Traditional Chinese medicine anti-tumor curative effect monitoring method based on impedance spectroscopy analysis

Abstract

The invention provides a traditional Chinese medicine anti-tumor curative effect monitoring method based on impedance spectroscopy analysis. Belongs to the technical field of biological medicine. The method comprises the following steps: collecting biological impedance spectroscopy data, and preprocessing the biological impedance spectroscopy data; extracting multi-scale features from the preprocessed bio-impedance spectroscopy data; based on a feature fusion algorithm, the multi-scale features are organically fused to form comprehensive and accurate feature vectors; constructing a hybrid model; generating a dynamic curative effect monitoring report; and monitoring data and a treatment scheme are shared in real time. Through deep analysis of the time domain, the frequency domain and the space on the biological impedance signal, the biological impedance signal can be comprehensively analyzed from different dimensions, richer signal features can be captured, advanced features which cannot be recognized by a traditional method can be mined, and the analysis accuracy is improved.

Description

Technical Field

[0001] The present invention provides a method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectrum analysis, belonging to the field of biomedical technology. Background Art

[0002] With the continuous development of bioimpedance measurement technology, bioimpedance spectrum (BIS) has been widely used in the field of biomedicine, such as disease diagnosis, health monitoring, evaluation of body fluid distribution, etc. However, traditional bioimpedance spectrum data processing methods mainly rely on simple time-domain or frequency-domain feature extraction, unable to fully explore the relationship between complex time-domain, frequency-domain and spatial features, and the feature extraction is too single and cannot process high-dimensional data quickly and effectively. Summary of the Invention

[0003] The present invention provides a method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectrum analysis to solve the problems of too single feature extraction, inability to effectively process high-dimensional data and ignoring spatial features when analyzing bioimpedance spectrum data in the prior art: The method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectrum analysis proposed by the present invention includes: S1. Collect bioimpedance spectrum data and perform preprocessing; S2. Extract multi-scale features from the preprocessed bioimpedance spectrum data; based on the feature fusion algorithm, organically fuse the multi-scale features to form a comprehensive and accurate feature vector; S3. Construct a hybrid model; S4. Generate a dynamic monitoring report on the efficacy; S5. Share the monitoring data and treatment plan in real time.

[0004] The system for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectrum analysis proposed by the present invention includes a memory, a processor, and a computer program stored on the memory and executable on the memory. The processor executes the program to implement the method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectrum analysis as described in any one of the above.

[0005] Advantages of the present invention: The technical solution proposed by the present invention can comprehensively analyze bioimpedance signals in different dimensions by deeply analyzing bioimpedance signals in the time domain, frequency domain, and space, capture richer signal features, and can extract high-level features that cannot be recognized by traditional methods, improving the accuracy of analysis; and by fusing the analyzed features respectively, a comprehensive and accurate feature vector can be formed, thereby improving the comprehensive performance of the model; through acceleration based on GPU parallel computing and distributed computing frameworks, large-scale bioimpedance data sets can be processed, improving the efficiency of data processing, and accelerating the model training and prediction speed; by analyzing the spatial distribution of bioimpedance data through spatial autocorrelation functions and semivariograms, the correlation and influence between different biological regions can be revealed, providing a scientific basis for the analysis of the spatial distribution of diseases or regional health risk assessment; by constructing a spatial weight matrix and analyzing the mutual influence between various parts, it helps to understand the interconnectivity and interaction between different biological regions and improve the accuracy of diagnostic and prediction models. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0007] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not intended to limit the present invention.

[0008] One embodiment of the present invention is as Figure 1 shown, a method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectroscopy analysis, the method comprising: S1. Collect bioimpedance spectroscopy data through a wearable bioimpedance spectroscopy data acquisition device, wherein the data acquisition device includes a high-precision current source, a voltage measurement module, a multi-frequency signal generator, and a flexible electrode array, and preprocess the collected bioimpedance spectroscopy data through an adaptive filtering and denoising algorithm; S2. Extract multi-scale features from the preprocessed bioimpedance spectroscopy data, the multi-scale features including time-domain features (such as impedance amplitude, phase angle), frequency-domain features (such as spectral density, coherence), and spatial features (such as differences in impedance spectra of different parts); based on a feature fusion algorithm, organically fuse the multi-scale features to form a more comprehensive and accurate feature vector; S3. Based on the deep learning framework, construct a hybrid model based on convolutional neural network (CNN) and recurrent neural network (RNN), where CNN is used to extract local spatial features, and RNN is used to capture the dynamic changes of time series data; process the multi-scale feature vectors through the hybrid model to predict the anti-tumor efficacy of traditional Chinese medicine; and through the attention mechanism, enhance the model's ability to capture key features; adopt transfer learning and incremental learning methods to continuously optimize the model performance; S4. Embed the deep learning model into a wearable device to continuously monitor the changes in the characteristic parameters of the patient's bioimpedance spectrum in real time. The device can continuously collect data and transmit the monitoring results to the cloud for processing and analysis in real time, and generate a dynamic monitoring report of the treatment effect based on the model prediction results; S5. Based on the intelligent feedback mechanism, automatically adjust the traditional Chinese medicine treatment plan according to the monitoring report, such as adjusting the drug combination, dosage or frequency; and through the doctor-patient interaction platform, share the monitoring data and treatment plan in real time.

