074] (7) In the above formula, E is the classification accuracy of the KNN classifier, R is the length of the selected feature subset, D is the total number of features in the dataset. α and β are the weights of classification error rate and feature reduction respectively. Since accuracy is obviously more important, α Set to 0.95, β Set to 0.05.

[0075] The BLMGO2-KNN model was used to perform feature selection on routine blood test data and clinical data, identifying the optimal solution, or optimal feature subset. Feature selection can be viewed as a multi-objective optimization problem, with the conflicting objectives of minimizing the number of feature subsets and maximizing classification accuracy. Therefore, an objective function was established to normalize these two objectives into a single function.

[0076] Step S7: The loop is updated to the maximum number of iterations, and the optimal solution for predicting non-small cell lung cancer patients, that is, the optimal clinical feature subset, is output, thereby improving the accuracy of predicting the efficacy of immunochemotherapy for non-small cell lung cancer patients.

[0077] The embodiments of the present invention have the following beneficial effects: Based on the characteristics of clinical data of patients with non-small cell lung cancer, biological learning strategies are integrated into the moss optimization algorithm. By screening and extracting the most relevant features, the prediction performance of immunochemotherapy efficacy in patients with non-small cell lung cancer is improved. This helps medical professionals more accurately identify individuals in the non-small cell lung cancer patient population who may benefit from long-term progression-free survival, thereby optimizing treatment plans and resource allocation.

[0078] Table 5 The method for accurately predicting the efficacy of immunochemotherapy for non-small cell lung cancer according to the present invention

[0079] Table 5 shows the results of the 10-fold cross-validation experiment of the bBLMGO-KNN model. The ACC value reached 89.647%, the sensitivity was 85.455%, the specificity was 92.692%, the accuracy was 90.795%, the MCC value was 79.334%, and the F-measure reached 87.685%.

[0080] Table 6 is a table showing the system effect of the method for accurately predicting the efficacy of immunochemotherapy for non-small cell lung cancer according to the present invention.

[0081] Table 6 shows the results of the 10-fold cross-validation experiment for the bBLMGO-KNN model. The ACC value reached 89.647%, the sensitivity was 85.455%, the specificity was 92.692%, the accuracy was 90.795%, the MCC value was 79.334%, and the F-measure reached 87.685%.

[0082] Based on the same inventive concept, an embodiment of the present invention further provides a feature extraction system for medical clinical data, the system comprising: The data acquisition module is used to acquire medical clinical data sets, which include blood test data and clinical data sets.

[0083] The processing module replaces the moss reproduction mechanism of the moss optimization algorithm with a biogeographically based learning strategy, removes the cryptobiotic effect of the moss optimization algorithm, and introduces a two-mode propagation search to obtain the biological learning moss optimization algorithm. The biological learning moss optimization algorithm is used to find the subset of clinical features most relevant to sample classification during feature extraction in medical clinical datasets.

[0084] The classification module is used to classify the clinical feature subset that is most relevant to the sample classification using the K-nearest neighbor classifier to obtain the optimal clinical feature subset of medical clinical data.

[0085] Among them, the processing module includes: a feature selection module and an optimization module.

[0086] The feature selection module is used to initialize the moss population. Each moss individual represents the candidate medical clinical data and has an independent eigenvalue, resulting in a set of eigenvalues. Feature selection is performed on the set of eigenvalues ​​to obtain the optimal feature subset.

[0087] The optimization module uses a biogeographically informed learning strategy to construct a set of example vectors corresponding to the optimal feature subset. The particles in the example vector set are sorted according to the moss individual's historical optimal solution, obtaining a ranking value for each particle. Based on the ranking value of each particle, in-migration and out-migration rates are assigned to the moss individuals corresponding to the optimal feature subset. A two-mode propagation search is performed based on the in-migration and out-migration rates to dynamically update the positions of the moss individuals and determine the global optimal solution, which is the clinical feature subset most relevant to the sample classification.

[0088] Based on the same inventive concept, an embodiment of the present invention further provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of a feature extraction method for medical clinical data are implemented.

[0089] Based on the same inventive concept, an embodiment of the present invention further provides a storage medium storing a computer program, which implements the steps of a feature extraction method for medical clinical data when executed by a processor.

[0090] The above-described embodiments merely illustrate several implementations of the present invention. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the spirit of the present invention, and all such variations and improvements fall within the scope of protection of the present invention.

Claims

1. A feature extraction method for medical clinical data, characterized in that: include: Acquire a medical clinical data set, wherein the medical clinical data set includes: blood test data and a clinical data set; By replacing the moss reproduction mechanism of the moss optimization algorithm with a biogeographically based learning strategy, the cryptobiotic effect of the moss optimization algorithm is removed, and a two-mode propagation search is introduced to obtain a biological learning moss optimization algorithm. The biological learning moss optimization algorithm is used to find the clinical feature subset most relevant to sample classification during feature extraction of medical clinical datasets. The K-nearest neighbor classifier is used to classify the clinical feature subset that is most relevant to the sample classification, and the optimal clinical feature subset of medical clinical data is obtained; The biological learning moss optimization algorithm is used to find the clinical feature subset most relevant to sample classification in the feature extraction process of the medical clinical dataset, specifically including: Initialize the moss population, where each moss individual represents the candidate medical clinical data and has an independent eigenvalue, thereby obtaining a eigenvalue set; perform feature selection on the eigenvalue set to obtain the optimal feature subset; A biogeography-based learning strategy was used to construct a set of example vectors corresponding to the optimal feature subset. The particles in the example vector set were sorted according to the moss individual's own historical optimal solution to obtain a ranking value for each particle. In-migration and out-migration rates were assigned to the moss individuals corresponding to the optimal feature subset based on the ranking value of each particle. A two-mode propagation search was performed based on the in-migration and out-migration rates to dynamically update the positions of the moss individuals and obtain the global optimal solution, which was the clinical feature subset most relevant to the sample classification.

