A method and system for pressure regulation and monitoring of intubation of a nasogastric tube

Abstract

The application relates to the technical field of nasogastric tube intubation, and discloses a pressure regulation and monitoring method and system for nasogastric tube intubation, wherein the pressure regulation and monitoring method for nasogastric tube intubation comprises the following steps: acquiring real-time pressure monitoring data of a nasogastric tube intubation process, and generating a pressure mode feature vector; mapping the pressure mode feature vector to a low-dimensional hash space by using a region-sensitive hash algorithm, and generating a hash code; carrying out similar case retrieval in a multi-level hash index structure based on the hash code, and outputting a candidate similar pressure mode set; analyzing the similar pressure modes and corresponding historical control strategies retrieved, and generating a control strategy recommendation for a current intubation scene; the application solves the technical problem of real-time retrieval of similar pressure modes in a large-scale historical case library, realizes millisecond-level pressure mode identification and control strategy recommendation, and improves the decision-making efficiency and safety of the nasogastric tube intubation process.

Description

Technical Field

[0001] This invention relates to the field of nasogastric tube insertion technology, and more specifically, to a method and system for pressure regulation and monitoring during nasogastric tube insertion. Background Technology

[0002] In the clinical procedure of nasogastric tube intubation, real-time monitoring and control of pressure changes are necessary to ensure safety and success. A large amount of pressure monitoring data is generated during intubation, containing a rich variety of pressure change patterns, each often corresponding to a specific physiological state and corresponding control strategy. In practical applications, when encountering a certain pressure pattern, the ability to quickly retrieve similar pressure patterns and their corresponding successful control strategies from a historical case database will greatly improve the safety and efficiency of the intubation procedure.

[0003] However, existing stress pattern recognition and retrieval systems suffer from serious efficiency problems. Traditional sequential comparison methods require calculating the similarity between the current stress pattern and each case in the historical case database one by one. When the historical case database reaches tens or even hundreds of thousands of records, the time complexity of this linear search method is O(n log n). ,in This represents the number of historical cases. In clinical scenarios requiring real-time responses, such retrieval delays are unacceptable.

[0004] Furthermore, stress pattern data typically possesses high-dimensional characteristics, and traditional similarity calculation methods suffer from high computational complexity in high-dimensional spaces, further exacerbating the retrieval efficiency problem. This results in the system's inability to find similar historical stress patterns within milliseconds, thus hindering the timely provision of effective control strategy suggestions to healthcare professionals and impacting the safety and success rate of nasogastric tube intubation. Summary of the Invention

[0005] This invention provides a method and system for pressure regulation and monitoring during nasogastric tube insertion, solving the technical problem of low retrieval efficiency in traditional sequential alignment methods in related technologies.

[0006] This invention provides a method for pressure regulation and monitoring during nasogastric tube insertion, comprising:

[0007] Real-time pressure monitoring data during nasogastric tube intubation is acquired, and pressure pattern feature vectors are generated.

[0008] The stress pattern feature vector is mapped to a low-dimensional hash space using the region-sensitive hashing algorithm to generate hash codes.

[0009] Based on hash encoding, similar case retrieval is performed in a pre-established multi-level hash index structure, and a candidate set of similar pressure patterns is output.

[0010] Analyze the retrieved candidate similar pressure patterns and their corresponding historical control strategies to generate a control strategy recommendation for the current intubation scenario, and adjust the pressure based on the control strategy recommendation;

[0011] The region-sensitive hashing algorithm includes a hash function based on random projection and a hash function based on a random hyperplane. The hash function is selected according to the similarity measurement method of the stress pattern feature vector.

[0012] Furthermore, the pressure mode feature vector includes a pressure mode time-domain feature vector, a pressure mode frequency-domain feature vector, and a correlation feature vector, and is generated by the following steps:

[0013] The sliding window algorithm is used to calculate the time-domain feature vector of the pressure pattern from the real-time pressure monitoring data, including the pressure mean, variance, peak value and valley value within each window;

[0014] The Fast Fourier Transform algorithm is used to extract features from real-time pressure monitoring data to obtain the frequency domain feature vector of the pressure mode, including the main frequency component and the spectral energy distribution.

[0015] The corresponding autocorrelation function and cross-correlation function are calculated based on real-time pressure monitoring data, and a correlation feature vector is generated based on the autocorrelation function and cross-correlation function.

[0016] Furthermore, the hash function based on random projection Defined as:

[0017] ;

[0018] in, For feature vectors, It is a random vector sampled from a standard normal distribution. From Random numbers sampled from a uniform distribution It is the width parameter of the hash bucket. This indicates that a rounding down operation is being performed.

[0019] Furthermore, the hash function based on the random hyperplane Defined as:

[0020] ;

[0021] in, For feature vectors, It is a random vector sampled from a standard normal distribution. Represents a sign function, when Returns 1 when Returns -1 when Returns 0.