[0009] The working principle and effects of the above technical solutions are as follows: By continuously monitoring the bioimpedance spectrum data of patients in real time, the changes in the treatment effect in the patient's body can be accurately captured, so as to dynamically adjust the traditional Chinese medicine treatment plan according to the actual response of the patient, thereby optimizing the treatment effect and avoiding drug abuse or unnecessary treatment interventions; The application of multi-scale feature fusion technology and deep learning model improves the accuracy of efficacy prediction. Through the processing ability of deep learning, information in multiple dimensions such as time domain, frequency domain and space can be comprehensively considered to accurately monitor the treatment effect, improve the accuracy of monitoring and the precision of subsequent judgment; Combining the attention mechanism can enable the model to focus on the features that have a greater impact on the treatment effect, improving the reliability of the prediction results; By adopting transfer learning and incremental learning, the model can continuously self-optimize as the patient's treatment process changes, adapt to new bioelectrical changes, and ensure that even in complex and long-term treatment processes, the treatment strategy can be adjusted in real time to achieve the optimal treatment effect; The doctor-patient interaction platform provides doctors with real-time data of patients, and through automatic adjustment of the plan, reduces the burden on doctors, and at the same time enhances the patient's sense of participation and trust in the treatment; Real-time sharing of data and treatment plans can help doctors and patients to have more interaction and feedback, form a good cooperative relationship, and improve the treatment effect; The use of wearable devices enables patients to conduct routine monitoring at home or in daily life, avoiding the cumbersome and inconvenient traditional hospital detection methods, and improving patient compliance and treatment continuity.

[0010] In one embodiment of the present invention, the S1 includes: S11. Make the electrode array with soft and skin - conforming flexible materials, which contains multiple electrode points to meet the data acquisition requirements of different body parts. The design of the electrode array needs to consider factors such as electrode spacing, shape, material, etc. to ensure that the collected bio - impedance spectrum data is representative. Based on a high - precision current source, inject a weak alternating current into the organism in a constant - current source mode, and at the same time use a high - precision voltage measurement module to detect the response voltage and calculate the bio - impedance. The precision of the current source and voltage measurement module needs to reach the micro - ampere level and micro - volt level to reduce measurement errors. Generate an alternating current signal covering the key frequency range of the bio - impedance spectrum through a multi - frequency signal generator. The frequency range is adjusted according to specific requirements, but usually should include low - frequency (e.g., below 1 kHz), medium - frequency (e.g., 1 kHz to 1 MHz), and high - frequency (e.g., above 1 MHz) components. For example, low - frequency signals may more reflect the electrical properties of cells, while high - frequency signals may be related to factors such as the conductivity between tissues. S12. Based on a preset data acquisition strategy, including acquisition time, frequency, duration, etc., collect data. During the acquisition process, record the bio - impedance spectrum data in real - time, and at the same time record the patient's basic information, such as age, gender, tumor type, stage, etc. Among them, combining the basic information with the bio - impedance data helps with subsequent analysis and processing. S13. Adopt an adaptive filtering algorithm to dynamically adjust the filtering parameters according to the data characteristics.

[0011] The working principle and effects of the above - mentioned technical solutions are as follows: By using a high - precision current source and voltage measurement module, it is possible to accurately measure bio - impedance spectrum data, reduce errors, and improve the accuracy of the collected data. The adoption of multi - frequency signals ensures a wide coverage of different frequency bands of the bio - impedance spectrum, capturing the comprehensive electrical characteristics in the patient's body. The setting of the data acquisition strategy and the real - time recording function ensure the continuity and integrity of the data. Recording the patient's bio - impedance data in real - time and synchronously recording the patient's basic information can provide a comprehensive understanding of the patient's condition, further improving the precision of treatment. The application of the adaptive filtering algorithm guarantees the reliability of the data during the acquisition process, removes unnecessary noise, ensures excellent signal quality, and thus improves the data analysis effect. Combining the patient's basic information and bio - impedance data enables more accurate personalized treatment monitoring. Through this comprehensive analysis, the treatment effect can be more accurately evaluated, helping doctors adjust the treatment plan in a timely manner.

[0012] In one embodiment of the present invention, S13 includes: Among the collected bio-impedance spectrum data, by setting thresholds or detecting specific abnormal patterns, such as suddenly jumping signals, automatically identify these abnormal data points through setting rules; and remove the obvious outliers caused by equipment failures, poor electrode contacts or environmental interferences; and further detect and eliminate potential data outliers by applying statistical methods (such as box plot analysis, Z-score normalization); for example, the Z-score normalization method calculates the deviation of each data point from the mean value, screens out the data with too large differences from other data points, so as to find the abnormal data; Smooth the bio-impedance data by using moving average filtering or Savitzky-Golay filtering; combine wavelet transform technology, decompose the signal into multiple scales, implement an adaptive denoising strategy for different frequency components, and retain the key bio-impedance features; Through an adaptive filtering algorithm based on the recursive least squares method, dynamically adjust the filter parameters according to the real-time data to suppress non-stationary noise; introduce intelligent algorithms (such as genetic algorithms, particle swarm optimization) to optimize the performance of the adaptive filter, and find the optimal combination of filter parameters through iterative search to improve the filtering efficiency and accuracy; Based on the preprocessed data, extract key bio-impedance features, such as phase angle, real and imaginary parts of impedance, etc.; adopt data augmentation techniques (such as time-domain translation, frequency-domain transformation) to increase data diversity and improve the generalization ability of the model; Establish a real-time data quality monitoring system, and timely detect and correct the deviations in the data processing process through preset thresholds and monitoring indicators.

[0013] Implement a closed-loop feedback mechanism, adjust the data acquisition strategy according to the data preprocessing results (such as increasing the number of sampling points, adjusting the frequency range), and form a cycle of continuous improvement.

[0014] The working principle and effects of the above technical solution are as follows: By initially cleaning the data, obvious outliers and potential data biases can be removed, thus improving the accuracy of the data; through wavelet transform and adaptive filtering algorithms, environmental noise and equipment errors can be removed, while the true bio-impedance signal is retained, enhancing the clarity and usability of the signal, thereby providing an accurate data source for subsequent steps such as modeling and analysis; the combination of moving average filtering, Savitzky-Golay filtering, and wavelet transform not only effectively smooths the data but also effectively removes noise, enhancing the useful information in the signal and avoiding interference from high-frequency noise; the adaptive filtering algorithm of recursive least squares can dynamically adjust the filtering parameters, optimize for different types of noise, and ensure high-quality output of the signal; through the extraction of key bio-impedance features, accurate parameters can be provided for subsequent analysis, contributing to applications such as disease prediction and treatment monitoring; data augmentation techniques effectively increase the diversity of the dataset, enhancing the generalization ability of the model, enabling the model to adapt to different physiological conditions and environmental changes, and improving the accuracy of the analysis; through the real-time data quality monitoring system, deviations and errors in data processing can be immediately detected, avoiding the impact of bad data on subsequent analysis; the closed-loop feedback mechanism ensures continuous improvement of the data acquisition and processing strategies, thereby enhancing the flexibility and adaptability of the system and improving the robustness of the entire system.