2. A method for extracting features from medical clinical data according to claim 1, characterized in that: Obtaining the medical clinical data set through electronic medical records, wherein the medical clinical data set specifically includes: demographic characteristics, disease characteristics, pre-treatment hematological indicators, immune characteristics, and efficacy outcome indicators; The efficacy outcome indicators include: efficacy response and survival endpoints; the efficacy response includes: disease remission group and disease non-remission group; the survival endpoint includes: long survival group and short survival group.

3. A method for extracting features from medical clinical data according to claim 1, characterized in that: The feature selection of the feature value set to obtain the optimal feature subset specifically includes: Binary encoding is performed on each moss individual and the eigenvalue set is encoded through a conversion function, so as to perform feature selection according to the position of the eigenvalue of each moss individual in the eigenvalue set, thereby obtaining a plurality of feature subsets; For each feature subset, the weight of the feature subset in the feature extraction process is determined, the fitness value of the feature subset is obtained, and the feature subset with a fitness value greater than a set threshold is taken as the optimal feature subset.

4. A method for extracting features from medical clinical data according to claim 3, characterized in that: The eigenvalue set is encoded by the conversion function based on the following formula to obtain the binary encoding value of the initial moss population: ； ； in, For the The first The individual's binary coded values, For the The first The individual's eigenvalues, is a random number in the range [0, 1], To negate the sign, is the conversion function, is the initial moss population; The weight of the feature subset in the feature extraction process is determined based on the following formula to obtain the fitness value of the feature subset: ； ； in, For the The fitness value of moss individuals, is the classification error rate, is the length of the feature subset, is the characteristic total length, is the weight of the classification error rate, is the weight for selecting the number of feature subsets, is the number of correct classifications in the sample, is the number of misclassified samples.

5. The method for extracting features from medical clinical data according to claim 1, wherein: A set of example vectors corresponding to the optimal feature subset is constructed using a biogeography-based learning strategy based on the following formula: = ]； ； Based on the following formula, the particles in the example vector set are sorted according to the moss individual's own historical optimal solution to obtain the ranking value of each particle: = N- 1 , = N- 2 , … = 0 ； Based on the following formula, the in-migration rate and out-migration rate are assigned to the moss individuals corresponding to the optimal feature subset according to the ranking value of each particle: ； in, is the particle’s own historical optimal solution, sN is the particle subscript, is the immigration rate, is the emigration rate, N is the sort value of the particle, rank is the sorting function, is an example vector, is a set of example vectors, i is the index.

6. A method for extracting features from medical clinical data according to claim 5, characterized in that: The dual-mode propagation search is performed based on the immigration rate and the emigration rate to dynamically update the position of the moss individual, specifically including: The moss individual with the best self-historical optimal solution has the lowest immigration rate and tends to update its position toward itself; The moss individual with the worst historical optimal solution has the highest immigration rate and tends to update its position toward other moss individuals; The particle with the best historical optimal solution has the highest migration rate and is used as the update direction by other moss individuals; The particle with the worst historical optimal solution has the lowest migration rate and is not used as an update direction by other moss individuals.

7. A medical clinical data feature extraction system based on a medical clinical data feature extraction method according to any one of claims 1 to 6, characterized in that: include: A data acquisition module, configured to acquire a medical clinical data set, wherein the medical clinical data set includes: blood test data and clinical data set; a processing module for replacing the moss reproduction mechanism of the moss optimization algorithm with a biogeographically based learning strategy, removing the cryptobiotic effect of the moss optimization algorithm, and introducing a bimodal propagation search to obtain a biological learning moss optimization algorithm; and finding a subset of clinical features most relevant to sample classification during feature extraction of a medical clinical dataset using the biological learning moss optimization algorithm; The classification module is used to classify the clinical feature subset that is most relevant to the sample classification using a K-nearest neighbor classifier to obtain the optimal clinical feature subset of the medical clinical data; Wherein, the processing module includes: a feature selection module and an optimization module; The feature selection module is used to initialize the moss population, each moss individual represents the candidate medical clinical data and has an independent feature value, to obtain a feature value set; feature selection is performed on the feature value set to obtain an optimal feature subset; The optimization module is used to construct a set of example vectors corresponding to the optimal feature subset using a biogeography-based learning strategy, sort the particles in the example vector set according to the moss individual's own historical optimal solution to obtain a ranking value for each particle; assign an in-migration rate and an out-migration rate to the moss individual corresponding to the optimal feature subset based on the ranking value of each particle; perform a bimodal propagation search based on the in-migration rate and the out-migration rate to dynamically update the position of the moss individual to obtain a global optimal solution, which is the clinical feature subset most relevant to the sample classification.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the processor implements the steps of a method for extracting features from medical clinical data according to any one of claims 1 to 6.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for extracting features from medical clinical data according to any one of claims 1 to 6 are implemented.

Images