Intelligent physical sign detection and auxiliary decision-making system based on data fusion

Abstract

The invention provides an intelligent physical sign detection and auxiliary decision making system based on data fusion. The system is applied to the technical field of physical sign detection and adjuvant therapy, and comprises an acquisition and preprocessing module used for acquiring and preprocessing multi-modal human body physical sign data of a patient; the model construction and optimization module is used for constructing a data feature extraction and layer strategy combined with a random forest and a ReliefF algorithm to carry out model training, and predicting to obtain a sign detection result; the stimulating electrode adjusting module is used for dynamically adjusting the intensity, the frequency and the position of a stimulating electrode by adopting a feedback mechanism of a self-adaptive proportional-integral-differential control algorithm according to the predicted physical sign detection result; and the data visualization module is used for presenting the multi-modal human body sign data, the sign detection result and the electrode stimulation real-time condition in real time in a data visualization mode, and assisting doctors in operation and decision making. The accuracy, practicability and scientificity of the medical health detection system are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of physical sign detection and auxiliary treatment, and in particular to an intelligent physical sign detection and auxiliary decision-making system based on data fusion. Background Art

[0002] With the advancement of healthcare, vital sign detection and decision-making systems are playing an increasingly important role in disease diagnosis, treatment optimization, and health management. However, traditional vital sign detection methods have numerous limitations, such as low data processing efficiency, insufficient accuracy, and difficulty in comprehensively analyzing multimodal data. With the rapid development of wearable device technology, flexible sensors and stimulation electrodes can collect vast amounts of medical data, providing a rich source of information for vital sign detection and medical decision-making. However, how to efficiently process and analyze this data to extract valuable information and provide a reliable basis for medical decision-making has become a pressing issue.

[0003] Existing vital sign detection and decision-making support systems often only process single-modality data, making it difficult to fully reflect a patient's health status. Furthermore, due to cross-domain disparities and data scarcity in medical data, traditional machine learning algorithms, while highly effective in disease classification and prediction, often struggle with novel disease categories or when data is limited. Furthermore, traditional electrode stimulation systems lack adaptive regulation capabilities and are unable to dynamically adjust to a patient's physiological changes, thus impacting treatment effectiveness. Summary of the Invention

[0004] This invention provides an intelligent vital sign detection and decision-making support system based on data fusion. It addresses technical issues such as the difficulty of traditional methods in effectively processing and analyzing multimodal data, poor detection accuracy and precision when faced with data scarcity or cross-domain differences, and the inability of traditional electrode stimulation systems to dynamically adjust to patients' physiological changes, making it difficult to meet the personalized needs of different patients. By efficiently processing, stratifying features, intelligently analyzing, and adaptively adjusting electrodes on data collected by flexible pressure stimulation sensors, this invention improves the accuracy and practicality of medical health detection systems, enhances the precision of vital sign detection, and enhances the scientific nature of medical decision-making.

[0005] To achieve the above objectives, the following technical solutions are implemented: an intelligent vital sign detection and decision support system based on data fusion, comprising: An acquisition and preprocessing module is used to acquire and preprocess the patient's multimodal vital sign data, including heart rate, blood pressure, skin resistance, and electromyographic signals; The model building and optimization module is used to construct a data feature extraction and layer strategy that combines the random forest and ReliefF algorithms for model training, and integrates it into the cross-domain low-sample egocentric recognition task based on multimodal input and unlabeled target data to optimize the classification task of the target data. The vital sign detection results are obtained through dynamic mask generation and multiple inference integrated predictions; a stimulation electrode adjustment module for dynamically adjusting the intensity, frequency, and position of the stimulation electrodes based on the predicted vital sign detection results using a feedback mechanism of an adaptive proportional-integral-differential control algorithm; The data visualization module is used to present multimodal human vital sign data, vital sign detection results and real-time electrode stimulation status in a data visualization manner in real time to assist doctors in operation and decision-making.

[0006] Furthermore, the construction combines data feature extraction and layer strategy of random forest and ReliefF algorithms for model training, and integrates it into a cross-domain low-sample egocentric recognition task based on multimodal input and unlabeled target data to optimize the classification task of target data, and obtains vital sign detection results through dynamic mask generation and multiple inference integrated predictions, including: Construct a cross-domain low-sample learning task and divide the multimodal human body sign dataset into labeled source datasets D s and the unlabeled target dataset D Tu , target dataset D Tu It is further divided into support set S and query set Q for model training and testing; Construct a data feature extraction and stratification strategy combining random forest and ReliefF algorithms to achieve stratified screening of target data features; The data feature extraction and stratification strategy is applied to multiple data training stages, and the classification performance is optimized by combining domain adaptation with multimodal distillation, spatiotemporal cube reconstruction, and integrated mask reasoning to predict the vital sign detection results.

[0007] Furthermore, a data feature extraction and stratification strategy combining random forest and ReliefF algorithms is constructed to achieve stratified screening of target data features, including: Randomly sample multiple subsets from the original sample data, each subset is used to train a decision tree, and finally a random forest is constructed; After constructing the random forest, ReliefF is used to calculate the importance weight of the features, implement hierarchical screening of target data features, and obtain the weight of the features; According to the feature weight sorting, the feature data extracted from multimodal human vital sign data by random forest and ReliefF algorithms are divided into high-weight set, medium-weight set and low-weight set, and features are evenly sampled in each set.