[0015] In one embodiment of the present invention, S2 includes: S21. Extract basic time-domain features, where the basic time-domain features include impedance amplitude and phase angle, and further extract advanced time-domain features, where the advanced time-domain features include impedance change rate and impedance stability index; calculate the fractal dimension of the impedance spectrum based on fractal analysis; S22. Based on spectral density and coherence, add frequency-domain features, where the frequency-domain features include spectral slope and spectral intercept; use wavelet transform to decompose the signal into wavelet coefficients at different scales, and extract features such as energy and entropy at each scale; S23. Use spatial autocorrelation function and semivariogram to analyze the spatial distribution and correlation of bio-impedance spectrum data at different locations, and extract advanced spatial features, where the advanced spatial features include spatial autocorrelation length and anisotropy; S24. Construct a spatial weight matrix, and based on adjacency relationships or distance metrics, evaluate the mutual influence between each location and extract spatial interaction features; S25. Transform the original bio-impedance spectrum data, where the transformation includes adding random noise and frequency offset to generate diverse training samples; and apply feature selection methods to remove redundant and irrelevant features; use principal component analysis to map the high-dimensional feature space to a low-dimensional space and retain key information; S26. In the feature extraction stage, based on a hierarchical feature fusion framework, time-domain, frequency-domain, and spatial features are fused at different levels. For example, time-domain and frequency-domain features can be first fused into composite features and then fused with spatial features. S27. Based on a deep learning model, perform representation learning on the features, automatically learn the complex relationships and latent structures between the features, and utilize the parallel computing power of the GPU to accelerate the feature extraction and fusion process. Based on a distributed computing framework, divide the large-scale dataset into small chunks and process them in parallel on multiple nodes. Through feature alignment techniques such as dynamic time warping (DTW) or longest common subsequence (LCS), align the features from different sources in time or space. S28. Use feature standardization and normalization methods to eliminate the dimensional difference and numerical range difference between the features. Apply mutual information to evaluate the difference between different features and identify the features with unique information. S29. Combine feature selection methods and clustering analysis to identify and remove redundant features, reduce the overlap and duplicate information between the features. Use random forest to evaluate the importance of each feature and weight the features according to their importance. Based on the dynamic adjustment mechanism of feature importance, dynamically adjust the feature weights according to the performance of the model on the validation set.

[0016] The working principle and effects of the above technical solution are as follows: By extracting features at multiple levels (time domain, frequency domain, space), the solution can comprehensively analyze bioimpedance signals from different dimensions, capture richer signal characteristics, and thus improve the accuracy and reliability of the model; Through techniques such as fractal analysis, spectral density, wavelet transform, and spatial autocorrelation function, advanced features that cannot be identified by traditional methods can be mined, thereby improving the accuracy and depth of data analysis; Through principal component analysis and feature selection methods, the dimensionality of the data can be effectively reduced, unnecessary features can be reduced, and overfitting of the model in high-dimensional space can be avoided, thus enhancing the generalization ability of the model in practical applications; Through feature selection, clustering analysis, and random forest to evaluate feature importance, redundant features that are not helpful for prediction can be removed, ensuring that the features used have strong discrimination ability, and improving the accuracy and efficiency of analysis; Through acceleration based on GPU parallel computing and distributed computing frameworks, large-scale bioimpedance datasets can be processed, significantly improving data processing efficiency, and accelerating model training and prediction speed; By adopting a hierarchical feature fusion framework, time-domain, frequency-domain, and spatial features are fused at different levels, enhancing the model's comprehensive understanding of signal features, contributing to improving the overall performance and stability of the model, and thus enhancing the accuracy of prediction; By analyzing the spatial distribution of bioimpedance data through spatial autocorrelation function and semivariogram, the correlation and influence between different biological regions can be revealed, providing a scientific basis for the analysis of the spatial distribution of diseases or regional health risk assessment; By constructing a spatial weight matrix and analyzing the mutual influence between parts, it helps to understand the interconnectivity and interaction between different biological regions, and improves the accuracy of diagnostic and prediction models; Through standardization and normalization methods, the dimensional difference and numerical range difference between different features are eliminated, ensuring that the influence of each feature in the model is relatively balanced, and contributing to improving the stability and reliability of the algorithm; By evaluating the difference between different features through mutual information, features with unique information can be identified, further optimizing the feature selection process, and improving the accuracy of the model; By dynamically adjusting the weights of features and optimizing according to the performance of the model on the validation set, it can adapt to different datasets and application scenarios, and enhance the adaptability and flexibility of the model; Through in-depth analysis of bioimpedance data, it can provide strong support for medical diagnosis, and improve the accuracy of disease prediction, early diagnosis, and health assessment; Through precise feature extraction and analysis, a more personalized health monitoring plan can be provided, helping doctors and medical institutions to provide customized treatment plans or health management strategies; Through distributed computing and parallel computing frameworks, massive bioimpedance data can be processed, especially suitable for applications in big data backgrounds such as health monitoring, clinical research, and public health fields; Accelerating the feature extraction and model training process, providing support for real-time health monitoring and warning systems, and enabling rapid response and decision-making.

