Intelligent analysis and treatment recommendation method and system based on ventilator data

Abstract

The present application relates to the technical field of breathing machine, disclose a kind of intelligent analysis and treatment suggestion method and system based on breathing machine data, comprising: the curve similarity between each curve data in the calculation curve data, utilize curve similarity to carry out problem aggregation to breathing common problem;Curve-problem relationship between the curve data and the set of respiratory problems is constructed;When there is breathing machine failure in the problem set category, the breathing machine is self-detected, and when there is breathing machine failure in the self-detection result;Reduce the data category of monitoring data, construct the numerical adjustment sequence of reduced data category;The improvement effect of value in numerical adjustment sequence on reduced set category is analyzed, and the reward model of splicing data is generated using the improvement effect;The state set corresponding to the reduced data category and the sign data is constructed, and the cumulative reward corresponding to the numerical adjustment sequence is calculated.The present application can improve the data analysis and treatment suggestion of breathing machine.

Description

TECHNICAL FIELD

[0001] The present application relates to a method and system for intelligent analysis and treatment recommendation based on ventilator data, belonging to the technical field of ventilators. BACKGROUND

[0002] As an effective means to artificially replace the self-ventilation function, the ventilator has been widely used in respiratory failure caused by various reasons, anesthesia and respiratory management during major surgery, respiratory support treatment and emergency resuscitation, and occupies a very important position in the field of modern medicine. The ventilator is a vital medical device that can prevent and treat respiratory failure, reduce complications, and save and prolong the lives of patients. The basic setting parameter data of the ventilator and the characteristic data generated by the patient using the ventilator.

[0003] With the progress of medical technology, the ventilator is more and more widely used in clinical practice. However, the traditional ventilator mainly provides mechanical ventilation support, and lacks in-depth analysis of the patient's condition and personalized treatment recommendations. Therefore, how to use the data generated by the ventilator for intelligent analysis to generate accurate treatment recommendations has become a problem to be solved in the current medical field. The user of the ventilator needs to manually upload data or the device automatically uploads data to a digital sleep management platform to generate a digital chart, and a professional person gives a treatment recommendation after interpreting the chart information. The treatment recommendation is a recommendation for adjusting the instrument parameters of the ventilator. When the patient's treatment data is uploaded, the doctor needs to spend a certain amount of time to carefully analyze these data. Since the amount of data can be very large, this makes the analysis work particularly complex and time-consuming, so the doctor cannot immediately give a treatment recommendation. In addition, as the amount of data increases, the required analysis cost will also increase accordingly, which may affect the efficiency and cost-effectiveness of the entire treatment process.

[0004] At present, the data analysis and treatment recommendation of the ventilator are not intelligent enough. SUMMARY

[0005] The present application provides a method and system for intelligent analysis and treatment recommendation based on ventilator data, which aims to improve the intelligence of data analysis and treatment recommendation of the ventilator.

[0006] To achieve the above-mentioned purpose, the present application provides a method for intelligent analysis and treatment recommendation based on ventilator data, which comprises:

[0007] Collecting common respiratory problems of the ventilator, simulating curve data of the common respiratory problems, calculating curve similarity between each curve data in the curve data, using the curve similarity to aggregate the common respiratory problems to obtain a respiratory problem set, wherein the curve data includes respiratory frequency curve, tidal volume curve, airway pressure curve and respiratory ratio curve.

[0008] Construct a curve-problem relationship between the curve data and the set of respiratory problems, collect monitoring data from the ventilator, input the monitoring data into the curve-problem relationship, and output the problem set category corresponding to the monitoring data through the curve-problem relationship. The monitoring data includes respiratory rate, tidal volume, airway pressure, and respiratory ratio.

[0009] Check if any of the aforementioned problem categories contain ventilator malfunctions;

[0010] When a ventilator malfunction exists in the problem set category, a problem self-detection is performed on the ventilator to obtain the self-detection result, and it is determined whether the self-detection result indicates that the ventilator malfunction exists.

[0011] When the self-detection result indicates a ventilator malfunction, after repairing the ventilator malfunction, the current data of the ventilator is collected again, the current problem category corresponding to the current data is identified using the curve-problem relationship, and the process returns to step S3 above.

[0012] When the self-detection results do not show any ventilator malfunction, the number of problems in the problem set category is reduced to obtain a reduced set category. Based on the reduced set category, the data category of the monitoring data is reduced to obtain a reduced data category. The numerical adjustment order of the reduced data category is then constructed.

