Humidification oxygen therapy parameter optimization assisting method and system based on attention mechanism

Abstract

The invention discloses a humidification oxygen therapy parameter optimization auxiliary method and system based on an attention mechanism, and relates to the technical field of medical optimization. According to the method, respiration physiological parameters and equipment operation parameters in the humidification oxygen therapy process are synchronously collected through a medical sensor array, and the parameters are preprocessed in a layered mode to obtain a standardized feature matrix; inputting the parameters into an attention mechanism neural network, distributing dynamic weights through a parameter attention layer to obtain a parameter weighted feature matrix, strengthening time sequence features through a time sequence attention layer, extracting key breathing state features, and generating optimized feature vectors; and synchronously calculating oxygenation efficiency, breathing load and airway comfort based on the vector, and outputting a treatment effect optimization result and a parameter optimization suggestion. The method can accurately optimize the treatment state and dynamically optimize the parameters, is especially suitable for scenes such as vertebroplasty, and improves the treatment safety and the comfort level of a patient.

Description

Technical Field

[0001] This invention proposes an auxiliary method and system for optimizing humidified oxygen therapy parameters based on attention mechanisms, belonging to the field of equipment data technology. Background Technology

[0002] In the field of medical optimization technology, humidified oxygen therapy, as a commonly used respiratory support method, is widely applied in clinical scenarios such as vertebroplasty. Its core requirement is to ensure patient safety and comfort through precise control of oxygen therapy parameters. Currently, optimization of humidified oxygen therapy often employs traditional fixed threshold optimization methods. This involves pre-setting fixed physiological and equipment parameter thresholds, issuing warnings only when parameters exceed these thresholds. This approach cannot dynamically adapt to changes in the patient's physiological state during treatment. For example, fluctuations in respiratory rate and blood oxygen saturation due to postural adjustments and stress responses during surgery are difficult for traditional optimization systems to capture in real time and adjust parameters accordingly. Furthermore, existing technologies lack the ability to deeply process optimization data, often optimizing only a single parameter without correlating respiratory physiological parameters with equipment operating parameters. This results in an inability to comprehensively assess the actual oxygen therapy effect on the patient, easily leading to parameter control lags. This can potentially cause excessively high or low oxygen concentrations, insufficient humidification, etc., affecting treatment outcomes and even increasing the risk of patient complications. In addition, although some optimization systems have certain data processing capabilities, they have not introduced intelligent feature extraction and weight allocation mechanisms, making it impossible to accurately identify key parameters that affect treatment effects. This results in parameter optimization lacking specificity and failing to meet the clinical demand for personalized and precise oxygen therapy optimization. Summary of the Invention

[0003] This invention provides an attention-based method and system for optimizing humidified oxygen therapy parameters, addressing the technical problems of insufficient multi-parameter correlation, inaccurate identification of key parameters, lag in response to time-series changes, and lack of targeted parameter optimization in existing humidified oxygen therapy optimization techniques. The technical solution adopted is as follows: An auxiliary method for optimizing humidified oxygen therapy parameters based on attention mechanisms includes the following steps: Multiple respiratory physiological parameters and equipment operating parameters during humidified oxygen therapy are collected synchronously using a medical sensor array. The parameters are subjected to hierarchical preprocessing to obtain a standardized feature matrix; The standardized feature matrix is ​​input into a preset attention mechanism neural network, and dynamic weights are assigned to different optimization parameters through a parameter attention layer to obtain a parameter-weighted feature matrix; The parameter weighted feature matrix is ​​enhanced by a temporal attention layer to extract key respiratory state features and generate a humidified oxygen therapy treatment state feature vector. Based on the humidified oxygen therapy treatment status feature vector, oxygenation efficiency, respiratory load, and airway comfort are calculated simultaneously, and a humidified oxygen therapy recommendation report is output.

[0004] Furthermore, the synchronous acquisition of multiple parameters via a medical sensor array also includes the following specific steps: The multi-channel optimization system is activated, and the temperature of the inhaled gas is collected in real time through the nasopharyngeal temperature sensor. After being processed by the signal calibration module, the gas is collected at a preset frequency. The system acquires inhaled oxygen concentration data through an oxygen concentration optimization module, collects actual airflow velocity through a flow sensor, collects blood oxygen saturation data through a pulse oximeter, records respiratory rate through a respiratory rate optimizer, and acquires end-tidal carbon dioxide partial pressure data through a carbon dioxide optimization module.