[0017] In one embodiment of the present invention, the S26 includes: Determine the objectives and principles of feature fusion, that is, to maximize the fusion of features from different domains without losing key information, so as to improve the prediction performance and generalization ability of the model; according to the physical meaning and biological interpretation of features, conduct preliminary classification and screening of time-domain, frequency-domain, and spatial features to ensure that the features participating in the fusion are representative; Adopt linear or non-linear combination methods (such as weighted summation, product, polynomial combination) to preliminarily fuse time-domain features (such as impedance magnitude, phase angle, impedance change rate, impedance stability index) with frequency-domain features (such as spectral slope, spectral intercept, wavelet coefficient energy, entropy); Based on deep learning networks (such as convolutional neural network CNN, recurrent neural network RNN, or long short-term memory network LSTM), further learn and represent the fused features to capture the complex relationships and potential structures between features; Further fuse spatial features (such as spatial autocorrelation length, anisotropy, spatial interaction features) with the aforementioned time-frequency fusion features; Based on the network architecture of the multi-scale spatial attention mechanism, dynamically adjust the weights of spatial features by calculating the correlations between different spatial positions; The working principle and effects of the above technical solution are as follows: Through multi-dimensional feature fusion and different types of deep learning models, the inherent laws of biological signals can be captured more comprehensively; not only the prediction performance of the model on the training data is improved, but also the adaptability and generalization ability of the model to unseen data are enhanced; using deep learning models for feature representation learning can automatically discover the complex relationships between features, further improving the accuracy and stability of the model; by preliminarily screening and classifying features, it is ensured that only representative time-domain, frequency-domain, and spatial features participate in the fusion, avoiding the interference of redundant information and enhancing the discriminability of features; using linear or non-linear methods to combine features from different domains can extract more meaningful information without losing key information, enabling the model to process more complex biological signal data; using deep learning models to further learn the fused features can not only learn the direct associations between features, but also capture complex non-linear relationships and discover the latent structure in the data, thus making data analysis more thorough; traditional deep learning models have certain limitations when processing spatial data, so GNN or SCN is used to specifically process spatial data, enabling the model to more effectively learn spatial structures and relationships and improve the generalization ability of the model; GNN and SCN can process complex spatial graph structures or local spatial data, helping the model to more accurately model and predict spatial features while considering spatial interrelationships; the network architecture based on the multi-scale spatial attention mechanism can dynamically adjust the feature weights of each spatial position by calculating the correlations between different spatial positions, improving the model's ability to capture spatial features, enhancing the expression ability, improving the robustness and computational efficiency, and enhancing the model's performance in different tasks. Through the fusion of multiple features and the representation learning of deep learning, a large amount of training data can be fully utilized and the risk of model overfitting can be reduced; the synergistic effect of time-domain, frequency-domain, and spatial features can comprehensively understand the multi-dimensional information of biological signals. This cross-domain feature fusion can provide the model with richer background information, enabling the model to understand and predict data at multiple levels.

[0018] In one embodiment of the present invention, S27 includes: S271. Input the fused features into a deep learning model for representation learning; the deep learning model automatically extracts feature representations useful for the prediction task by learning the complex relationships and latent structures between features; S272. Use cross-validation to evaluate the performance of the model on the validation set to ensure the effectiveness and accuracy of the feature representation; utilize the parallel computing power of the GPU to accelerate the training and inference processes of the deep learning model; by optimizing steps such as data loading and model calculation, reduce the calculation time and improve the processing efficiency; S273. Based on the GPU parallel computing algorithm and data structure, further improve the computing speed and resource utilization according to the characteristics of bioimpedance spectroscopy data and the requirements of deep learning models; S274. Based on the distributed computing framework (such as Apache Spark, TensorFlow Distributed, etc.), split the large-scale bioimpedance spectroscopy dataset into small pieces and process them in parallel on multiple nodes; through steps such as optimizing data partitioning and task scheduling, achieve efficient parallel processing of data.

[0019] S275. Based on the deep learning training framework of data parallel processing, conduct distributed training on the model by exchanging intermediate results and gradient information between network layers or nodes; S276. For features from different sources or at different time points, adopt feature alignment techniques (such as dynamic time warping DTW, longest common subsequence LCS, etc.) to align them in time or space to ensure the consistency and comparability of features; S277. On the basis of feature alignment, adopt efficient feature fusion algorithms and data structures to perform fast fusion and calculation of features; during the training process of the deep learning model, adopt optimization methods such as regularization and dropout to prevent the model from overfitting and improve the generalization performance of the model; S278. Based on the dynamic adjustment mechanism of feature importance, dynamically adjust the weights of features in the network and the structure of the model according to the performance of the model on the validation set to achieve adaptive optimization of the model.

[0020] The working principle and effects of the above technical solution are as follows: By using a deep learning model to learn the complex relationships and potential structures between features, the model can automatically extract feature representations useful for the prediction task, which helps to improve the model's understanding and prediction ability of biosignal data. Especially when facing high-dimensional and complex data, it can effectively capture important patterns and hidden features in the data. By evaluating the model's performance through cross-validation, the generalization ability of the model on the validation set can be ensured, thereby improving the accuracy of the model. At the same time, using GPU parallel computing to accelerate the training and inference processes can reduce the computing time and improve the training efficiency of the model. For the characteristics of bioimpedance spectroscopy data, combined with GPU parallel computing algorithms to further optimize the computing speed and resource utilization rate, the processing efficiency can be improved. Especially when processing large-scale and high-dimensional data, it can effectively avoid computing bottlenecks and ensure the high efficiency and real-time performance of model training. By using a distributed computing framework, splitting a large-scale dataset into small pieces and processing them in parallel on multiple nodes can significantly improve the utilization rate of computing resources and accelerate the processing process of large-scale data. Through data partitioning and task scheduling optimization, the efficiency of large-scale data processing can be improved, which is especially suitable for deep learning tasks that require large amounts of data for training. Through a data parallel processing framework, distributed training of the model is carried out to ensure efficient computing during the training process and the processing of larger-scale models. By adopting feature alignment technology, ensuring that features from different sources or different time points can be consistently aligned in time or space can eliminate the impacts caused by time differences, signal noise, or measurement errors, improve the comparability of features, and thus enhance the robustness of the model. Efficient feature fusion algorithms and regularization methods help prevent model overfitting and enhance the generalization performance of the model. Feature fusion technology can effectively integrate multi-source data in a multi-dimensional feature space and improve the model's understanding ability of complex biosignals. Based on the dynamic adjustment mechanism of feature importance, by dynamically adjusting the weights of features in the network and the structure of the model according to the model's performance on the validation set, further adaptive optimization of the model is achieved.