[0022] Furthermore, the pre-established multi-level hash index structure includes Each hash table uses a different combination of hash functions. The value ranges from 2 to 10 and is dynamically adjusted according to the size of the historical case library and the requirements for retrieval efficiency. The multi-level hash index structure is constructed based on the historical pressure pattern feature vector and its corresponding hash code in the historical case library.

[0023] Furthermore, the generation of the candidate similar pressure pattern set also includes the following steps:

[0024] For each similar case in the candidate similar stress pattern set, calculate its similarity score with the stress pattern corresponding to the retrieved hash code;

[0025] Sort the similarity scores and output the top-ranked similarity scores. The stress patterns of similar cases were used as the final search results, among which The value range is from 5 to 20.

[0026] Furthermore, the similarity score is calculated using one of the following similarity functions:

[0027] Euclidean distance function :

[0028] ;

[0029] Cosine similarity function :

[0030] ;

[0031] Mahalanobis distance function :

[0032] ;

[0033] in, The retrieved hash code corresponds to the pressure pattern feature vector. For similar cases, the stress pattern feature vectors are... The covariance matrix of the eigenvectors. express The i-th component, express The i-th component, The dimension of the feature vector. This indicates the transpose operation.

[0034] Furthermore, the generation of the recommended control strategy includes the following steps:

[0035] Extract the control strategy parameters used in each similar case, including intubation speed, angle adjustment, and pressure threshold;

[0036] The parameters of each control strategy are normalized, and the normalized similarity scores of similar cases are used as weights to perform weighted fusion of the normalized control strategy parameters to generate recommended control strategy values ​​suitable for the current intubation scenario.

[0037] The recommended values ​​of the control strategy are restored to their original dimensions through inverse transformation, and then truncated and adjusted in combination with the effective range of parameters in clinical practice.

[0038] Furthermore, the generation of the recommended control strategy also includes the following steps:

[0039] By using collaborative filtering algorithms to analyze the correlation between similar cases, we can identify control strategy combinations that have a success rate exceeding a preset success rate threshold under similar stress patterns and generate control strategy recommendations based on case correlation.

[0040] Rare stress pattern cases are input into a generative adversarial network (GAN), and virtual case data is synthesized by the generator in the GAN. The virtual case data is then integrated into the historical case library.

[0041] Obtain real-time feedback data after pressure adjustment based on control strategy recommendations, and update the multi-level hash index structure based on the real-time feedback data, including adding the new pressure mode corresponding cases and their control effects to the historical case library, and dynamically adjusting the hash function parameters used by the hash tables contained in the multi-level hash index structure.

[0042] This invention provides a pressure regulation and monitoring system for nasogastric tube insertion, used to perform the aforementioned pressure regulation and monitoring method for nasogastric tube insertion, comprising:

[0043] The data acquisition module is used to acquire real-time pressure monitoring data during the nasogastric tube insertion process;

[0044] The feature generation module is used to generate pressure pattern feature vectors based on real-time pressure monitoring data.

[0045] The hash encoding generation module uses a region-sensitive hashing algorithm to map the stress pattern feature vector to a low-dimensional hash space to generate a hash code; wherein, the region-sensitive hashing algorithm includes a hash function based on random projection and a hash function based on a random hyperplane, and the hash function is selected according to the similarity measurement method of the stress pattern feature vector;

[0046] The similar case retrieval module performs similar case retrieval based on hash encoding in a pre-established multi-level hash index structure and outputs a set of candidate similar pressure patterns.

[0047] The control strategy generation and adjustment module is used to analyze the retrieved candidate similar pressure patterns and their corresponding historical control strategies, generate control strategy recommendations for the current intubation scenario, and allow medical staff to adjust the pressure based on the control strategy recommendations.

[0048] The human-computer interaction interface is used to display search results and control strategy recommendations to medical staff, and to receive operational feedback.

[0049] The beneficial effects of this invention are as follows:

[0050] 1) This invention provides a method and system for pressure regulation and monitoring during nasogastric tube insertion, which extracts features from real-time pressure monitoring data from multiple dimensions including time domain, frequency domain, and correlation, thereby improving the comprehensiveness and accuracy of the acquired data;

[0051] 2) This invention provides a method and system for pressure regulation and monitoring during nasogastric tube insertion. It utilizes a region-sensitive hashing algorithm to map pressure pattern feature vectors to a low-dimensional hash space, reducing data storage space and computational complexity. The region-sensitive hashing algorithm includes a hash function based on random projection and a hash function based on a random hyperplane. The hash function is selected according to the similarity measurement method of the pressure pattern feature vectors. This step ensures that similar pressure pattern feature vectors also have high similarity in the hash space, facilitating subsequent similar case retrieval.