[0008] Furthermore, the data feature extraction and stratification strategy is applied to multiple data training stages, and the classification performance is optimized by combining domain adaptation with multimodal distillation, spatiotemporal cube reconstruction, and integrated mask reasoning to predict the vital sign detection results, including: Define the class discriminative features consisting of the sum of the RGB features of the student encoder and the multimodal features of the teacher encoder after projection; Self-supervised pre-training via masked autoencoders, optimizing reconstruction loss and cross entropy loss; During the multimodal distillation phase, random forest and ReliefF feature screening are introduced to enable the distillation process to adaptively select the most discriminative features. This means extracting class-discriminative features from preprocessed multimodal data, optimizing the decision boundary for disease classification, and thus achieving efficient knowledge transfer and model optimization in the target domain. This includes the introduction of multimodal distillation loss and feature alignment using a teacher encoder and an RGB student encoder. The classifier is trained on the support set S. The encoder-decoder architecture is introduced into the model to extract the deep features of the data and optimize the classification loss through spatiotemporal cube reconstruction. Dynamic mask generation and multiple reasoning integration are used to predict the query set Q and output the final prediction result, namely the vital sign detection result.

[0009] Furthermore, the classifier is trained on the support set S, and an encoder-decoder architecture is introduced into the model to extract deep features of the data and optimize the classification loss through spatiotemporal cube reconstruction, including: Based on the limited labeled samples in the support set S, an encoder-decoder architecture is introduced into the model; The encoder-decoder architecture is used to extract deep features of the data and perform space-time cube reconstruction to optimize classification loss. The specific method of space-time cube reconstruction is as follows: Randomly masking some input features through spatiotemporal masks, forcing the model to reason with missing information; For the masked spatiotemporal cube data, decoder reconstruction is used for data augmentation to restore the data blocked by the mask and generate new sample variants to obtain reconstructed data, that is, to generate enhanced samples for expanding the diversity of training data; The reconstructed data is further encoded to extract features.

[0010] Furthermore, the dynamic mask generation and multiple reasoning integration are used to predict the query set Q and output the final prediction result, i.e., the physical sign detection result, including: During inference, dynamic mask generation is used to generate multiple mask data according to the set mask ratio. These mask data represent different degrees of occlusion of the query set samples. Through multiple inferences, multiple prediction results corresponding to all mask variants are obtained; The multiple prediction results corresponding to all mask variants are averaged to obtain the final prediction result.

[0011] Furthermore, the feedback control formula of the adaptive proportional-integral-differential control algorithm is: ; in, For the moment Electrode stimulation intensity at is the error, that is, the difference between the current vital sign detection value and the target vital sign value, and ; To predict the physical sign test results, is the target sign value; , , : are proportional-integral-differential control parameters, representing proportional, integral, and differential gains, which are respectively based on the predicted physical sign detection results Adaptive gain adjustment is performed based on the changes in

[0012] Furthermore, the proportional, integral, and differential gains 、 、 , according to the predicted physical sign test results The adaptive gain adjustment process is as follows: The control gain is adjusted by weighting the confidence of the vital sign data to adapt to the needs under different physiological states. The gain adjustment formula is as follows: ; ; ; in, , , are the updated proportional-integral-derivative control parameters, representing the updated proportional, integral, and differential gains respectively; , , are the proportional-integral-derivative control parameters before updating, representing the proportional, integral, and differential gains before updating respectively; , and are the confidence levels calculated for the proportional gain, integral gain, and derivative gain, respectively; 、 and For proportional gain , integral gain and differential gain The calculated confidence weighted sum; 、 and Based on the Individual sign test results Calculated proportional gain , integral gain and differential gain confidence level.

[0013] Furthermore, the data visualization module is used to present multimodal human vital sign data, vital sign detection results, and real-time electrode stimulation status in a data visualization manner in real time to assist doctors in operation and decision-making, including: The feature embedding space t-SNE method is used to display the distribution of multimodal data, the classification decision boundary is displayed through support vector machine kernel function mapping, and the visualization results are dynamically updated to assist medical decision-making.

[0014] Furthermore, the data visualization process includes: The feature embedding space t-SNE method is used to project the high-dimensional features of the class discriminant features extracted from the preprocessed multimodal data into a two-dimensional space through t-SNE. This is used to show the distribution relationship of different modal data in the embedding space and help doctors understand the basis for model decision-making; After the feature space of the class-discriminative feature F(x) is mapped by the support vector machine's kernel function, the decision boundary of the disease classification can be intuitively displayed, assisting doctors in evaluating the classification logic of the model. The kernel function mapping of the support vector machine is used to display the nonlinear decision boundary of high-dimensional data, projecting the decision boundary onto a two-dimensional plane and using different colors to identify different categories of vital sign data, helping doctors quickly identify high-risk cases. When the patient's vital signs data changes, the model immediately updates the visualization results, achieving dynamic real-time updates of the visualization results and providing the most timely clinical decision support.