[0021] In one embodiment of the present invention, the S274 includes: Before distributed computing, preprocess the large-scale bioimpedance spectroscopy data. The preprocessing includes cleaning, removing noise, outliers, or missing data, and performing standardization processing to ensure the consistency and comparability of the data; According to the characteristics of the data (such as timestamps, sample IDs, etc.), intelligently split the dataset into multiple small pieces, each small piece containing a relatively independent and complete data subset; at the same time, construct an efficient data index to facilitate subsequent rapid retrieval and parallel processing; Use a distributed file system, such as Hadoop HDFS or Amazon S3, to store the sharded dataset on multiple physical nodes for data redundancy backup and load balancing; According to the characteristics and processing requirements of bioimpedance spectroscopy data, comprehensively evaluate the applicability of distributed computing frameworks such as Apache Spark, TensorFlow Distributed, and Dask, and select the most suitable framework for deployment; for example, if a large amount of data needs to be processed in a short time and real-time performance is required, then choose Apache Spark; Based on the hardware resources of the cluster (such as the number of CPUs, GPUs, and memory size), dynamically adjust the resource allocation of each node so that tasks can be efficiently and evenly distributed among all nodes; at the same time, through the task scheduling algorithm, reduce the task waiting time and resource idle rate; For the network communication bottleneck in distributed computing, adopt technologies such as data compression and serialization optimization to reduce the amount of data transmission; use high-speed network technologies such as RDMA (Remote Direct Memory Access) to improve the data transmission rate between nodes; According to the processing logic of bioimpedance spectroscopy data, through data parallel processing strategies, such as data parallelism, model parallelism, or hybrid parallelism; perform data parallel processing; For specific analysis tasks of bioimpedance spectroscopy data (such as feature extraction, classification prediction, etc.), adopt efficient algorithms and data structures, such as the Fast Fourier Transform (FFT), K-means clustering, and sparse matrix storage, to accelerate the calculation process; During the distributed processing process, real-time monitor the load conditions of each node, and through multiple directions, the multiple methods include dynamically adjusting task allocation or starting standby nodes to perform load balancing; at the same time, based on the fault tolerance mechanism, recover the computing tasks when a node fails; After the distributed processing is completed, integrate the processing results of all nodes, and ensure the integrity and accuracy of the results through the verification mechanism; Based on the integrated results, comprehensively evaluate the model using methods such as cross-validation and AUC-ROC curve analysis; according to the evaluation results, dynamically adjust the model parameters, feature selection, or algorithm implementation.

[0022] The working principle and effects of the above technical solution are as follows: By preprocessing, denoising, removing outliers and missing data from large-scale bioimpedance spectroscopy data and performing normalization processing, the consistency and comparability of the data can be ensured, the errors caused by inconsistent data quality can be reduced, and the accuracy of subsequent analysis can be improved; intelligently segmenting the data set and performing efficient data indexing and storage can optimize the data reading and query speed and significantly improve the processing efficiency; using a distributed file system to store data can efficiently manage massive data and perform redundant backups among multiple nodes to ensure data security and high availability; based on the selection and deployment of a distributed computing framework, the system can automatically scale according to the data volume and computing requirements to ensure the scalability of large-scale data processing; by dynamically adjusting node resource allocation and load balancing, hardware resources can be effectively utilized, computing bottlenecks can be avoided, and computing performance can be improved; technologies such as data compression and serialization optimization are adopted to reduce the data transmission volume, reduce the network communication burden, and improve the data transmission efficiency; high-speed network technology is used to increase the data transmission rate between nodes, further reduce the data exchange time and latency, and enhance the distributed computing performance; a data parallel processing strategy is adopted. According to the processing logic of bioimpedance spectroscopy data, the tasks are divided into smaller and more independent parts to increase the parallelism of computing and shorten the processing time; in specific analysis tasks, the computing is accelerated by applying efficient algorithms and data structures to improve the computing speed of the algorithms; the load conditions of the distributed system are monitored in real time, and the waiting time and resource idleness are reduced through task scheduling and dynamic task allocation algorithms to improve the resource utilization rate; when a node fails, the system can recover the computing task through a fault tolerance mechanism to ensure the high reliability of the system and the continuity of the computing process; after the distributed computing is completed, a verification mechanism is adopted to ensure the integrity and accuracy of the processing results of each node and reduce the problem of result inconsistency caused by distributed computing; evaluation methods such as cross-validation and AUC-ROC curves are used to comprehensively evaluate the model to dynamically adjust model parameters, feature selection or algorithms to achieve model optimization and accuracy improvement; in scenarios that require real-time performance and high throughput, the needs of fast data processing and real-time analysis can be met by selecting an appropriate distributed framework.

[0023] In one embodiment of the present invention, the S275 includes: Evaluate the applicability of different distributed training frameworks (such as TensorFlow Distributed, Horovod, PyTorch Distributed, etc.) according to the characteristics of bioimpedance spectroscopy data and the requirements of the deep learning model; Based on the architecture of the parameter server and the worker nodes, effectively combine data parallelism and model parallelism; among them, the parameter server is responsible for storing and updating model parameters, and the worker nodes are responsible for processing data and calculating gradients; dynamically adjust the number of nodes and resource configuration according to the cluster resources and task requirements; Split the large-scale bio-impedance spectroscopy dataset into small pieces according to specific strategies (such as random sharding, feature-based sharding), and distribute them to different worker nodes; Design a model parallelization strategy for the specific structure of the deep learning model (such as convolutional layers, fully connected layers, etc.); allocate different parts of the model to different nodes, and perform data synchronization and gradient exchange between nodes through network communication; After calculating the gradients on the worker nodes, use a gradient aggregation algorithm to merge the gradients of multiple nodes into a global gradient; and reduce the amount of gradient transmission through gradient compression techniques (such as quantization, sparsification, etc.); The parameter server updates the model parameters according to the aggregated gradients, and distributes the updated parameters to each worker node; based on the consistency maintenance mechanism, synchronize the model parameters on all nodes; During the distributed training process, adopt a fault tolerance mechanism to handle problems such as node failures or network interruptions; when a failure occurs, be able to quickly resume the training task; at the same time, use elastic recovery technology to quickly resume the training process according to the state before the failure; Dynamically adjust the hyperparameters according to the performance of the model on the validation set, where the hyperparameters include the learning rate and the batch size; at the same time, adjust the weights of the features in the network and the structure of the model according to the feature importance dynamic adjustment mechanism (such as described in S278).