[0052] 3) This invention provides a method and system for regulating and monitoring pressure during nasogastric tube insertion. Based on hash encoding, it performs similar case retrieval in a pre-constructed multi-level hash index structure and outputs a set of candidate similar pressure patterns. The multi-level hash index structure contains multiple independent hash tables, each using a different combination of hash functions, which improves the accuracy and efficiency of similar case retrieval. Furthermore, the number of independent hash tables can be dynamically adjusted according to actual conditions, enabling the system to flexibly adapt to different sizes of historical case databases and retrieval efficiency requirements, thereby enhancing the system's adaptability and practicality.

[0053] 4) This invention provides a method and system for pressure regulation and monitoring during nasogastric tube insertion. It comprehensively considers parameters of similar cases, correlations between cases, and rare cases, and generates control strategy recommendations through multiple methods to make the recommended control strategies more suitable for the current insertion scenario. Furthermore, this invention updates the multi-level hash index structure and historical case library based on real-time feedback data, so that the system can continuously learn and adapt to new scenarios, and realize continuous optimization and dynamic evolution of the system. Attached Figure Description

[0054] Figure 1 This is a flowchart of the pressure regulation and monitoring method for nasogastric tube insertion according to the present invention. Detailed Implementation

[0055] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0056] Example 1:

[0057] This embodiment provides a method for pressure regulation and monitoring during nasogastric tube insertion, such as... Figure 1 As shown, the method includes the following steps:

[0058] Step 100: Obtain real-time pressure monitoring data during the nasogastric tube insertion process and generate a pressure pattern feature vector.

[0059] This step involves acquiring real-time pressure monitoring data during cannulation using pressure sensors, including parameters such as pressure value sequences, pressure change rates, pressure peak values, and pressure waveform characteristics. Before feature extraction, the raw pressure data (i.e., real-time pressure monitoring data) is preprocessed, including noise reduction, outlier detection and handling, data smoothing, and standardization to eliminate data bias caused by differences in measurement environments and equipment. Standardization uses the Z-score standardization method to convert each pressure measurement value into a standard normal distribution with a mean of 0 and a standard deviation of 1, ensuring that pressure data of different dimensions and ranges can be effectively compared and calculated. After preprocessing, features are extracted from these data to generate feature vectors describing the current pressure pattern. The dimensions of the feature vectors are determined according to the actual application requirements and typically include information from multiple dimensions such as time-domain features, frequency-domain features, and statistical features.

[0060] It should be noted that various feature extraction methods can be used to generate stress pattern feature vectors.

[0061] For example, step 101: Analyze the pressure data sequence of real-time pressure monitoring data using the sliding window method, calculate the pressure mean, variance, peak value and valley value within each window, and output the pressure mode time domain feature vector;

[0062] Step 102: Input the pressure data sequence into the Fast Fourier Transform algorithm to extract the main frequency components and spectral energy distribution, and output the pressure mode frequency domain feature vector;

[0063] Step 103: Calculate the autocorrelation function and cross-correlation function based on the pressure data sequence, identify the periodic pattern of pressure change, and output the correlation feature vector.

[0064] Furthermore, the sliding window method in step 101 receives a time series as input. :

[0065] ;

[0066] in , , They are the 1st, 2nd, and 3rd respectively. Each time series unit outputs a feature vector. :

[0067] ;

[0068] in , , Corresponding to The first, second, and m features each correspond to a statistic, such as mean, variance, peak value, and trough value.

[0069] Furthermore, the input received by the Fast Fourier Transform algorithm in step 102 is a time series. The output is a frequency domain feature vector. :

[0070] ;

[0071] in , , These represent the amplitudes of the 1st, 2nd, and kth corresponding frequency components, respectively.

[0072] Furthermore, the autocorrelation and cross-correlation function analyses in step 103 receive a time series input. ;

[0073] The output is a relevance feature vector:

[0074] ;

[0075] in , , They represent the 1st, 2nd, and 3rd respectively. The correlation coefficient under a time delay.

[0076] Step 200: Use the region-sensitive hashing algorithm to map the stress pattern feature vector to a low-dimensional hash space to generate a hash code.

[0077] This step employs the Region Sensitive Hash (LSH) algorithm to map high-dimensional stress pattern feature vectors to a low-dimensional hash space. To ensure the effectiveness of the hash mapping, the feature vectors generated in step 100 are first normalized, scaling the feature values ​​of different dimensions to a uniform numerical range to eliminate the impact of dimensional differences on hash calculation. For time-domain, frequency-domain, and correlation features, a min-max normalization method is used to map them to the [0,1] interval, ensuring that all types of features have the same weight in the hash calculation. The LSH algorithm designs a special family of hash functions so that similar feature vectors are mapped to the same hash bucket with a higher probability, while dissimilar feature vectors are mapped to the same hash bucket with a lower probability. For the input... 3D feature vector ,use An independent hash function Perform mapping and generate Bit hash code :

[0078] ;

[0079] , , These represent the hash codes for the 1st, 2nd, and kh bits, respectively.