[0015] Compared with the prior art, the present invention achieves the following beneficial effects: (1) Through multimodal data fusion technology, the system can integrate data from different sensors (such as heart rate, blood pressure, skin resistance, electromyography, etc.), thereby more comprehensively reflecting the patient's health status. By adopting an improved cross-domain low-sample learning method, the system can still accurately classify new categories of diseases even when there are cross-domain differences in medical data or limited data, thereby improving the accuracy of physical sign detection. Combining random forest and ReliefF algorithms for feature extraction and stratification effectively removes redundant features, improving the effectiveness and computational efficiency of features.

[0016] (2) Through dynamic mask generation and multiple inference integration, the system can improve the robustness and accuracy of the model while ensuring computational efficiency, reducing the bias and instability that may be caused by a single inference. Combined with domain adaptation and multimodal distillation strategies, the system can achieve efficient knowledge transfer and model optimization in the target domain, improving the model's adaptability in new environments.

[0017] (3) Based on the predicted vital sign detection results, the present invention uses an adaptive proportional-integral-differential control algorithm to dynamically adjust the intensity, frequency, and position of the stimulation electrodes to ensure the accuracy and safety of stimulation. By adjusting the control gain based on the confidence level of the vital sign data, the system can achieve personalized electrode stimulation adjustment based on the needs of different physiological states, thereby improving the treatment effect.

[0018] (4) This invention establishes an individualized physiological model and combines real-time vital sign data with historical data to achieve personalized adjustment of stimulation electrode parameters to meet the individual needs of different patients. It also provides an intuitive data visualization interface, enabling doctors to better understand the patient's health status and treatment effects, thereby formulating more personalized treatment plans.

[0019] It should be understood that the contents described in the summary of the invention are not intended to limit the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above and other features, advantages and aspects of the embodiments of the present invention will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. The accompanying drawings are provided for a better understanding of the present invention and do not constitute a limitation of the present invention. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, among which: Figure 1 This is a system flow chart of an intelligent vital sign detection and decision support system based on data fusion according to an embodiment of the present invention; Figure 2 This is a module diagram of an intelligent vital sign detection and decision support system based on data fusion according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] In this document, the term "and / or" simply describes a relationship between related objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, the character " / " in this document generally indicates that the related objects are in an "or" relationship.

[0023] The present invention provides an intelligent vital sign detection and decision-making support system based on data fusion. The core technologies include the following aspects: Through multimodal data fusion and improved cross-domain low-sample learning, the model is trained using labeled source data and unlabeled target data. The random forest and ReliefF algorithms are combined to divide multi-weight features. Integrated masked reasoning is performed by combining customized domain adaptation, multimodal distillation strategies, and spatiotemporal cube reconstruction. This allows for accurate classification of new disease categories even when medical data has cross-domain differences or limited data. Based on the results of vital sign detection, an improved feedback control mechanism is used to adaptively adjust the electrode stimulation intensity. Through customized gain adjustment, the electrode position, intensity, and frequency are dynamically adjusted to ensure that the electrode stimulation scheme can be optimally adjusted according to changes in vital sign data and individual physiological models. By combining multiple data visualization technologies, the complexity of high-dimensional data is transformed into an intuitive graphical dynamic display. When the patient's vital signs data changes, the model can immediately update the visualization results to assist doctors in operations and decision-making.

[0024] Figure 1 A flow chart of an intelligent vital sign detection and decision support system based on data fusion is shown; Figure 2 Figure 1 shows a module diagram of an intelligent vital sign detection and decision support system based on data fusion. Figure 1 and Figure 2 As shown, an intelligent vital sign detection and decision support system 100 based on data fusion includes: The acquisition and preprocessing module 101 is used to acquire and preprocess the patient's multimodal vital sign data. The multimodal vital sign data includes heart rate, blood pressure, skin resistance, electromyographic signals and other physiological signals, thereby forming a rich set of multimodal data streams. These data not only cover traditional vital sign monitoring parameters, but also introduce more sophisticated stimulation electrode feedback signals.

[0025] To efficiently process data from diverse devices and modalities, the present invention supports large-scale data integration and management through a data warehouse. First, through data extraction, the system acquires vital sign data from multiple data sources (including flexible sensors and stimulation electrodes) in real time or periodically. After conversion, this raw data undergoes standardization, adaptive filter-based denoising, and missing value filling to ensure data consistency and integrity.

[0026] The model building and optimization module 102 is used to construct a data feature extraction and layer strategy that combines the random forest and ReliefF algorithms for model training, and integrate it into a cross-domain low-sample egocentric recognition task based on multimodal input and unlabeled target data to optimize the classification task of the target data, and obtain the vital sign detection results through dynamic mask generation and multiple inference integrated predictions; The stimulation electrode adjustment module 103 is used to dynamically adjust the intensity, frequency and position of the stimulation electrodes according to the predicted vital sign detection results using a feedback mechanism of an adaptive proportional-integral-differential control algorithm; The data visualization module 104 is used to present multimodal human vital sign data, vital sign detection results, and real-time electrode stimulation status in a data visualization manner in real time to assist doctors in operation and decision-making.