[0024] The working principle and effects of the above technical solution are as follows: By evaluating the applicability of different distributed training frameworks, the most suitable framework for the current task and hardware environment can be selected to ensure efficient distributed training on large-scale datasets, which can improve training efficiency, accelerate data processing, optimize the use of hardware resources, reduce communication overhead, and improve fault tolerance. The flexibility and scalability of different frameworks enable the system to adapt to various training tasks and maximize performance using cluster resources; Based on the design of the parameter server and worker node architecture, combining the advantages of data parallelism and model parallelism, the computing tasks are efficiently allocated on different nodes, and the parameter server is used to manage and update model parameters to ensure the stability and consistency of the training process; Dividing the bioimpedance spectroscopy dataset into small pieces according to a specific strategy and distributing them to different worker nodes can significantly improve the parallelism of data processing, reduce the load on a single node, and improve the overall training efficiency; Through the model parallelization strategy for the specific structure of deep learning models, different parts of the model are allocated to different nodes for computing to solve the problem of insufficient memory and computing resources on a single node and ensure that the model can be trained on large-scale datasets; After the gradients are calculated on each worker node, a gradient aggregation algorithm is used to merge the gradients of multiple nodes and generate the global gradient. Through this strategy, the bottleneck problem caused by gradient calculation on a single node is avoided, and the computing efficiency is improved at the same time; The gradient compression technology is used to reduce the amount of gradient transmission, which not only reduces the network bandwidth pressure but also improves the communication efficiency of distributed training, enabling large-scale distributed training to proceed smoothly under limited bandwidth conditions; During the distributed training process, the fault tolerance mechanism can timely handle problems such as node failures or network interruptions, ensuring that the training process is not interrupted due to the failure of individual nodes, which can greatly improve the robustness and reliability of the system and ensure the smooth progress of long-term and large-scale training tasks; Using the elastic recovery technology, the training process can be quickly restored according to the state before the failure, avoiding the waste of resources caused by starting training from scratch, and ensuring the rapid recovery of training tasks in case of node failures or other emergencies; According to the performance of the model on the validation set, hyperparameters such as the learning rate and batch size are dynamically adjusted to optimize the training process according to the actual situation and further improve the training efficiency and convergence speed of the model; Through the feature importance dynamic adjustment mechanism, the weights of features can be adjusted according to the contribution of different features to model prediction, enabling the model to adaptively focus on the most meaningful features and improve the performance of the model; Adjust the model structure, including the number of layers, activation functions, etc. in the deep learning model, to further optimize the network structure and improve the prediction ability and generalization ability of the model; Based on the task requirements and cluster resource conditions, dynamically adjust the number of nodes and resource allocation.This kind of dynamic resource scheduling can ensure that computing tasks receive sufficient resource support and avoid waste of resources; when the data volume or computing load changes, the system can increase or decrease computing nodes in real time and optimize resource allocation to meet the requirements of tasks of different scales, improving the flexibility and adaptability of the training process.

[0025] In one embodiment of the present invention, step S3 includes: S31. Construct a convolutional neural network (CNN) structure, including an input layer, a convolutional layer, a pooling layer, and a fully connected layer; the convolutional layer is used to extract local features and spatial relationships in the feature vector; the pooling layer is used to reduce the feature dimension and computational amount; the fully connected layer is used to map the features to the prediction result; construct a recurrent neural network (RNN) structure, including an input layer, a hidden layer, and an output layer; the RNN is used to capture the time series characteristics and dependencies in the feature vector; S32. Fuse the outputs of the CNN and RNN through feature concatenation to form the final output of the hybrid model; divide the data set into a training set, a validation set, and a test set; the training set is used to train the model; the validation set is used to adjust the model parameters; the test set is used to evaluate the model performance; S33. Evaluate the performance of the model through cross-validation and early stopping method; cross-validation can be used to evaluate the generalization ability of the model; the early stopping method can be used to prevent the model from overfitting; S34. Introduce an attention mechanism into the hybrid model to enhance the model's ability to capture key features and dynamically adjust the weights according to the importance of the features; S35. Use transfer learning to transfer the weights of the pre-trained model to the current task to accelerate the training process; and adopt an incremental learning method to continuously incorporate newly collected data and update the model parameters.