[0080] It should be noted that the choice of region-sensitive hash function depends on the similarity measurement method of feature vectors.

[0081] Step 201: When using Euclidean distance to measure similarity, use a hash function based on random projection. :

[0082] ;

[0083] in It is a random vector sampled from a standard normal distribution. From Random numbers sampled from a uniform distribution It is the width parameter of the hash bucket. Indicates to The underlying function for rounding down;

[0084] Step 202: When using cosine similarity as a metric, use a hash function based on a random hyperplane. :

[0085] ;

[0086] in It is a random vector sampled from a standard normal distribution. Represents a sign function, when Returns 1 when Returns -1 when Returns 0.

[0087] Furthermore, in this embodiment, the input received by the region-sensitive hashing algorithm is a stress pattern feature vector. :

[0088] ;

[0089] This is a high-dimensional vector that contains time-domain features, frequency-domain features, and correlation features;

[0090] The algorithm outputs a low-dimensional hash code. :

[0091] ;

[0092] in , , Represents low-dimensional hash encoding The 1st, 2nd, and zkth bits are binary values ​​of 0 or 1. It is much smaller than the original dimension of the feature vector.

[0093] Furthermore, the random projection hash function in step 201 receives a feature vector as input. and its corresponding random vector and random offset The output is an integer hash value; the random projection hash function discretizes the continuous feature space into a finite number of hash buckets by calculating the projection of the feature vector onto a random vector and then quantizing it. To ensure the consistency of the computational dimensions, the random vector... Using normalized feature vectors Constructed with the same dimensions, and with each component sampled from a standard normal distribution, the dot product is... Become a dimensionless quantity; random offset Bucket width parameter It is also designed as a dimensionless parameter to ensure the consistency of dimensions throughout the hash function. In the specific implementation, a random vector... The dimension is the same as the feature vector, and the elements follow a standard normal distribution; random offset. from Uniform sampling within a given interval; bucket width parameter Based on the data distribution characteristics, a value of 4.0 is used in this implementation. This value was determined through experimental testing and can be adjusted between 1.0 and 10.0. A smaller value is acceptable. Larger values ​​make hashes more sensitive (leading to more hash buckets), while larger values... The value makes the hash more robust (resulting in fewer hash buckets); when the feature space dimension is high (e.g., more than 100 dimensions), a projection dimension reduction strategy is adopted, where each hash function only uses a random subset of the feature vector to calculate the projection, in order to reduce computational complexity. The size of the subset is usually set to 10%-30% of the total dimension.

[0094] Furthermore, the input received by the random hyperplane hash function in step 202 is a feature vector. and its corresponding random vector The output is a binary hash value (1 or -1). The random hyperplane hash function achieves a binary partition of the feature space by determining which side of the random hyperplane the feature vector lies on, thus preserving the cosine similarity relationship between vectors. In practical applications, random vectors... The generation uses a pseudo-random number generator combined with a fixed seed value (e.g., a seed value of 42) to ensure the reproducibility of the hash function, and its dimension is similar to that of the feature vector. Similarly, each component follows a standard normal distribution. Independent sampling; to improve hash stability, the input feature vector is... First, perform L2 normalization, that is... Obtain the normalized feature vectors To eliminate interference caused by differences in vector lengths; and to handle the boundary case where the dot product result is exactly 0, rules are set when... Returning 1 at the time ensures the complete definition of the hash function.

[0095] Step 300: Based on hash encoding, perform similar case retrieval in a multi-level hash index structure and output a set of candidate similar pressure patterns.

[0096] This step utilizes the generated hash code to perform fast retrieval within a pre-established multi-level hash index structure. The multi-level hash index structure contains... Each hash table uses a separate hash table, with a different combination of hash functions. For the pressure pattern feature vector of the retrieval, the corresponding hash bucket is located in each hash table, and historical cases from all hash buckets are merged into a candidate set. This multi-level index structure reduces the time complexity of retrieval from... Reduce to ,in It is the average number of cases in each hash bucket. It represents the total number of historical cases, usually .

[0097] In this embodiment of the application, in order to further improve the retrieval accuracy, step 301 is also included in addition to step 300:

[0098] For each historical case in the candidate similar stress pattern set, calculate its precise similarity score with the query stress pattern, sort them according to the similarity score, and output the top cases with the highest similarity. Each pressure pattern is used as the final search result. In this embodiment, The value ranges from 5 to 20, and is set according to the complexity of the specific clinical scenario and the time required for decision-making. In emergency situations, a smaller value (e.g., 5-10) is used to quickly provide recommendations, while in routine operations, a larger value (e.g., 15-20) is used to improve the diversity and comprehensiveness of the recommendations. The similarity score can be obtained through a function... The function calculates and returns a numerical value indicating the degree of similarity between two feature vectors, depending on the selected metric (such as Euclidean distance, cosine similarity, etc.).