[0013] Collect vital sign data of patients on the ventilator, analyze the improvement effect of the values ​​in the numerical adjustment order on the reduced set category based on the narrowed data category and the vital sign data, and use the improvement effect to generate a reward model for the spliced ​​data;

[0014] Construct a state set corresponding to the narrowed data category and the vital sign data. Calculate the cumulative reward corresponding to the numerical adjustment order based on the reward model and the state set. Use the cumulative reward to filter the target adjustment order from the numerical adjustment order and determine the treatment recommendation for the ventilator based on the target adjustment order.

[0015] Optionally, calculating the curve similarity between each curve data in the curve data includes:

[0016] Obtain the respiratory rate curve, tidal volume curve, airway pressure curve, and respiratory ratio curve from the curve data;

[0017] The similarity of the first curve corresponding to the respiratory rate curve is calculated using the following formula:

[0018]

[0019] Among them, DKC(x i ,x j ) represents the similarity of the first curve, x i This represents the respiratory rate curve for the i-th common breathing problem, x j This represents the respiratory rate curve for the j-th common breathing problem, where Φ() represents the dynamic kernel-related kernel function, Φ(x i ) represents x i The kernel function, Φ(x) j ) represents x j kernel function, R i R represents the set of factors related to the i-th common breathing problem. j Let represent the set of factors related to the j-th common breathing problem, and α represent the preset similarity threshold;

[0020] Calculate the similarity between the second curve and the tidal volume curve;

[0021] Calculate the similarity of the third curve corresponding to the airway pressure curve;

[0022] Calculate the similarity of the fourth curve corresponding to the respiratory ratio curve;

[0023] The first curve similarity, the second curve similarity, the third curve similarity, and the fourth curve similarity are used as the curve similarity between each curve data in the curve data.

[0024] Optionally, the step of aggregating common breathing problems using the curve similarity to obtain a set of breathing problems includes:

[0025] Obtain the first curve similarity, the second curve similarity, the third curve similarity, and the fourth curve similarity from the curve similarity;

[0026] Determine whether the similarity of the first curve, the second curve, the third curve, and the fourth curve is not less than a preset similarity threshold;

[0027] When the similarity of the first curve, the similarity of the second curve, the similarity of the third curve, and the similarity of the fourth curve are all not less than a preset similarity threshold, the first target common problem corresponding to the similarity of the first curve, the similarity of the second curve, the similarity of the third curve, and the similarity of the fourth curve are obtained from the common breathing problems;

[0028] The common problems of the first objective are aggregated into a set of breathing problems;

[0029] When the similarity of the first curve, the similarity of the second curve, the similarity of the third curve, and the similarity of the fourth curve are not all less than a preset similarity threshold, the second target common problem corresponding to the similarity of the first curve, the similarity of the second curve, the similarity of the third curve, and the similarity of the fourth curve are obtained from the common breathing problems;

[0030] The common problems of the second objective are categorized as a set of breathing problems.

[0031] Optionally, constructing the curve-problem relationship between the curve data and the set of breathing problems includes:

[0032] Obtain the respiratory rate curve, tidal volume curve, airway pressure curve, and respiratory ratio curve corresponding to the curve data;

[0033] The respiratory rate curve, the tidal volume curve, the airway pressure curve, and the respiratory ratio curve are converted into vector matrices.

[0034] Temporal features of each row in the vector matrix are extracted using a gated loop unit;

[0035] The spatial features of each column in the vector matrix are extracted using a convolutional neural network;

[0036] The temporal features and spatial features are spatiotemporally fused using an attention mechanism to obtain fused features;

[0037] Analyze the classification probability values ​​of the fused features with respect to the set of breathing problems;

[0038] The classification probability values ​​are used to identify the curve-problem relationship between the curve data and the set of breathing problems.

[0039] Optionally, the step of performing a self-detection of the ventilator to obtain the self-detection result includes:

[0040] To obtain the ventilator malfunction information;

[0041] The ventilator malfunction is taken as the current state of the ventilator;

[0042] Query the transition states and transition conditions of the current state within the historical time period;

[0043] Based on the transferable conditions, the ventilator is conditionally transferred to obtain the transfer conditions;

[0044] Identify the transfer status of the ventilator under the transfer conditions;

[0045] Determine whether the transition state is consistent with the transferable state;

[0046] When the transfer state is consistent with the transferable state, the existence of the ventilator malfunction will be taken as the self-detection result;

[0047] When the transfer state is inconsistent with the transferable state, the absence of the ventilator malfunction will be taken as the self-detection result.