[0026] The working principle and effects of the above technical solution are as follows: By using a convolutional neural network to extract local features and spatial relationships, it can help the model better identify important information in the data. Especially for datasets with spatial or local features such as images and time series, it can improve the model's expressive ability and accuracy; the introduction of a recurrent neural network can handle the temporal dependence in time series data and capture the temporal relationships existing in the data, enabling the model to better process dynamic data and sequential data; by concatenating the features of the outputs of CNN and RNN to form a hybrid model, the advantages of both can be fully utilized, enhancing the model's learning ability for complex data features and avoiding the limitations that may exist in a single model. This hybrid structure enables the model to adapt to more complex tasks, improving the accuracy and reliability of predictions; through cross-validation, the model can more comprehensively evaluate its performance on different datasets, thus avoiding overfitting to the training set. Cross-validation enhances the model's generalization ability, making its performance on unknown data more robust; the early stopping method can prevent overfitting during the model training process. By monitoring the performance changes of the model on the validation set and stopping the training process early, it avoids ineffective training and ensures the efficiency and generalization ability of the model; the introduction of the attention mechanism enables the model to focus more on the key features in the data, reducing the interference of redundant information and improving the effectiveness of feature selection. Dynamically adjusting the weights according to the importance of features enables the model to automatically focus on the part that has the greatest impact on the prediction result during the training process, further enhancing the model's prediction ability; through transfer learning, the model weights that have been trained on other tasks can be transferred to the current task, greatly accelerating the model training process. Especially when the amount of data is insufficient, transfer learning can effectively improve the model performance and reduce the consumption of computing resources; the incremental learning method enables the model to perform online learning when the data is continuously increasing, continuously updating the model parameters and maintaining the latest state of the model, thus coping with the challenge of rapid data changes in the real environment. This method can continuously improve the accuracy of the model and adapt to the changing data distribution; through the comprehensive application of the hybrid model, cross-validation, early stopping method, attention mechanism, transfer learning and incremental learning, the technical solution plays an important role in improving the prediction performance, generalization ability and training efficiency of the model. Especially for tasks such as large-scale datasets, time series data, and image data, the solution can significantly improve the overall performance of the model and solve multiple challenges in the data processing and training process.

[0027] In one embodiment of the present invention, S4 includes: S41. Embedding the deep learning model into the wearable device; connecting the wearable device to the embedded system through wireless communication technology; S42. Design a dynamic monitoring report template for treatment efficacy, including patient basic information, monitoring data, prediction results and suggestions; use visualization techniques, such as line charts, bar charts, radar charts, etc., to intuitively display the monitoring data and prediction results; S43. Automatically generate a dynamic monitoring report for treatment efficacy based on the real-time monitoring data and the prediction results of the deep learning model, and push it to patients and doctors through the doctor-patient interaction platform or text messages.

[0028] The working principle and effects of the above technical solutions are as follows: Embedding the deep learning model into the wearable device can achieve real-time monitoring of the patient's health status. Through the combination of the embedded system and wireless communication technology, the data can be transmitted to the doctor in real time, ensuring the timeliness and accuracy of the monitoring data. The prediction ability of the deep learning model can also help doctors identify potential health problems in advance and provide timely intervention measures; the designed dynamic monitoring report template for treatment efficacy not only includes the patient's basic information, but also can combine the real-time monitoring data and the prediction results of the deep learning model to automatically generate a personalized health report. This method can provide a customized health management plan for patients, help doctors better track the treatment effect of patients, and make adjustments according to the actual situation of patients; through the application of visualization techniques such as line charts, bar charts, and radar charts, complex monitoring data and prediction results can be displayed in an intuitive and easy-to-understand way, which helps doctors quickly understand the patient's health status, and can also enhance the patient's understanding of their own health data and improve their cooperation with the treatment plan; the automatically generated dynamic monitoring report for treatment efficacy can be sent to patients and doctors in a timely manner through the doctor-patient interaction platform or text message push, which can greatly improve the efficiency of doctor-patient communication, ensure that patients can obtain necessary health information at any time during the treatment process, and doctors can quickly make corresponding adjustments or suggestions; through continuous real-time data monitoring and dynamic report push, patients can timely understand their own health status and actively cooperate with the treatment. Doctors can provide more accurate and timely medical advice based on the dynamically changing data, further improving the treatment effect and patient compliance; the function of automatically generating reports and pushing can reduce the workload of doctors, enabling them to concentrate more time and energy on the analysis of complex cases and treatment decisions, thus effectively reducing the work pressure of the hospital and improving the utilization efficiency of medical resources; the integration of the deep learning model can not only improve the accuracy of prediction, but also effectively analyze massive health data in the big data era. This technical solution demonstrates the application prospect of deep learning in the field of medical health, promotes the development of intelligent medicine, and improves the quality of medical services.

[0029] In one embodiment of the present invention, the S5 includes: S51. Based on the intelligent feedback algorithm, automatically adjust the traditional Chinese medicine treatment plan according to the prediction results and treatment goals in the dynamic monitoring report of treatment efficacy; S52. Provide decision support for intelligent feedback based on an expert system or knowledge base. The expert system or knowledge base may contain information such as the properties of traditional Chinese medicine, compatibility taboos, and dosage adjustment rules. Through the doctor-patient interaction platform, monitor data and treatment plans are shared in real time.

[0030] The working principle and effects of the above technical solution are as follows: Based on the intelligent feedback algorithm, according to the prediction results and treatment goals in the dynamic monitoring report of the treatment effect, the traditional Chinese medicine treatment plan is automatically adjusted, which not only improves the flexibility of the treatment plan but also can timely respond to the changes in the patient's condition to ensure that the patient obtains the best treatment effect. Through the automatic adjustment of the algorithm, the error of manual intervention is reduced, and the accuracy of treatment is improved; through the expert system or knowledge base, intelligent feedback can obtain effective decision support. The expert system can provide information on the properties of traditional Chinese medicine, compatibility taboos, dosage adjustment rules, etc., to help doctors make scientific and reasonable decisions in complex treatment scenarios, avoid treatment mistakes caused by human negligence, and improve the quality and safety of medical services; combined with the characteristics of traditional Chinese medicine treatment, the intelligent algorithm can make personalized adjustments to the treatment plan according to the individual differences of patients (such as constitution, condition, drug reaction, etc.), making the treatment plan more in line with the actual needs of patients, and also better improving the treatment effect and reducing side effects; through the doctor-patient interaction platform, doctors and patients can share monitor data and treatment plans in real time, enhancing the efficiency and transparency of communication. Patients can better understand their treatment process, and doctors can adjust the plan at any time according to new data, thus ensuring the dynamic adjustment of the treatment process and improving the continuity and consistency of treatment.

[0031] An embodiment of the present invention, a traditional Chinese medicine anti-tumor efficacy monitoring system based on impedance spectroscopy analysis, includes a memory, a processor, and a computer program stored on the memory and executable on the memory. The processor executes the program to implement the traditional Chinese medicine anti-tumor efficacy monitoring method based on impedance spectroscopy analysis as described in any one of the above.