[0099] Furthermore, a multi-level hash index structure is a type of structure composed of... A data structure consisting of several independent hash tables, each using a different combination of hash functions, improves the recall rate of retrieval. In this embodiment, The value ranges from 2 to 10, and is dynamically adjusted according to the size of the historical case database and retrieval efficiency requirements. When the case database is large (e.g., exceeding 100,000 records), the value approaches the upper limit to improve retrieval recall. The multi-level hash index structure receives a stress-mode hash code as input;

[0100] The output is a set of candidate similar cases. :

[0101] ;

[0102] in , , These are the 1st, 2nd, and mth historical stress pattern cases, respectively.

[0103] Furthermore, the precise similarity calculation in step 301 receives the retrieved stress pattern feature vector as input. and the feature vector of each case in the candidate case set The output is a similarity score. ;

[0104] To ensure dimensional consistency across different feature dimensions, the feature vectors are standardized twice before similarity calculation to eliminate any possible residual dimensional differences.

[0105] Depending on the application scenario, similarity can be calculated using the Euclidean distance function. :

[0106] ;

[0107] Stress pattern feature vector and the feature vector of each case in the candidate case set , express The i-th component, express The i-th component.

[0108] Cosine similarity function :

[0109] ;

[0110] Mahalanobis distance function :

[0111] ;

[0112] These metrics can accurately reflect the degree of similarity between stress patterns, among which The covariance matrix represents the features. The cosine similarity function, by dividing the dot product of vectors by the product of their magnitudes, naturally eliminates the influence of dimensions; the Mahalanobis distance function, by introducing the inverse of the covariance matrix as weights, appropriately compensates for the dimensional differences and correlations among the feature dimensions. This indicates the transpose operation.

[0113] Step 400: Analyze the retrieved similar pressure patterns and their corresponding historical control strategies to generate a recommended control strategy for the current intubation scenario.

[0114] This step analyzes similar historical cases and extracts the control strategy parameters used in each case, including intubation speed, angle adjustment, and pressure threshold. Since these control strategy parameters have different units and dimensions (e.g., intubation speed in mm / s, angle in degrees, and pressure threshold in kPa), each parameter needs to be normalized before fusion to eliminate the impact of dimensional differences on the weighted average. These normalized control strategy parameters are then weighted and fused according to similarity weights to generate recommended control strategy values ​​suitable for the current intubation scenario. Finally, the fused normalized parameter values ​​are restored to their original dimensions through inverse transformation and truncated based on the effective range of parameters in clinical practice to ensure that the recommended control parameters are within a safe and effective range. The output control strategy includes specific operational parameters and implementation suggestions, presented in an intuitive and easy-to-understand manner for medical staff to reference.

[0115] In this embodiment of the application, in order to improve the accuracy of the control strategy recommendation, step 401 is further included in addition to step 400:

[0116] By using collaborative filtering algorithms to analyze the correlation between similar cases, we can identify control strategy combinations that have a success rate exceeding a preset success rate threshold (e.g., 80%) under similar stress patterns and generate strategy recommendations based on case correlation.

[0117] Step 402: Input rare stress pattern cases into the generative adversarial network, and synthesize similar but varied virtual case data through the generator network to expand the case library of rare scenarios and improve the recommendation ability in scarce scenarios.

[0118] Step 403: Update the hash index structure based on real-time feedback data, add new stress mode cases and their control effects to the historical case library, dynamically adjust the hash function parameters, and continuously optimize retrieval accuracy and recommendation quality.

[0119] Step 403 specifically employs a time decay weighting model, assigning different weights to historical cases at different time points. The weight calculation formula is as follows:

[0120]

[0121] in Indicates the current time. Timestamps representing historical cases The time decay coefficient ranges from [0.001, 0.1], and is set to 0.01 in this embodiment. The hash function parameters aim to maximize retrieval recall, minimize normalized retrieval time, and maximize retrieval precision. The dynamic adjustment of the hash function parameters employs a time-based multi-objective optimization method, with the objective function being... :

[0122] ;

[0123] in It is the number of hash tables. It is the hash code length. It is the hash bucket width parameter. Indicates the retrieval recall rate. This represents the normalized retrieval time. This represents the time relevance score, used to measure the accuracy of retrieving recent cases. , and To balance the coefficients of each objective and satisfy In this embodiment, the values ​​are set to 0.4, 0.3, and 0.3, respectively.