[0027] Furthermore, a data feature extraction and layer strategy combining random forest and ReliefF algorithms is constructed for model training, and integrated into a cross-domain low-sample egocentric recognition task based on multimodal input and unlabeled target data to optimize the classification task of the target data. Vital sign detection results are obtained through dynamic mask generation and multiple inference integrated predictions, including: (1) Construct a cross-domain low-sample learning task and divide the multimodal human body sign dataset into a labeled source dataset Ds and an unlabeled target dataset D Tu , the cross-domain low-shot learning task aims to utilize the labeled source dataset Ds and the unlabeled target data D Tu Train the target dataset The new categories are classified, the data are divided by mode, and the categories of the source dataset and the target dataset have no overlap; the target dataset D Tu Further divided into support set S (including Class, each class data) and query set Q (same as support set category) for model training and testing.

[0028] On this basis, the data feature extraction and stratification strategy described later is used to optimize model training and improve classification accuracy.

[0029] (2) Construct a data feature extraction and stratification strategy that combines the random forest and ReliefF algorithms to achieve stratified screening of target data features; In order to enhance the modality alignment and feature adaptability of source data and target data, the present invention constructs a feature stratification strategy based on the random forest and ReliefF algorithms.

[0030] In the source dataset and unlabeled target datasets In the feature learning stage, this paper introduces the Bootstrap self-sampling method to randomly sample multiple subsets from the original sample data. Each subset is used to train a decision tree, and finally a random forest is constructed. After the random forest is constructed, ReliefF is used to calculate the importance weights of the features, thereby achieving hierarchical screening of the target data features: ; in, :feature The weight of : The number of samples in the query set Q; : Feature differences with samples of the same category; : Feature differences with samples of different categories; : The number of nearest neighbor samples.

[0031] According to feature weight Sorting, the feature data extracted from multimodal human vital sign data by random forest and ReliefF algorithms are divided into: high-weight set (top 30%), medium-weight set (30%-70%), and low-weight set (bottom 30%).

[0032] Features are sampled evenly within each set to avoid low-weight features from having too much influence on the model and to improve the discriminative ability of the target dataset.

[0033] (3) The data feature extraction and stratification strategy is applied to multiple data training stages, and the classification performance is optimized by combining domain adaptation with multimodal distillation, spatiotemporal cube reconstruction, and integrated mask reasoning to predict the vital sign detection results.

[0034] During the data training process, the present invention applies the above-mentioned data feature extraction and stratification strategy to multiple data training stages to optimize classification performance.

[0035] Specifically, the data feature extraction and stratification strategy is applied to multiple data training stages, and the classification performance is optimized by combining domain adaptation with multimodal distillation, spatiotemporal cube reconstruction, and integrated mask inference to predict the vital sign detection results, including: (3.1) Define a class-discriminative feature consisting of the sum of the RGB features of the student encoder and the multimodal features of the teacher encoder after projection to improve cross-domain adaptation capabilities; Adopt cross-domain low-sample learning, whose goal is to utilize source or labeled data and target domain unlabeled data , for the target domain data Perform classification. Define class discriminant features : ; in, The features extracted by the RGB Student Encoder for the input data x. The RGB Student Encoder is a model trained for disease classification. The features extracted by the teacher encoder (TeacherEncoder) for the mth modal data x. The teacher encoder is a pre-trained model used to extract effective feature representations from multimodal data; It is the Feature Projection Layer, which is used to project features from different modalities into the same feature space for subsequent fusion and classification tasks; m is the modality index, which is used to traverse all modalities in order to fuse features from different modalities; Represents the fused class discriminative features, which are obtained by fusing the features extracted by the student encoder with the multimodal features extracted by the teacher encoder. This fused feature is used for subsequent classification tasks.

[0036] (3.2) Self-supervised pre-training is performed through masked autoencoders to optimize reconstruction loss (including source data reconstruction loss and target data reconstruction loss) and cross entropy loss; In the self-supervised pre-training process based on masked autoencoders, the model is trained on the source dataset. and target dataset Perform joint training to optimize the objective function: ; Where, Reconstruct the loss for the source data; Reconstruct the loss for the target data; is the cross-extraction loss on the source data; : Trade-offs in hyperparameters.

[0037] in: ; ; ; in, is the original input data; For the reconstructed data; and The source datasets and target dataset The number of samples; For the source dataset The true label of the nth sample in (usually a one-hot encoding or a numerical representation of the class label); The model is the source dataset The predicted probability (or predicted output) of the nth sample in is usually a probability distribution, which indicates the probability that the sample belongs to each category. In the classification task, if the model is a softmax regression or similar probabilistic model, It is the value of the element in the probability vector output by the model that corresponds to the true category label (that is, the model predicts the probability that the sample belongs to the true category).

[0038] (3.3) In the multimodal distillation stage, random forest and ReliefF feature screening are introduced to enable the distillation process to adaptively select the most discriminative features. That is, class-discriminative features are extracted from the preprocessed multimodal data to optimize the decision boundary for disease classification, thereby achieving efficient knowledge transfer and model optimization in the target domain. This includes: introducing a multimodal distillation loss and using a teacher encoder and an RGB student encoder for feature alignment; In the multimodal distillation stage, this paper introduces random forest and ReliefF feature screening, which enables the distillation process to adaptively select the most discriminative features, thereby achieving efficient knowledge transfer and model optimization in the target domain. While ensuring the generalization ability of the model, the discriminative ability of the RGB student encoder in the target domain is improved, and the L2 norm is used to optimize the distillation loss: ; in, Features extracted for RGB student encoder; The real modality features extracted for the teacher encoder; is the multimodal distillation loss.