[0032] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these changes and modifications.

Claims

1. A method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectroscopy analysis, characterized in that The method includes: S1. Collect bio-impedance spectrum data and perform preprocessing; S2. Extract multi-scale features from the preprocessed bio-impedance spectrum data; based on a feature fusion algorithm, organically fuse the multi-scale features to form a comprehensive and accurate feature vector; S3. Construct a hybrid model; S4. Generate a dynamic monitoring report on the curative effect; S5. Share the monitoring data and treatment plan in real time.

2. The method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectroscopy according to claim 1, wherein The said S1 includes: S11. Collect different data through a data acquisition device; S12. Record the bio-impedance spectrum data and the basic information of the patient in real time; S13. Adopt an adaptive filtering algorithm to dynamically adjust the filtering parameters according to the data characteristics.

3. The method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectroscopy according to claim 2, wherein The said S13 includes: Perform preprocessing on the collected bio-impedance spectrum data; Perform multi-scale decomposition on the signal and implement an adaptive denoising strategy for different frequency components; Dynamically adjust the filter parameters according to the real-time data to optimize the performance of the adaptive filter; Extract key bio-impedance features; Discover and correct the deviations in the data processing process through preset thresholds and monitoring indicators; Form a continuously improving cycle.

4. The method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectroscopy according to claim 1, characterized in that, The said S2 includes: S21. Extract basic time-domain features and further extract advanced time-domain features, and calculate the fractal dimension of the impedance spectrum based on fractal analysis; S22. On the basis of spectral density and coherence, add frequency-domain features, adopt wavelet transform to decompose the signal into wavelet coefficients of different scales, and extract the features at each scale; S23. Use the spatial autocorrelation function and semi-variogram to analyze the spatial distribution and correlation of bio-impedance spectrum data in different parts, and extract advanced spatial features; S24. Construct a spatial weight matrix, and based on the adjacency relationship or distance metric, evaluate the mutual influence between each part and extract spatial interaction features; S25. Transform the original bio-impedance spectrum data to generate diverse training samples; and apply a feature selection method to remove redundant and irrelevant features; adopt principal component analysis to map the high-dimensional feature space to a low-dimensional space and retain the key information; S26. In the feature extraction stage, based on a hierarchical feature fusion framework, fuse time-domain, frequency-domain, and spatial features at different levels; S27. Based on a deep learning model, perform representation learning on the features, utilize the parallel computing ability of the GPU to accelerate the feature extraction and fusion process; based on a distributed computing framework, divide the large-scale data set into small pieces and process them in parallel on multiple nodes; through feature alignment technology, align the features from different sources in time or space; S28. Use feature standardization and normalization methods to eliminate the dimensional difference and numerical range difference between features; apply mutual information to evaluate the difference between different features and identify the features with unique information; S29. Combine the feature selection method and clustering analysis to identify and remove redundant features, use random forest to evaluate the importance of each feature, and weight the features according to the importance; based on the dynamic adjustment mechanism of feature importance, dynamically adjust the feature weights according to the performance of the model on the validation set.

5. The method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectroscopy according to claim 4, wherein The said S26 includes: Determine the objectives and principles of feature fusion, and conduct preliminary classification and screening of time-domain, frequency-domain, and spatial features according to the physical meaning and biological interpretation of the features; Adopt linear or non-linear combination methods to preliminarily fuse time-domain features and frequency-domain features; Based on a deep learning network, further learn and represent the fused features to capture the complex relationships and potential structures between the features; Further fuse spatial features with time-frequency fused features; Based on the network architecture of the multi-scale spatial attention mechanism, dynamically adjust the weights of spatial features by calculating the correlations between different spatial positions.

6. The method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectroscopy according to claim 4, characterized in that, The S27 includes: Input the fused features into a deep learning model for representation learning; Use cross-validation to evaluate the performance of the model on the validation set, and utilize the parallel computing power of the GPU to accelerate the training and inference processes of the deep learning model; Aim at the characteristics of bioimpedance spectroscopy data and the requirements of the deep learning model, and improve the computing speed and resource utilization rate based on GPU parallel computing algorithms and data structures; Based on the distributed computing framework, divide the large-scale bioimpedance spectroscopy dataset into small pieces and process them in parallel on multiple nodes; Based on the deep learning training framework of data parallel processing, conduct distributed training of the model by exchanging intermediate results and gradient information between network layers or nodes; For features from different sources or different time points, adopt feature alignment techniques to align them in time or space; On the basis of feature alignment, adopt efficient feature fusion algorithms and data structures for fast feature fusion and calculation; Based on the dynamic adjustment mechanism of feature importance, dynamically adjust the weights of features in the network and the structure of the model according to the performance of the model on the validation set.

7. The method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectroscopy according to claim 1, wherein The S3 includes: S31. Construct a convolutional neural network structure; S32. Fuse the outputs of the CNN and RNN to form the final output of the hybrid model; S33. Evaluate the performance of the model; S34. Dynamically adjust the weights according to the importance of the features; S35. Continuously incorporate newly collected data to update the model parameters.

8. The method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectroscopy according to claim 1, wherein The S4 includes: S41. Embed the deep learning model into a wearable device; S42. Design a template for the dynamic monitoring report of the curative effect; S43. Automatically generate a dynamic monitoring report of the curative effect.

9. The method for monitoring the anti-tumor efficacy of traditional Chinese medicine based on impedance spectroscopy according to claim 1, wherein The S5 includes: S51. Automatically adjust the traditional Chinese medicine treatment plan; S52. Provide decision support for intelligent feedback; share monitoring data and treatment plans in real time.

10. A traditional Chinese medicine anti-tumor efficacy monitoring system based on impedance spectroscopy analysis, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the memory. The processor executes the program to implement the method for monitoring the anti-tumor curative effect of traditional Chinese medicine based on impedance spectroscopy analysis according to any one of claims 1-9.

Images