[0124] Furthermore, the weighted fusion strategy in step 400 receives the retrieved set of similar cases as input. :

[0125] ;

[0126] and their corresponding similarity scores :

[0127] ;

[0128] and control strategy parameter P:

[0129] ;

[0130] The output is a recommended control strategy after fusion. ;

[0131] , , They represent the 1st, 2nd, and 3rd respectively. A number of similar cases; , , They represent the 1st, 2nd, and 3rd respectively. Similarity scores of similar cases , , They represent the 1st, 2nd, and 3rd respectively. Each control strategy parameter;

[0132] The weighted fusion strategy uses a similarity-weighted average to ensure that control strategies for more similar cases receive a higher impact weight.

[0133] Specifically, for each control strategy parameter First, a minimum-maximum normalization process is performed to obtain the normalized control strategy parameters. :

[0134]

[0135] in and These are the minimum and maximum values ​​of the parameter across all candidate cases, respectively.

[0136] Then, the normalized similarity score is used. :

[0137] ;

[0138] As weights, they are used in the weighted average formula. :

[0139] ;

[0140] Calculate the recommended value for this parameter;

[0141] Finally, the recommended values ​​in the normalized space are transformed back into the original parameter space to obtain the recommended values ​​for the normalized parameters. :

[0142] ;

[0143] Threshold limits are applied based on the effective range of the parameters to ensure that the final recommended value is within a safe and effective range.

[0144] Furthermore, the collaborative filtering algorithm in step 401 is a recommendation algorithm based on historical data analysis of user preferences or item similarity. In this embodiment, the collaborative filtering algorithm receives a stress pattern-control strategy matrix as input. Each row represents a stress mode, each column represents a control strategy parameter, and the matrix elements indicate the success rate of using the control strategy under that stress mode. Because the matrix... Different control strategy parameters (such as speed, angle, and pressure threshold) have different dimensions and scoring ranges. Before applying the collaborative filtering algorithm, the matrix is ​​first normalized to convert the success scores of each control strategy parameter into a uniform 0-1 range, eliminating the influence of dimensional differences on similarity calculation. Column normalization specifically employs the Z-score standardization method, i.e., normalizing the matrix... Each column Calculate its mean and standard deviation Then for each element in that column Perform the transformation:

[0145] ;

[0146] The standardized values ​​are then mapped to the Sigmoid function. The interval, since the Sigmoid function is a commonly used technique in this field, will not be elaborated upon here. The output of the collaborative filtering algorithm is the optimal combination of control strategies for the current stress mode. The output is a recommended value in the normalized space, which needs to be mapped back to the original parameter space through an inverse transformation and converted into specific clinical operation guidelines, such as insertion at a speed of 2.5 mm / s, maintaining a 15-degree elevation angle, pausing and waiting for 2 seconds when the pressure exceeds 1.8 kPa, and other instructions that medical staff can directly execute. The collaborative filtering algorithm uses a similarity function. Calculate stress mode and The similarity between them, in this embodiment, is actually achieved using cosine similarity, that is:

[0147] ;

[0148] in To control the total number of strategy parameters;

[0149] Predicting the current pressure pattern using a weighted average function. Control strategy The score is expressed by the formula:

[0150] ;

[0151] in It is the set of neighbors most similar to the current stress pattern. In this implementation, the neighbor set... The value of is between 10 and 50, and is dynamically adjusted based on data distribution characteristics and algorithm performance. and These represent the stress mode-control strategy matrix respectively. The element in the i-th row and s-th column and the element in the j-th row and s-th column;

[0152] When the data is dense and the similarity distribution is uniform Use smaller values ​​(e.g., 10-20) to improve recommendation accuracy;

[0153] When the data is sparse or the similarity distribution is uneven Take a larger value (e.g., 30-50) to increase the robustness of the recommendation.

[0154] Similarity function The value range is [-1, 1], where 1 represents complete similarity, 0 represents no correlation, and -1 represents complete oppositeness.

[0155] Furthermore, the generative adversarial network in step 402 consists of two neural networks: a generator and a discriminator. The generator network receives a random noise vector as input. Rare stress pattern characteristics The output is the synthesized pressure mode features. The discriminator network receives stress pattern features (real or synthetic) as input and outputs the probability that the feature is real data. The training objective of the generative adversarial network is to enable the generator network to generate synthetic data that the discriminator network cannot distinguish between real and synthetic data. In practical implementations, random noise vectors... We use a standard normal distribution with dimension 100 for sampling, that is:

[0156] ;

[0157] Rare stress pattern characteristics Principal component analysis was used to reduce the dimensionality to 50 dimensions, retaining more than 90% of the variance information, ensuring that the main feature structure was captured while reducing computational complexity;

[0158] The generative adversarial network is trained using the Wasserstein distance-based loss function WGAN-GP, where the gradient penalty coefficient is... Discriminator training iterations The learning rates for both the generator and the discriminator are set to 0.0002, and training stability is guaranteed by batch normalization and gradient clipping (with a threshold of 0.01).