[0039] (3.4) The classifier is trained on the support set S. The encoder-decoder architecture is introduced into the model to extract the deep features of the data and optimize the classification loss through spatiotemporal cube reconstruction, including: Based on the limited labeled samples in the support set S, an encoder-decoder architecture is introduced into the model. This encoder-decoder architecture is not only used to extract deep features of the data, but also to deal with the sample scarcity problem caused by scarce labels through spatiotemporal cube reconstruction.

[0040] The specific method of space-time cube reconstruction is as follows: During training, spatiotemporal masks are used to randomly block some input features, forcing the model to reason with missing information. This randomness not only simulates the common data missing problem in practical applications, but also enables the model to explore the implicit patterns in the data as much as possible with limited information. For masked spatiotemporal cube data, decoder reconstruction is used for data augmentation. In this process, the decoder not only recovers the data blocked by the mask, but also generates new sample variants to obtain reconstructed data, that is, to generate enhanced samples, which are used to expand the diversity of training data. The reconstructed data is further extracted through the encoder and used together with the original samples to train the classifier, ensuring that the model can learn more effective features with limited samples.

[0041] Specifically, the mask reconstruction loss is optimized by the following formula: ; in, is the newly trained classifier; is the encoding feature of the support set; is the cross-picking loss.

[0042] (3.5) Dynamic mask generation and multiple reasoning integration are used to predict the query set Q and output the final prediction result, i.e., the physical sign detection result.

[0043] To reduce computational costs, the present invention balances the consumption of computing resources and the requirements of prediction accuracy through dynamic mask generation and multiple inference integration. It not only reduces the redundant information of input data through the mask mechanism, but also further enhances the robustness and accuracy of the model by integrating the prediction results of multiple mask variants.

[0044] Specifically, during inference, the present invention adopts dynamic mask generation, according to the set mask ratio Generate multiple mask data, which represent different degrees of occlusion of the query set samples. Specifically, the mask ratio This determines which features are randomly masked each time data is input, while others are retained. This approach allows the present invention to simulate the incomplete or partially missing data scenarios that may occur in the real world while maintaining computational efficiency. By performing multiple inferences, multiple predictions corresponding to all mask variations are obtained. By integrating these multiple predictions, the bias and instability that may arise from a single inference are mitigated.

[0045] The samples of query set Q are masked at the ratio generate mask data, the final prediction result is obtained by averaging the output results of all mask variants: ; in, is the final prediction result or final prediction probability; For input data processed with different masks (i.e., samples in the query set Q), during the inference process, different masks are applied to the original input data through dynamic mask generation technology to generate multiple mask variants. ; For RGB StudentEncoder, the j-th mask variant Extracted features; The number of mask variants inferred for the ensemble, i.e., the number of mask data generated dynamically; is a newly trained classifier, which is trained based on the encoder-decoder architecture and is used to classify the input data.

[0046] The final prediction result is obtained by integrating the prediction results of multiple mask variants , which represents the model's final classification result for samples in the query set Q, or serves as the basis for subsequent decisions. This formula improves the robustness and accuracy of the model by generating different mask data multiple times, performing multiple inferences, and then integrating the prediction results of multiple mask variants.

[0047] The stimulation electrode adjustment module 103 dynamically adjusts the intensity, frequency and position of the stimulation electrodes based on the predicted vital sign detection results using a feedback mechanism of an adaptive proportional-integral-differential control algorithm, specifically including: In order to achieve personalized stimulation electrode adjustment, the present invention proposes a dynamic feedback control mechanism. This mechanism automatically adjusts the intensity of the stimulation electrode based on real-time vital sign detection results, such as muscle response, neural activity or heart rate fluctuations. , ensuring the precision and safety of stimulation and maximizing the therapeutic effect.

[0048] Specifically, the physical sign test results obtained by the above prediction This data reflects an individual's health status or physiological signals, including but not limited to muscle electrical activity, heart rate variability, or other neurophysiological signals. By analyzing this data, we can monitor an individual's physiological status in real time and provide accurate feedback for adjusting the stimulation electrodes.

[0049] To achieve this goal, the present invention introduces a feedback mechanism based on an adaptive proportional-integral-differential control algorithm. This mechanism not only dynamically adjusts the intensity, frequency, and position of the stimulation electrodes based on the results of vital sign detection, but also has good response speed and flexibility. Unlike traditional proportional-integral-differential control, the control mechanism of the present invention takes into account the nonlinear characteristics and time-varying nature of individual physiological changes, making electrode adjustment more intelligent and personalized. The feedback control formula is as follows: ; in, For the moment Electrode stimulation intensity at is the error, that is, the difference between the current vital sign detection value and the target vital sign value, and ; For the predicted physical sign test results (such as heart rate, blood pressure, electromyographic signals, neural activity, etc.), is the target physical sign value or the expected physical sign value; , , : are proportional-integral-differential control parameters, representing proportional, integral, and differential gains, which are respectively based on the predicted physical sign detection results Adaptive gain adjustment is performed based on the changes in

[0050] Through real-time error feedback, the intensity of electrode stimulation can be adjusted based on real-time changes in individual physiological signals. For example, when heart rate fluctuates significantly, the control system automatically reduces stimulation intensity to avoid overstimulation. Similarly, when muscle response becomes fatigued or unstable, the system automatically adjusts frequency and intensity to avoid side effects caused by overstimulation.