[0005] Furthermore, obtaining the parameter-weighted feature matrix also includes the following steps: By optimizing the embedded processing unit of the host computer, an adaptive filtering algorithm is used to denoise the respiratory rate signal, extract respiratory cycle feature parameters, and analyze the irregularity of the respiratory waveform using wavelet transform to obtain preprocessed respiratory feature data. A multi-parameter preprocessing unit then uses a Z-score normalization algorithm to process temperature, oxygen concentration, and flow rate data. The Z-score normalization formula is as follows: in, For the original data points, This is the mean of this type of parameter. Standard deviation The data is normalized; the dynamic trend of blood oxygen saturation is extracted through the time series analysis module, and the minimum value and fluctuation range of blood oxygen saturation within a set time period are calculated using the sliding window algorithm to obtain the dynamic characteristic data of oxygenation; the above data is uniformly converted into a standardized feature matrix through the matrix integration unit. in, As a preset time dimension, The preset number of features.

[0006] Furthermore, obtaining the parameter-weighted feature matrix also includes the following specific steps: The standardized feature matrix is ​​input into the first fully connected layer to obtain the initial weight vector of parameters. The current oxygenation state is determined based on the blood oxygen saturation level and respiratory rate, and the corresponding basic weight matrix is ​​called. The initial parameter weight vector is multiplied by the basic weight matrix and then normalized by Softmax to obtain the parameter attention weights. The weights of oxygen concentration and flow rate under hypoxia and respiratory rate and airway pressure under high respiratory load are emphasized. The parameter attention weights are then multiplied element-wise by the original feature matrix to generate the parameter weighted feature matrix.

[0007] Furthermore, the extraction of key respiratory state features through the temporal attention layer also includes: The parameter-weighted feature matrix is ​​input into a bidirectional LSTM layer to extract the temporal feature matrix. The temporal feature matrix is ​​then subjected to global average pooling and input into the second fully connected layer to generate temporal attention weights, focusing on key time nodes such as a sudden increase in respiratory rate and a sudden decrease in blood oxygen saturation. Multiply the temporal attention weights by the temporal feature matrix, and then perform global max pooling to obtain the optimized feature vector of humidified oxygen therapy treatment status.

[0008] In some embodiments, during the extraction of key respiratory state features through a temporal attention layer, the parameter-weighted feature matrix obtained after parameter attention processing is first input into a bidirectional Long Short-Term Memory (LSTM) layer. The bidirectional LSTM layer has the ability to capture temporal correlations of data simultaneously from past to future and from future to past, comprehensively mining the hidden temporal dimension information in the parameter-weighted feature matrix, such as the changing trends of different respiratory parameters with the progress of surgery and the dynamic coupling relationships between parameters, ultimately outputting a temporal feature matrix covering multi-dimensional temporal features. After obtaining the temporal feature matrix, a global average pooling operation is first performed on it. This operation reduces the data dimensionality while preserving the overall temporal trend by calculating the average value of the temporal feature matrix in the time dimension, avoiding interference from outliers at single time points in subsequent processing, and providing a more stable feature basis for the generation of subsequent attention weights. Subsequently, the pooled features are input into a second fully connected layer, which learns the importance of different temporal features for respiratory state assessment through nonlinear transformation, thereby generating corresponding temporal attention weights. During the weight learning process, the model automatically emphasizes key time points such as sudden increases in respiratory rate and sudden drops in blood oxygen saturation, making the temporal attention weights for these points significantly higher than those for regular time points. This ensures that subsequent feature extraction focuses on moments that decisively determine the respiratory status. After generating the temporal attention weights, they are multiplied element-wise with the temporal feature matrix. Through weight allocation, key temporal features are strengthened while non-key features are suppressed, highlighting the feature information of moments with abnormal respiratory status. Finally, a global max pooling operation is performed on the multiplied feature matrix to select the maximum value for each feature channel. These maximum values ​​are then integrated into a lower-dimensional, more information-concentrated humidified oxygen therapy treatment status optimization feature vector. This vector not only contains key respiratory status features but also accurately reflects the impact of different time points on the overall oxygen therapy status, providing core data support for subsequent treatment effect calculations and parameter optimization suggestions.

[0009] Furthermore, it also includes the model training steps: Clinical data on the application of humidified oxygen therapy in vertebroplasty were collected and divided into training, validation and test sets. Data augmentation was performed by adding respiratory parameter fluctuation noise and time offset. A multi-task loss function was constructed, taking into account oxygenation effect, breathing comfort and equipment parameter matching. The AdamW optimizer was used to train the attention mechanism neural network. Training was stopped when the validation set loss no longer decreased, and the model was deployed to the intraoperative optimization AI unit.