[0159] Furthermore, the generator network is a type of feedforward neural network, and its layer structure includes:

[0160] The input layer receives conditional information from a random noise vector and rare stress pattern features, where the rare stress pattern features are normalized before input.

[0161] Multiple fully connected hidden layers, each followed by batch normalization and ReLU activation functions, are used for feature transformation and non-linear mapping;

[0162] The output layer generates a synthetic feature vector with the same dimensions as the features of the real stress pattern, and uses the Tanh activation function to scale the output values ​​to the range of [-1, 1].

[0163] The generated synthetic feature vectors need to be transformed back to the original feature space through inverse normalization, and then converted into a clinically interpretable stress pattern representation by the feature decoder. This representation includes typical stress waveform parameters, key point time series, and descriptions of stress change patterns, making it easier for healthcare professionals to understand and use. The generator network is trained using an adversarial loss function. :

[0164] ;

[0165] in Represents the mathematical expectation. Represents a generator function. Let represent the discriminator function, whose goal is to maximize the probability that the discriminator classifies a generated sample as a real sample. In actual training, to ensure the continuity and physical plausibility of the generated stress pattern features over time, a time smoothness constraint term is added to the generator loss function. :

[0166] ;

[0167] in Indicates the synthetic features at time points The value, The weight coefficient for the smoothness constraint ranges from [0.01, 0.1], and is set to 0.05 in this embodiment. The complete generator loss function is as follows: :

[0168] ;

[0169] By optimizing this loss function, we can ensure that the generated stress patterns maintain realism while also satisfying the temporal continuity constraint.

[0170] Furthermore, the discriminator network is a type of feedforward neural network, and its layer structure includes:

[0171] The input layer receives the feature vectors of the stress pattern.

[0172] Multiple fully connected hidden layers, each followed by a Leaky ReLU activation function, extract discriminative representations of features;

[0173] The output layer uses a sigmoid activation function to output probability values ​​between 0 and 1, representing the likelihood that the input data is real data. The discriminator network is trained using a binary cross-entropy loss function. :

[0174] ;

[0175] The goal is to maximize the log probability of the real samples and the inverse log probability of the synthetic samples. To ensure the discriminator's sensitivity to temporal characteristics, a time-dependent constraint is added to the discriminator's loss function. :

[0176] ;

[0177] in and These represent the real features and synthetic features at different time points. The value, and The feature value at the previous time point, This is the weighting coefficient for the time correlation constraint, with a value range of [0.1, 0.5], and is set to 0.3 in this embodiment. The complete discriminator loss function... for:

[0178] ;

[0179] By optimizing this loss function, the discriminator can identify the physical plausibility of stress patterns over time.

[0180] It is understood that data preprocessing methods known to those skilled in the art include data cleaning, data transformation, and data reduction. Data transformation includes type conversion and normalization and standardization. Although the dimensions and types of data were omitted in the description of the preceding embodiments, data preprocessing is a technical knowledge known to those skilled in the art and a prerequisite step in data processing. Therefore, the previously described well-known data preprocessing steps were not described independently.

[0181] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for pressure regulation and monitoring during nasogastric tube insertion, characterized in that, Includes the following steps: Real-time pressure monitoring data during nasogastric tube intubation was acquired. Before generating the pressure pattern feature vector, the real-time pressure monitoring data was preprocessed, including noise reduction, outlier detection and processing, data smoothing and standardization. The standardization process adopted the Z-score standardization method to convert each pressure measurement value into a standard normal distribution with a mean of 0 and a standard deviation of 1. After preprocessing, the pressure pattern feature vector was generated. The stress pattern feature vector is mapped to a low-dimensional hash space using the region-sensitive hashing algorithm to generate hash codes. Based on hash encoding, similar case retrieval is performed in a pre-established multi-level hash index structure, and a candidate set of similar pressure patterns is output. Analyze the retrieved candidate similar pressure patterns and their corresponding historical control strategies. During the analysis, assign time decay weights to historical cases. The weight calculation formula is as follows: ,in Indicates the current time. Timestamps representing historical cases The time decay coefficient ranges from 0.001 to 0.

1. A control strategy recommendation for the current intubation scenario is generated, and pressure is adjusted based on the recommended control strategy. By using collaborative filtering algorithms to analyze the correlation between similar cases, we can identify control strategy combinations that have a success rate exceeding a preset success rate threshold under similar stress patterns and generate control strategy recommendations based on case correlation. Rare stress pattern cases are input into a generative adversarial network (GAN), where a generator in the GAN synthesizes virtual case data. This virtual case data is then integrated into a historical case database. The loss function of the generator in the GAN includes a time smoothness constraint term. ,in Indicates the synthetic features at time points The value, The weight coefficients for smoothness constraints are used to define the loss function of the discriminator in the generative adversarial network, which includes a time-related constraint term to identify the physical plausibility of stress patterns in the time dimension. Obtain real-time feedback data after stress adjustment based on control strategy recommendations, and update the multi-level hash index structure based on the real-time feedback data, including adding the new stress mode corresponding cases and their control effects to the historical case library, and dynamically adjusting the hash function parameters used by the hash tables contained in the multi-level hash index structure. The dynamic adjustment adopts a time-based multi-objective optimization method, and the objective functions include maximizing the retrieval recall rate, minimizing the normalized retrieval time, and maximizing the time relevance score. The region-sensitive hashing algorithm includes a hash function based on random projection and a hash function based on a random hyperplane. The hash function is selected according to the similarity measurement method of the stress pattern feature vector.