[0051] In order to improve the sensitivity of the adjustment, , , According to Adaptive gain adjustment (incorporating dynamic weights) is performed based on changes in the vital sign data. The control gain is weighted by the confidence level of the vital sign data to adapt to the needs of different physiological states. For example, when the vital sign signal fluctuates greatly or there is uncertainty, the system automatically reduces the gain value to reduce the risk of over-adjustment. When the signal is stable, the gain will automatically increase to ensure more precise stimulation adjustment. The gain adjustment formula is as follows: ; ; ; in, , , are the updated proportional-integral-derivative control parameters, representing the updated proportional, integral, and differential gains respectively; , , are the proportional-integral-derivative control parameters before updating, representing the proportional, integral, and differential gains before updating respectively; , and are the confidence levels calculated for the proportional gain, integral gain, and derivative gain, respectively; 、 and For proportional gain , integral gain and differential gain The calculated confidence weighted sum; 、 and Based on the Individual sign test results Calculated proportional gain , integral gain and differential gain confidence level.

[0052] The optimization of electrode parameters, i.e., the dynamic adjustment of the intensity, frequency and position of the stimulation electrodes, is as follows: By establishing an individualized physiological model, the system can simulate the response of different physiological signals to electrode stimulation and make predictive adjustments. This physiological model can not only be trained based on an individual's historical data, but can also be optimized in conjunction with real-time vital signs, thereby providing a more personalized and precise electrode adjustment solution. By combining real-time vital sign data and the physiological model, multiple electrode parameters are adjusted through gradient optimization to ensure that electrode stimulation maintains optimal results throughout the treatment process: ; , : represent the electrode parameters (position, intensity, frequency) before and after updating respectively; : learning rate; : Electrode parameter optimization loss function and physical sign detection results Related: ; , :Represents the Predicted results and expected sign values of individual sign tests.

[0053] The data visualization module 104 is used to present multimodal human vital sign data, vital sign detection results, and real-time electrode stimulation status in a data visualization manner in real time to assist doctors in operation and decision-making, including: The feature embedding space t-SNE method is used to display the distribution of multimodal data, the classification decision boundary is displayed through support vector machine kernel function mapping, and the visualization results are dynamically updated to assist medical decision-making.

[0054] t-SNE (t-Distributed Stochastic Neighbor Embedding) is a nonlinear dimensionality reduction technique particularly suitable for visualizing high-dimensional data. It projects high-dimensional data into a lower-dimensional space (usually two or three dimensions) while preserving both the local and global structure between data points.

[0055] The discriminative feature F(x) is a feature representation derived from the source and target data through cross-domain low-sample learning. It contains key information for classification. In the present invention, F(x) is a high-dimensional feature vector that includes features from different modalities (such as heart rate, blood pressure, skin resistance, and electromyography). The feature representation obtained by projecting F(x) into two-dimensional space using the t-SNE method is called an embedded feature. This projection process preserves the key structural information in F(x), allowing doctors to intuitively observe the distribution relationship between data from different modalities on a two-dimensional plane.

[0056] A support vector machine (SVM) is a binary classification model that separates data points of different categories by finding a hyperplane and maximizing the distance (i.e., margin) between this hyperplane and the nearest data point. For nonlinearly separable data, the SVM can map the data into a high-dimensional space by introducing a kernel function, thereby finding a linearly separable hyperplane. In this paper, the kernel function of the support vector machine (such as the radial basis function (RBF)) is used to map the class-discriminative features F(x) into a high-dimensional space. This mapping process preserves the nonlinear structural information in F(x), enabling the SVM to learn more complex decision boundaries.

[0057] Furthermore, the data visualization process of the data visualization module 104 includes: The feature embedding space t-SNE method is used to project the high-dimensional features of the class discriminant features extracted from the preprocessed multimodal data into a two-dimensional space through t-SNE. This is used to show the distribution relationship of different modal data in the embedding space and help doctors understand the basis for model decision-making; Optionally, the specific process of t-SNE projection includes: (1) Data preparation: Extract the class discriminant feature F(x) from the classification model as the input data of t-SNE. (2) Parameter setting: Set the parameters of t-SNE, such as perplexity, number of iterations, learning rate, etc. Perplexity is a key parameter that controls the balance between t-SNE in preserving local structure and global structure. (3) Run t-SNE: Input F(x) into the t-SNE algorithm to obtain a two-dimensional embedding feature. (4) Visualization: Plot the two-dimensional embedding feature on a two-dimensional plane, and represent different categories of data points with different colors or shapes. Doctors can understand the relationship between different modal data and the model's ability to distinguish different categories by observing the distribution of these data points.