[0010] In some embodiments, during the model training phase, clinical data on the application of humidified oxygen therapy in vertebroplasty surgery are collected. This data includes key information such as the patient's intraoperative respiratory physiological parameters, the operating parameters of the humidified oxygen therapy equipment, and corresponding treatment effect feedback. After collection, the data is divided into training, validation, and test sets according to clinical data processing standards. The division process fully considers the diversity and representativeness of the data, ensuring that each dataset reflects the intraoperative oxygen therapy status of different patients. To improve the model's generalization ability, data augmentation is also required. Specifically, respiratory parameter fluctuation noise is added to simulate the natural fluctuations in parameters caused by changes in patient position and stress responses during surgery. Simultaneously, time-series offset is introduced to simulate slight time synchronization deviations that may occur during data acquisition by the equipment, making the training data more closely resemble actual clinical scenarios.

[0011] Furthermore, a multi-task loss function is constructed, which comprehensively incorporates three core dimensions: oxygenation effect, respiratory comfort, and equipment parameter matching degree. Oxygenation effect is related to key physiological indicators such as patient blood oxygen saturation; respiratory comfort refers to feedback data such as patient airway pressure and respiratory rate; and equipment parameter matching degree focuses on the suitability of equipment output oxygen concentration and humidification temperature to patient needs. Through reasonable weight allocation, the function can comprehensively measure the difference between the model's predicted results and the ideal clinical state. During training, the AdamW optimizer is used to update the parameters of the attention mechanism neural network. The optimizer effectively balances training speed and convergence stability, avoiding overfitting. During training, the validation set loss is continuously monitored. When the validation set loss no longer decreases after several rounds of training, the model is considered to have reached its optimal training state, and training is stopped. The final optimized model is then deployed to the intraoperative optimization AI unit, enabling it to provide real-time support for adjusting humidification oxygenation parameters during actual vertebroplasty surgery.

[0012] Furthermore, the simultaneous calculation of oxygenation efficiency, respiratory load, and airway comfort also includes the following specific steps: calculating the probability distribution of oxygenation adequacy through the fully connected layer, taking the maximum value as the oxygenation efficiency rating result, regressing to calculate the predicted value of breathing work as the respiratory load index, and combining temperature and humidity parameters to calculate the airway mucosal comfort score; updating and optimizing the results in real time, triggering an early warning when blood oxygen saturation or respiratory rate is abnormal, and automatically generating parameter adjustment suggestions.

[0013] In some embodiments, when simultaneously calculating oxygenation efficiency, respiratory load, and airway comfort, the humidified oxygen therapy treatment status monitoring feature vector obtained by the monitoring system is first input into a pre-defined fully connected layer. The fully connected layer calculates an oxygenation adequacy probability distribution based on key parameters such as blood oxygen saturation and oxygen concentration in the feature vector through nonlinear transformation. This distribution covers probability values ​​corresponding to different oxygenation states. The maximum value in the distribution is then selected and mapped to the corresponding oxygenation efficiency rating, such as "Excellent," "Good," or "Needs Optimization," thus intuitively reflecting the current oxygenation support effect of oxygen therapy on the patient. For the calculation of respiratory load, a regression analysis method is used, with parameters such as airway pressure and respiratory flow rate in the monitoring feature vector as independent variables. These are substituted into a pre-trained work of breathing prediction model, and regression calculation is performed using the following formula: .

[0014] In the formula, , These are the model training weight coefficients corresponding to airway pressure and respiratory flow rate, respectively. The predicted work of breathing is calculated for the bias term. This value directly corresponds to the respiratory load index; a higher value indicates that the patient needs to consume more energy to breathe, resulting in a greater respiratory load. For airway comfort scoring, humidification temperature and relative humidity parameters are extracted from the monitoring data and combined with a clinically validated comfort assessment model. After converting the temperature and humidity parameters into standardized scores, a weighted sum is used to obtain the airway mucosal comfort score. A higher score indicates that the airway mucosa is in a more suitable humidified environment, resulting in better patient comfort. After completing the calculation of the above three indicators, the system updates the optimization results in real time and displays them on the intraoperative monitoring host interface. Simultaneously, the system continuously compares real-time data of blood oxygen saturation and respiratory rate with normal threshold ranges. When blood oxygen saturation exceeds the normal range or respiratory rate fluctuates abnormally, an early warning mechanism is immediately triggered, notifying medical staff through audio and visual prompts. Based on the degree to which the current indicators deviate from the normal range and combined with preset parameter adjustment rules, the system automatically generates specific parameter adjustment suggestions, such as adjusting oxygen concentration and humidification temperature, providing accurate references for medical staff to quickly optimize oxygen therapy plans.