2. The method according to claim 1, characterized in that, The pressure mode feature vector includes a pressure mode time-domain feature vector, a pressure mode frequency-domain feature vector, and a correlation feature vector, and is generated by the following steps: The sliding window algorithm is used to calculate the time-domain feature vector of the pressure pattern from the real-time pressure monitoring data, including the pressure mean, variance, peak value and valley value within each window; The Fast Fourier Transform algorithm is used to extract features from real-time pressure monitoring data to obtain the frequency domain feature vector of the pressure mode, including the main frequency component and the spectral energy distribution. The corresponding autocorrelation function and cross-correlation function are calculated based on real-time pressure monitoring data, and a correlation feature vector is generated based on the autocorrelation function and cross-correlation function.

3. The method according to claim 1, characterized in that, The hash function based on random projection Defined as: ； in, For feature vectors, It is a random vector sampled from a standard normal distribution. From Random numbers sampled from a uniform distribution It is the width parameter of the hash bucket. This indicates that a rounding down operation is being performed.

4. The method according to claim 1, characterized in that, The hash function based on the random hyperplane Defined as: ； in, For feature vectors, It is a random vector sampled from a standard normal distribution. Represents a sign function, when Returns 1 when Returns -1 when Returns 0.

5. The method according to claim 1, characterized in that, The pre-established multi-level hash index structure includes Each hash table uses a different combination of hash functions. The value ranges from 2 to 10 and is dynamically adjusted according to the size of the historical case library and the requirements for retrieval efficiency. The multi-level hash index structure is constructed based on the historical pressure pattern feature vectors and their corresponding hash codes in the historical case library.

6. The method according to claim 1, characterized in that, The generation of the candidate similar stress pattern set also includes the following steps: For each similar case in the candidate similar stress pattern set, calculate its similarity score with the stress pattern corresponding to the retrieved hash code; Sort the similarity scores and output the top-ranked similarity scores. The stress patterns of similar cases were used as the final search results, among which The value range is from 5 to 20.

7. The method according to claim 6, characterized in that, The similarity score is calculated using one of the following similarity functions: Euclidean distance function : ； Cosine similarity function : ； Mahalanobis distance function : ； in, The retrieved hash code corresponds to the pressure pattern feature vector. For similar cases, the stress pattern feature vectors are... The covariance matrix of the eigenvectors. express The i-th component, express The i-th component, The dimension of the feature vector. This indicates the transpose operation.

8. The method according to claim 6, characterized in that, The generation of the recommended control strategy includes the following steps: Extract the control strategy parameters used in each similar case, including intubation speed, angle adjustment, and pressure threshold; The parameters of each control strategy are normalized, and the normalized similarity scores of similar cases are used as weights to perform weighted fusion of the normalized control strategy parameters to generate recommended control strategy values ​​suitable for the current intubation scenario. The recommended values ​​of the control strategy are restored to their original dimensions through inverse transformation, and then truncated and adjusted in combination with the effective range of parameters in clinical practice.

9. A pressure regulation and monitoring system for nasogastric tube insertion, used to perform the pressure regulation and monitoring method for nasogastric tube insertion according to any one of claims 1-8, characterized in that, include: The data acquisition module is used to acquire real-time pressure monitoring data during the nasogastric tube insertion process; The feature generation module is used to generate pressure pattern feature vectors based on real-time pressure monitoring data. The hash encoding generation module uses a region-sensitive hashing algorithm to map the stress pattern feature vector to a low-dimensional hash space to generate a hash code; wherein, the region-sensitive hashing algorithm includes a hash function based on random projection and a hash function based on a random hyperplane, and the hash function is selected according to the similarity measurement method of the stress pattern feature vector; The similar case retrieval module performs similar case retrieval based on hash encoding in a pre-established multi-level hash index structure and outputs a set of candidate similar pressure patterns. The control strategy generation and adjustment module is used to analyze the retrieved candidate similar pressure patterns and their corresponding historical control strategies, generate control strategy recommendations for the current intubation scenario, and allow medical staff to adjust the pressure based on the control strategy recommendations. The human-computer interaction interface is used to display search results and control strategy recommendations to medical staff, and to receive operational feedback.

Images