[0058] After the feature space of the class-discriminative feature F(x) is mapped by the support vector machine's kernel function, the decision boundary of the disease classification can be intuitively displayed, assisting doctors in evaluating the classification logic of the model. The kernel function mapping of the support vector machine is used to display the nonlinear decision boundary of high-dimensional data, projecting the decision boundary onto a two-dimensional plane and using different colors to identify different categories of vital sign data, helping doctors quickly identify high-risk cases. Optionally, the visualization process of the classification boundary is as follows: (1) Data preparation: Similar to t-SNE, extract the class discriminative feature F(x) from the classification model. (2) Training SVM: Use F(x) as input data to train an SVM model. During the training process, select appropriate kernel functions and parameters (such as penalty parameter C, kernel function parameter γ, etc.). (3) Decision boundary projection: Project the decision boundary of the SVM onto a two-dimensional plane. This is usually achieved by selecting two feature dimensions (such as the two dimensions after t-SNE projection) and calculating the decision boundary on these dimensions. (4) Visualization: Plot the data points and decision boundaries of different categories on a two-dimensional plane, using different colors to identify data points of different categories. Doctors can evaluate the classification logic and performance of the model by observing the position and shape of the decision boundary.

[0059] When patient vital signs change, the model instantly updates the visualization, achieving dynamic, real-time updates and providing timely clinical decision support. Specifically, the new F(x) is input into the t-SNE and SVM models to generate new two-dimensional embedding features and decision boundaries. The visualization interface is then updated to display the latest data distribution and decision boundaries. Physicians can adjust and optimize treatment plans based on the updated visualization results. Furthermore, the system can further train and optimize the model based on physician feedback.

[0060] According to the above-mentioned embodiments of the present invention, comprehensive analysis and accurate classification of multimodal data are achieved through multimodal data fusion and cross-domain low-sample learning technology; the robustness and accuracy of the system are improved through dynamic mask generation and multiple inference integration; the adaptive proportional-integral-differential control algorithm is adopted to achieve adaptive adjustment of electrode stimulation intensity, which can meet the patient's personalized treatment needs; and by combining multiple data visualization technologies, intuitive and dynamic data visualization support is provided.

[0061] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present invention can be achieved. This is not limited herein.

[0062] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. An intelligent vital sign detection and decision support system based on data fusion, characterized in that: include: An acquisition and preprocessing module is used to acquire and preprocess the patient's multimodal vital sign data, including heart rate, blood pressure, skin resistance, and electromyographic signals; The model building and optimization module is used to construct a data feature extraction and layer strategy that combines the random forest and ReliefF algorithms for model training, and integrates it into the cross-domain low-sample egocentric recognition task based on multimodal input and unlabeled target data to optimize the classification task of the target data. The vital sign detection results are obtained through dynamic mask generation and multiple inference integrated predictions; a stimulation electrode adjustment module for dynamically adjusting the intensity, frequency, and position of the stimulation electrodes based on the predicted vital sign detection results using a feedback mechanism of an adaptive proportional-integral-differential control algorithm; The data visualization module is used to present multimodal human vital sign data, vital sign detection results and real-time electrode stimulation status in a data visualization manner in real time to assist doctors in operation and decision-making.

2. The intelligent vital sign detection and decision support system based on data fusion according to claim 1 is characterized in that: in, The proposed model combines data feature extraction and layer strategies of the random forest and ReliefF algorithms for model training, and integrates them into a cross-domain low-sample egocentric recognition task based on multimodal input and unlabeled target data to optimize the classification task of the target data. Vital sign detection results are obtained through dynamic mask generation and multiple inference ensemble predictions, including: Construct a cross-domain low-sample learning task and divide the multimodal human body sign dataset into labeled source datasets D s and the unlabeled target dataset D Tu , target dataset D Tu It is further divided into support set S and query set Q for model training and testing; Construct a data feature extraction and stratification strategy combining random forest and ReliefF algorithms to achieve stratified screening of target data features; The data feature extraction and stratification strategy is applied to multiple data training stages, and the classification performance is optimized by combining domain adaptation with multimodal distillation, spatiotemporal cube reconstruction, and integrated mask reasoning to predict the vital sign detection results.

3. The intelligent vital sign detection and decision support system based on data fusion according to claim 2 is characterized in that: in, Construct a data feature extraction and stratification strategy that combines the random forest and ReliefF algorithms to achieve stratified screening of target data features, including: Randomly sample multiple subsets from the original sample data, each subset is used to train a decision tree, and finally a random forest is constructed; After constructing the random forest, ReliefF is used to calculate the importance weight of the features, implement hierarchical screening of target data features, and obtain the weight of the features; According to the feature weight sorting, the feature data extracted from multimodal human vital sign data by random forest and ReliefF algorithms are divided into high-weight set, medium-weight set and low-weight set, and features are evenly sampled in each set.