[0015] Secondly, the present invention also provides an attention-based humidification oxygen therapy parameter optimization assistance system, comprising: The parameter acquisition module includes a medical-grade sensor array and a humidified oxygen therapy device interface, used to simultaneously acquire respiratory physiological parameters and device operating parameters; The preprocessing module, connected to the parameter acquisition module, is used to perform hierarchical preprocessing on the parameters to obtain a standardized feature matrix; The parameter attention processing module is equipped with a parameter attention layer in a preset attention mechanism neural network and is connected to the preprocessing module. It is used to assign dynamic weights to different optimization parameters to obtain a parameter-weighted feature matrix. The temporal feature enhancement module is equipped with a temporal attention layer in a preset attention mechanism neural network and is connected to the parameter attention processing module. It is used to extract key respiratory state features and obtain an optimized feature vector of humidified oxygen therapy treatment state. The optimization module includes a multi-task output unit of the intraoperative optimization host, which is connected to the temporal feature enhancement module and is used to calculate the treatment effect and output parameter optimization suggestions.

[0016] Thirdly, the present invention also provides an attention-based humidified oxygen therapy parameter optimization auxiliary device, comprising: one or more processors; a memory; one or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, the programs comprising instructions for performing the attention-based humidified oxygen therapy parameter optimization auxiliary method.

[0017] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the attention mechanism-based humidification oxygen therapy parameter optimization assistance method.

[0018] Beneficial effects of this invention: The attention-based humidified oxygen therapy parameter optimization method, system, device, and storage medium provided by this invention effectively solves many problems existing in the prior art, bringing significant benefits in multiple aspects. First, by synchronously collecting respiratory physiological parameters and device operating parameters through a medical sensor array, it breaks through the limitations of independent optimization of single parameters in traditional optimization, achieving comprehensive acquisition of multi-dimensional data. This lays the foundation for subsequent accurate analysis, avoids optimization blind spots caused by data gaps, and ensures real-time monitoring of the patient's overall oxygen therapy status. Second, by introducing an attention mechanism neural network, dynamic weights are assigned to different parameters through a parameter attention layer. This allows for precise identification of key parameters that play a crucial role in oxygen therapy effectiveness, such as the patient's blood oxygen saturation and airway pressure during surgery. Compared to traditional fixed-weight optimization methods, this approach more specifically focuses on key data, improves the accuracy of parameter analysis, and avoids judgment biases caused by invalid data interference. Furthermore, the temporal attention layer enhances the extraction of temporal features of parameters, enabling real-time capture of dynamic changes in the patient's physiological state and equipment operating parameters. For example, during vertebroplasty, changes in the patient's respiratory rhythm due to procedural stimuli can be quickly responded to and corresponding optimized feature vectors generated. This achieves real-time parameter control, solving the problem of lag in traditional optimization systems and effectively reducing the risk of complications caused by parameter mismatch. Finally, based on the optimized feature vectors, oxygenation efficiency, respiratory load, and airway comfort are calculated simultaneously. This allows for the evaluation of oxygen therapy effectiveness from multiple dimensions, including efficacy and patient tolerance, and provides precise parameter optimization suggestions. This meets the needs of personalized and precise clinical treatment, especially in scenarios with high requirements for oxygen therapy stability, such as vertebroplasty, significantly improving treatment safety and patient comfort, and providing more reliable technical support for clinical respiratory support therapy. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method described in this invention; Figure 2 This is a system block diagram of the system described in this invention; Figure 3 This is a schematic diagram of the structure of the device described in this invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] refer to Figure 1 The present invention proposes an auxiliary method for optimizing humidified oxygen therapy parameters based on attention mechanisms, comprising the following steps: S1. Synchronously collect various respiratory physiological parameters and equipment operating parameters during humidified oxygen therapy treatment through a medical sensor array; S2. Perform hierarchical preprocessing on the parameters to obtain a standardized feature matrix; S3. Input the standardized feature matrix into a preset attention mechanism neural network, and assign dynamic weights to different optimization parameters through a parameter attention layer to obtain a parameter-weighted feature matrix; S4. The parameter weighted feature matrix is ​​enhanced by a temporal attention layer to extract key respiratory state features and generate a humidified oxygen therapy treatment state feature vector. S5. Based on the humidified oxygen therapy treatment status feature vector, simultaneously calculate oxygenation efficiency, respiratory load, and airway comfort, and output a humidified oxygen therapy recommendation report.

[0022] Preferably, the synchronous acquisition of multiple parameters via a medical sensor array further includes the following specific steps: The multi-channel optimization system is activated, and the temperature of the inhaled gas is collected in real time through the nasopharyngeal temperature sensor. After being processed by the signal calibration module, the gas is collected at a preset frequency. The system acquires inhaled oxygen concentration data through an oxygen concentration optimization module, collects actual airflow velocity through a flow sensor, collects blood oxygen saturation data through a pulse oximeter, records respiratory rate through a respiratory rate optimizer, and acquires end-tidal carbon dioxide partial pressure data through a carbon dioxide optimization module.