4. The intelligent vital sign detection and decision support system based on data fusion according to claim 2 is characterized in that: in, The data feature extraction and stratification strategy is applied to multiple data training stages. Domain adaptation and multimodal distillation, spatiotemporal cube reconstruction, and integrated mask inference are combined to optimize classification performance and predict vital sign detection results, including: Define the class discriminative features consisting of the sum of the RGB features of the student encoder and the multimodal features of the teacher encoder after projection; Self-supervised pre-training via masked autoencoders, optimizing reconstruction loss and cross entropy loss; During the multimodal distillation phase, random forest and ReliefF feature screening are introduced to enable the distillation process to adaptively select the most discriminative features. This means extracting class-discriminative features from preprocessed multimodal data, optimizing the decision boundary for disease classification, and thus achieving efficient knowledge transfer and model optimization in the target domain. This includes the introduction of multimodal distillation loss and feature alignment using a teacher encoder and an RGB student encoder. The classifier is trained on the support set S. The encoder-decoder architecture is introduced into the model to extract the deep features of the data and optimize the classification loss through spatiotemporal cube reconstruction. Dynamic mask generation and multiple reasoning integration are used to predict the query set Q and output the final prediction result, namely the vital sign detection result.

5. The intelligent vital sign detection and decision support system based on data fusion according to claim 4 is characterized in that: in, The classifier is trained on the support set S. The encoder-decoder architecture is introduced into the model to extract deep features of the data and optimize the classification loss through spatiotemporal cube reconstruction, including: Based on the limited labeled samples in the support set S, an encoder-decoder architecture is introduced into the model; The encoder-decoder architecture is used to extract deep features of the data and perform space-time cube reconstruction to optimize classification loss. The specific method of space-time cube reconstruction is as follows: Randomly masking some input features through spatiotemporal masks, forcing the model to reason with missing information; For the masked spatiotemporal cube data, decoder reconstruction is used for data augmentation to restore the data blocked by the mask and generate new sample variants to obtain reconstructed data, that is, to generate enhanced samples for expanding the diversity of training data; The reconstructed data is further encoded to extract features.

6. The intelligent vital sign detection and decision support system based on data fusion according to claim 4 or 5, characterized in that: in, The dynamic mask generation and multiple reasoning integration are used to predict the query set Q and output the final prediction result, i.e., the physical sign detection result, including: During inference, dynamic mask generation is used to generate multiple mask data according to the set mask ratio. These mask data represent different degrees of occlusion of the query set samples. Through multiple inferences, multiple prediction results corresponding to all mask variants are obtained; The multiple prediction results corresponding to all mask variants are averaged to obtain the final prediction result.

7. The intelligent vital sign detection and decision support system based on data fusion according to claim 6 is characterized in that: in, The feedback control formula of the adaptive proportional-integral-differential control algorithm used is: ； in, For the moment Electrode stimulation intensity at is the error, that is, the difference between the current vital sign detection value and the target vital sign value, and ; To predict the physical sign test results, is the target sign value; , , : are proportional-integral-differential control parameters, representing proportional, integral, and differential gains, which are respectively based on the predicted physical sign detection results Adaptive gain adjustment is performed based on the changes in 8. The intelligent vital sign detection and decision support system based on data fusion according to claim 7 is characterized in that: in, Proportional, integral, and differential gains , , , according to the predicted physical sign test results The adaptive gain adjustment process is as follows: The control gain is adjusted by weighting the confidence of the vital sign data to adapt to the needs under different physiological states. The gain adjustment formula is as follows: ； ； ； in, , , are the updated proportional-integral-derivative control parameters, representing the updated proportional, integral, and differential gains respectively; , , are the proportional-integral-derivative control parameters before updating, representing the proportional, integral, and differential gains before updating respectively; , and are the confidence levels calculated for the proportional gain, integral gain, and derivative gain, respectively; 、 and For proportional gain , integral gain and differential gain The calculated confidence weighted sum; 、 and Based on the Individual sign test results Calculated proportional gain , integral gain and differential gain confidence level.

9. The intelligent vital sign detection and decision support system based on data fusion according to claim 4 is characterized in that: in, The data visualization module is used to present multimodal human vital sign data, vital sign detection results, and real-time electrode stimulation status in a data visualization manner in real time to assist doctors in operation and decision-making, including: The feature embedding space t-SNE method is used to display the distribution of multimodal data, the classification decision boundary is displayed through support vector machine kernel function mapping, and the visualization results are dynamically updated to assist medical decision-making.

10. The intelligent vital sign detection and decision support system based on data fusion according to claim 9, characterized in that: in, The data visualization process includes: The feature embedding space t-SNE method is used to project the high-dimensional features of the class discriminant features extracted from the preprocessed multimodal data into a two-dimensional space through t-SNE. This is used to show the distribution relationship of different modal data in the embedding space and help doctors understand the basis for model decision-making; After the feature space of the class-discriminative feature F(x) is mapped by the support vector machine's kernel function, the decision boundary of the disease classification can be intuitively displayed, assisting doctors in evaluating the classification logic of the model. The kernel function mapping of the support vector machine is used to display the nonlinear decision boundary of high-dimensional data, projecting the decision boundary onto a two-dimensional plane and using different colors to identify different categories of vital sign data, helping doctors quickly identify high-risk cases. When the patient's vital signs data changes, the model immediately updates the visualization results, achieving dynamic real-time updates of the visualization results and providing the most timely clinical decision support.

Images