[0023] Preferably, obtaining the parameter-weighted feature matrix further includes the following steps: By optimizing the embedded processing unit of the host computer, an adaptive filtering algorithm is used to denoise the respiratory rate signal, extract respiratory cycle feature parameters, and analyze the irregularity of the respiratory waveform using wavelet transform to obtain preprocessed respiratory feature data. A multi-parameter preprocessing unit then uses a Z-score normalization algorithm to process temperature, oxygen concentration, and flow rate data. The Z-score normalization formula is as follows: in, For the original data points, This is the mean of this type of parameter. Standard deviation The data is normalized; the dynamic trend of blood oxygen saturation is extracted through the time series analysis module, and the minimum value and fluctuation range of blood oxygen saturation within a set time period are calculated using the sliding window algorithm to obtain the dynamic characteristic data of oxygenation; the above data is uniformly converted into a standardized feature matrix through the matrix integration unit. in, As a preset time dimension, The preset number of features.

[0024] Preferably, obtaining the parameter-weighted feature matrix further includes the following specific steps: The standardized feature matrix is ​​input into the first fully connected layer to obtain the initial weight vector of parameters. The current oxygenation state is determined based on the blood oxygen saturation level and respiratory rate, and the corresponding basic weight matrix is ​​called. The initial parameter weight vector is multiplied by the basic weight matrix and then normalized by Softmax to obtain the parameter attention weights. The weights of oxygen concentration and flow rate under hypoxia and respiratory rate and airway pressure under high respiratory load are emphasized. The parameter attention weights are then multiplied element-wise by the original feature matrix to generate the parameter weighted feature matrix.

[0025] Preferably, the extraction of key respiratory state features through the temporal attention layer further includes: The parameter-weighted feature matrix is ​​input into a bidirectional LSTM layer to extract the temporal feature matrix. The temporal feature matrix is ​​then subjected to global average pooling and input into the second fully connected layer to generate temporal attention weights, focusing on key time nodes such as sudden increases in respiratory rate and sudden decreases in blood oxygen saturation. The temporal attention weights are multiplied by the temporal feature matrix and then subjected to global max pooling to obtain the optimized feature vector of humidified oxygen therapy treatment status.

[0026] Preferably, the procedure also includes a model training step: collecting clinical data on the application of humidified oxygen therapy in vertebroplasty surgery, dividing it into training set, validation set and test set, and performing data augmentation by adding respiratory parameter fluctuation noise and time series offset; constructing a multi-task loss function that comprehensively considers oxygenation effect, respiratory comfort and equipment parameter matching degree, training the attention mechanism neural network using the AdamW optimizer, stopping training when the validation set loss no longer decreases, and deploying the model to the intraoperative optimization AI unit.

[0027] Preferably, the simultaneous calculation of oxygenation efficiency, respiratory load, and airway comfort also includes the following specific steps: calculating the probability distribution of oxygenation adequacy through the fully connected layer, taking the maximum value as the oxygenation efficiency rating result, regressing to calculate the predicted value of breathing work as the respiratory load index, and combining temperature and humidity parameters to calculate the airway mucosal comfort score; updating and optimizing the results in real time, triggering an early warning when blood oxygen saturation or respiratory rate is abnormal, and automatically generating parameter adjustment suggestions.

[0028] refer to Figure 2 This invention also provides an attention-based humidification oxygen therapy parameter optimization assistance system, comprising: The parameter acquisition module includes a medical-grade sensor array and a humidified oxygen therapy device interface, used to simultaneously acquire respiratory physiological parameters and device operating parameters; The preprocessing module, connected to the parameter acquisition module, is used to perform hierarchical preprocessing on the parameters to obtain a standardized feature matrix; The parameter attention processing module is equipped with a parameter attention layer in a preset attention mechanism neural network and is connected to the preprocessing module. It is used to assign dynamic weights to different optimization parameters to obtain a parameter-weighted feature matrix. The temporal feature enhancement module is equipped with a temporal attention layer in a preset attention mechanism neural network and is connected to the parameter attention processing module. It is used to extract key respiratory state features and obtain an optimized feature vector of humidified oxygen therapy treatment state. The optimization module includes a multi-task output unit of the intraoperative optimization host, which is connected to the temporal feature enhancement module and is used to calculate the treatment effect and output parameter optimization suggestions.

[0029] In some embodiments, when this humidified oxygen therapy parameter optimization system is applied, the parameter acquisition module first starts working. Its medical-grade sensor array continuously captures the patient's respiratory physiological parameters during oxygen therapy, and simultaneously acquires various parameters from the humidified oxygen therapy device during operation via the device interface. This parallel acquisition and transmission of both types of parameters provides a complete data foundation for subsequent processing. The acquired parameters are directly sent to the preprocessing module, which performs tiered preprocessing operations based on the characteristics of different parameter types. Through data cleaning, standardization, and other steps, interference caused by differences in units and magnitudes between parameters is eliminated, ultimately forming a standardized feature matrix with a well-structured and consistent data, ensuring that subsequent modules can perform parameter analysis based on a unified standard. After preprocessing, the standardized feature matrix is ​​transmitted to the parameter attention processing module. This module's parameter attention layer, based on a pre-defined attention mechanism neural network algorithm, evaluates the importance of each parameter in the matrix and dynamically assigns corresponding weights to different optimized parameters. The weighting results are directly reflected in the generated parameter-weighted feature matrix, giving key parameters higher priority in subsequent analysis. Subsequently, the parameter-weighted feature matrix enters the temporal feature enhancement module, whose temporal attention layer focuses on the patterns of parameter changes over time, extracting feature information reflecting the patient's key respiratory states from the dynamic data. After feature filtering and integration, an optimized feature vector is generated that accurately characterizes the humidified oxygen therapy treatment status. Finally, the optimized feature vector is sent to the optimization module. The intraoperative optimization host multi-task output unit in this module synchronously performs calculations based on the feature vector, deriving treatment effect-related indicators such as oxygenation efficiency, respiratory load, and airway comfort during oxygen therapy. Based on these indicators, it generates parameter optimization suggestions tailored to the current patient condition and directly outputs them to the clinical operating terminal, providing clear guidance for medical staff to adjust oxygen therapy parameters.

[0030] Thirdly, the present invention also provides an attention-based humidified oxygen therapy parameter optimization auxiliary device, comprising: one or more processors; a memory; one or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, the programs comprising instructions for performing any of the attention-based humidified oxygen therapy parameter optimization auxiliary methods.

[0031] Figure 3 This is a schematic diagram of the structure of the attention-based humidified oxygen therapy parameter optimization auxiliary device provided in the embodiments of this application, with reference to... Figure 3 The attention-based humidified oxygen therapy parameter optimization auxiliary device includes: a processor 31, a memory 32, a communication device 33, an input device 34, and an output device 35. The number of processors 31 and the number of memories 32 in the attention-based humidified oxygen therapy parameter optimization auxiliary device can be one or more. The processor 31, memory 32, communication device 33, input device 34, and output device 35 of the attention-based humidified oxygen therapy parameter optimization auxiliary device can be connected via a bus or other means.

[0032] The memory 32, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the attention mechanism-based humidified oxygen therapy parameter optimization auxiliary method in any embodiment of this application. The memory 32 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function; the data storage area may store data created based on device usage, etc. Furthermore, the memory 32 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0033] The communication device 33 is used for data transmission.

[0034] The processor 31 executes various functional applications and data processing of the device by running software programs, instructions and modules stored in the memory 32, thereby realizing the above-mentioned humidified oxygen therapy parameter optimization auxiliary method based on the attention mechanism.

[0035] Input device 34 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the device. Output device 35 may include display devices such as a display screen.

[0036] The attention-based humidified oxygen therapy parameter optimization auxiliary device provided above can be used to perform the attention-based humidified oxygen therapy parameter optimization auxiliary method provided in the above embodiments, and has corresponding functions and beneficial effects.

[0037] This application also provides a storage medium containing computer-executable instructions. When executed by a computer processor, the computer-executable instructions are used for an attention-based humidified oxygen therapy parameter optimization assistance method, specifically including the following steps: synchronously collecting multiple respiratory physiological parameters and equipment operating parameters during humidified oxygen therapy treatment through a medical sensor array; performing hierarchical preprocessing on the parameters to obtain a standardized feature matrix; inputting the standardized feature matrix into a preset attention mechanism neural network, assigning dynamic weights to different optimization parameters through a parameter attention layer to obtain a parameter-weighted feature matrix; performing temporal feature enhancement processing on the parameter-weighted feature matrix through a temporal attention layer to extract key respiratory state features and generate a humidified oxygen therapy treatment state optimization feature vector; and based on the optimization feature vector, synchronously calculating oxygenation efficiency, respiratory load, and airway comfort, and outputting a humidified oxygen therapy suggestion report result.

[0038] Storage medium—any type of memory device or storage apparatus. The term "storage medium" is intended to include: mounting media, such as CD-ROMs, floppy disks, or magnetic tape devices; computer system memory or random access memory, such as DRAM, DDRRAM, SRAM, EDORAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (e.g., hard disks or optical storage); registers or other similar types of memory elements, etc. Storage media may also include other types of memory or combinations thereof. Furthermore, storage media may reside in a first computer system in which a program is executed, or may reside in a different second computer system connected to the first computer system via a network (such as the Internet). The second computer system can provide program instructions to the first computer for execution. The term "storage medium" may include two or more storage media residing in different locations (e.g., in different computer systems connected via a network). Storage media may store program instructions (e.g., specifically implemented as a computer program) executable by one or more processors. The attention-based humidified oxygen therapy parameter optimization method and system provided in this invention can effectively solve various problems in the prior art and bring significant benefits in many aspects. First, by using a medical sensor array to simultaneously collect respiratory physiological parameters and equipment operating parameters, it overcomes the limitations of independent optimization of a single parameter in traditional optimization, achieving comprehensive acquisition of multi-dimensional data. This lays the foundation for subsequent accurate analysis, avoids optimization blind spots due to data gaps, and ensures real-time monitoring of the patient's overall oxygen therapy status. Second, by introducing an attention mechanism neural network, dynamic weights are assigned to different parameters through a parameter attention layer. This allows for accurate identification of parameters that play a key role in the oxygen therapy effect. Compared to traditional fixed-weight optimization methods, it can more effectively focus on key data, improve the accuracy of parameter analysis, and avoid judgment biases caused by invalid data interference. Furthermore, the temporal attention layer can enhance the extraction of temporal features of parameters, capturing the dynamic changes in the patient's physiological state and equipment operating parameters in real time. It can quickly respond and generate corresponding optimized feature vectors, achieving real-time parameter control and solving the problem of lag in traditional optimization systems, effectively reducing the risk of complications caused by parameter mismatch. Finally, based on the optimized feature vectors, oxygenation efficiency, respiratory load, and airway comfort are calculated simultaneously, enabling the evaluation of oxygen therapy effects from multiple dimensions such as efficacy and patient tolerance. It also outputs precise parameter optimization suggestions, meeting the needs of personalized and precise clinical treatment. Especially in scenarios requiring high oxygen therapy stability, it can significantly improve treatment safety and patient comfort, providing more reliable technical support for clinical respiratory support therapy.

[0039] Of course, the computer-executable instructions provided in the embodiments of this application are not limited to the above-mentioned attention-based humidified oxygen therapy parameter optimization assistance method, but can also execute related operations in the attention-based humidified oxygen therapy parameter optimization assistance method provided in any embodiment of this application. Obviously, those skilled in the art can make various modifications and variations to this invention without departing from the spirit and scope of the invention. Therefore, if these modifications and variations of the invention fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for optimizing humidified oxygen therapy parameters based on attention mechanisms, characterized in that, Includes the following steps: Multiple respiratory physiological parameters and equipment operating parameters during humidified oxygen therapy are synchronously collected using a medical sensor array; the parameters are then preprocessed in a hierarchical manner to obtain a standardized feature matrix. The standardized feature matrix is ​​input into a preset attention mechanism neural network, and dynamic weights are assigned to different optimization parameters through a parameter attention layer to obtain a parameter-weighted feature matrix; The parameter-weighted feature matrix is ​​enhanced by a temporal attention layer to extract key respiratory state features and generate an optimized feature vector. Based on the optimized feature vector, oxygenation efficiency, respiratory load, and airway comfort are calculated simultaneously, and a humidified oxygen therapy recommendation report is output.

2. The auxiliary method for optimizing humidified oxygen therapy parameters based on attention mechanisms according to claim 1, characterized in that, The synchronous acquisition of multiple parameters via a medical sensor array also includes the following specific steps: The multi-channel optimization system is activated, and the temperature of the inhaled gas is collected in real time through the nasopharyngeal temperature sensor. After being processed by the signal calibration module, the gas is collected at a preset frequency. The system acquires inhaled oxygen concentration data through an oxygen concentration optimization module, collects actual airflow velocity through a flow sensor, collects blood oxygen saturation data through a pulse oximeter, records respiratory rate through a respiratory rate optimizer, and acquires end-tidal carbon dioxide partial pressure data through a carbon dioxide optimization module.

3. The auxiliary method for optimizing humidified oxygen therapy parameters based on attention mechanisms according to claim 1, characterized in that, The method for obtaining the parameter-weighted feature matrix also includes the following steps: By optimizing the embedded processing unit of the host computer, an adaptive filtering algorithm is used to denoise the respiratory frequency signal, extract respiratory cycle feature parameters, and use wavelet transform to analyze the irregularity of the respiratory waveform to obtain preprocessed respiratory feature data. Temperature, oxygen concentration, and flow rate data are processed using a multi-parameter preprocessing unit and the Z-score normalization algorithm. The dynamic trend of blood oxygen saturation is extracted by the time series analysis module, and the minimum value and fluctuation range of blood oxygen saturation within a set time period are calculated by the sliding window algorithm to obtain oxygenation dynamic characteristic data. The above data is uniformly converted into a standardized feature matrix through the matrix integration unit.

4. The auxiliary method for optimizing humidified oxygen therapy parameters based on attention mechanisms according to claim 1, characterized in that, The process of obtaining the parameter-weighted feature matrix also includes the following specific steps: The standardized feature matrix is ​​input into the first fully connected layer to obtain the initial weight vector of parameters. The current oxygenation state is determined based on the blood oxygen saturation level and respiratory rate, and the corresponding basic weight matrix is ​​called. The initial parameter weight vector is multiplied by the basic weight matrix and then normalized by Softmax to obtain the parameter attention weights. The weights of oxygen concentration and flow rate under hypoxia and respiratory rate and airway pressure under high respiratory load are emphasized. The parameter attention weights are then multiplied element-wise by the original feature matrix to generate the parameter weighted feature matrix.

5. The auxiliary method for optimizing humidified oxygen therapy parameters based on attention mechanisms according to claim 1, characterized in that, The extraction of key respiratory state features through the temporal attention layer also includes: The parameter-weighted feature matrix is ​​input into a bidirectional LSTM layer to extract the temporal feature matrix. The temporal feature matrix is ​​then subjected to global average pooling and input into the second fully connected layer to generate temporal attention weights, focusing on key time nodes such as a sudden increase in respiratory rate and a sudden decrease in blood oxygen saturation. Multiply the temporal attention weights by the temporal feature matrix, and then perform global max pooling to obtain the optimized feature vector of humidified oxygen therapy treatment status.

6. The auxiliary method for optimizing humidified oxygen therapy parameters based on attention mechanisms according to claim 1, characterized in that, It also includes the model training step: Clinical data on the application of humidified oxygen therapy in vertebroplasty were collected and divided into training, validation and test sets. Data augmentation was performed by adding respiratory parameter fluctuation noise and time offset. A multi-task loss function was constructed, taking into account oxygenation effect, breathing comfort and equipment parameter matching. The AdamW optimizer was used to train the attention mechanism neural network. Training was stopped when the validation set loss no longer decreased, and the model was deployed to the intraoperative optimization AI unit.

7. The auxiliary method for optimizing humidified oxygen therapy parameters based on attention mechanisms according to claim 1, characterized in that, The simultaneous calculation of oxygenation efficiency, respiratory load and airway comfort also includes the following specific steps: calculating the probability distribution of oxygenation adequacy through the fully connected layer, taking the maximum value as the oxygenation efficiency rating result, regressing to calculate the predicted value of breathing work as the respiratory load index, and combining temperature and humidity parameters to calculate the airway mucosal comfort score. The system updates optimization results in real time, triggers warnings when blood oxygen saturation or respiratory rate is abnormal, and automatically generates parameter adjustment suggestions.

8. A humidified oxygen therapy parameter optimization auxiliary system based on attention mechanisms, characterized in that, include: The parameter acquisition module includes a medical-grade sensor array and a humidified oxygen therapy device interface, used to simultaneously acquire respiratory physiological parameters and device operating parameters; The preprocessing module, connected to the parameter acquisition module, is used to perform hierarchical preprocessing on the parameters to obtain a standardized feature matrix; The parameter attention processing module is equipped with a parameter attention layer in a preset attention mechanism neural network and is connected to the preprocessing module. It is used to assign dynamic weights to different optimization parameters to obtain a parameter-weighted feature matrix. The temporal feature enhancement module is equipped with a temporal attention layer in a preset attention mechanism neural network and is connected to the parameter attention processing module. It is used to extract key respiratory state features and obtain an optimized feature vector of humidified oxygen therapy treatment state. The optimization module includes a multi-task output unit of the intraoperative optimization host, which is connected to the temporal feature enhancement module and is used to calculate the treatment effect and output parameter optimization suggestions.

9. An auxiliary device for optimizing humidified oxygen therapy parameters based on an attention mechanism, characterized in that, include: One or more processors; Memory; One or more computer programs, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors, the programs including instructions for performing the attention mechanism-based humidified oxygen therapy parameter optimization auxiliary method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the attention mechanism-based humidified oxygen therapy parameter optimization auxiliary method as described in any one of claims 1-7.

Images