[0048] Optionally, the step of narrowing down the data categories of the monitoring data based on the narrowed set categories to obtain narrowed data categories includes:

[0049] Obtain the ventilator faults corresponding to the narrowed set categories;

[0050] Based on the ventilator malfunction and the narrowed set categories, the categories to be removed from the monitoring data are identified using the following method:

[0051]

[0052] Where z represents the category to be removed, R k R represents the set of factors related to ventilator malfunction k. k′ This represents the set of relevant factors that narrow down the category k′.

[0053] The categories to be removed are removed from the monitoring data to obtain a narrowed data category.

[0054] Optionally, the step of analyzing the improvement effect of the numerical values ​​in the numerical adjustment order on the reduction of the set categories based on the reduced data categories and the vital sign data includes:

[0055] The reduced set categories are assigned category values ​​to obtain initial category values;

[0056] The following formula is used to construct the input data between the numerical adjustment order, the initial value of the category, the narrowed data category, and the vital sign data:

[0057]

[0058] Where X represents the input data, This represents the value in the numerical adjustment order at round t. This represents the set of values ​​whose categories are reduced in round t, where t=1. X3 represents the initial value of the category, X4 represents the set of narrowed data categories, and X5 represents the set of vital sign data.

[0059] Feature extraction is performed on the input data to obtain the extracted features;

[0060] Identify the hidden vector corresponding to the extracted features;

[0061] Output the next round distribution of the initial values ​​of the categories corresponding to the latent vector, the next round of categories for narrowing down the data categories, and the next round of vital signs for the vital sign data;

[0062] Determine whether there is a next round of values ​​in the numerical adjustment order;

[0063] If a next round value exists in the value adjustment order, then the next round value is selected from the value adjustment order.

[0064] Construct the next round of data between the next round of values, the next round of distribution, the next round of categories, and the next round of vital signs, and return to the above steps of extracting features from the input data to obtain the extracted features;

[0065] If there is no next round value in the numerical adjustment order, the next round distribution is weighted and summed to obtain a weighted sum value;

[0066] The weighted summation value is used as the improvement effect.

[0067] Optionally, constructing the state set corresponding to the narrowed data category and the vital sign data includes:

[0068] The following formula is used to construct the state set corresponding to the narrowed data categories and the vital sign data:

[0069] S = {s1, s2, ..., s} M}

[0070] s m ={s u ,s v}

[0071] Where S represents the set of states, s1, s2, ..., s M This represents the M states corresponding to the narrowed data categories and the vital sign data, s m Represents s1, s2, ..., s M The m-th state in the sequence, s u s represents any set of values ​​corresponding to a narrower data category. v It represents any set of values ​​corresponding to vital signs data.

[0072] Optionally, calculating the cumulative reward corresponding to the numerical adjustment order based on the reward model and the state set includes:

[0073] Based on the reward model and the state set, calculate the action value corresponding to the numerical adjustment order;

[0074] Calculate the state value corresponding to the numerical adjustment order based on the reward model and the state set;

[0075] The cumulative reward corresponding to the numerical adjustment order is determined by using the action value and the state value.

[0076] To address the above problems, the present invention also provides an intelligent analysis and treatment recommendation system based on ventilator data, the system comprising:

[0077] The problem aggregation module is used to collect common respiratory problems of ventilators, simulate curve data of the common respiratory problems, calculate the curve similarity between each curve data in the curve data, and aggregate the common respiratory problems using the curve similarity to obtain a set of respiratory problems. The curve data includes respiratory rate curve, tidal volume curve, airway pressure curve and respiratory ratio curve.

[0078] The category output module is used to construct the curve-problem relationship between the curve data and the set of respiratory problems, collect monitoring data from the ventilator, input the monitoring data into the curve-problem relationship, and output the problem set category corresponding to the monitoring data through the curve-problem relationship. The monitoring data includes respiratory rate, tidal volume, airway pressure, and respiratory ratio.

[0079] The fault query module is used to query whether there are any ventilator faults in the problem set categories;

[0080] The fault diagnosis module is used to perform a self-detection of the ventilator when a ventilator fault exists in the problem set category, obtain a self-detection result, and determine whether the self-detection result indicates that the ventilator fault exists.

[0081] The category identification module is used to, when the self-detection result indicates that the ventilator is faulty, after repairing the ventilator fault, collect the current data of the ventilator again, identify the current problem category corresponding to the current data using the curve-problem relationship, and return to step S3 above;

[0082] The sequence construction module is used to reduce the number of problems in the problem set category when the self-detection result does not show the ventilator fault, to obtain a reduced set category, and based on the reduced set category, reduce the data category of the monitoring data to obtain a reduced data category, and construct the numerical adjustment order of the reduced data category;

[0083] The reward generation module is used to collect the vital sign data of the patients on the ventilator, analyze the improvement effect of the values ​​in the numerical adjustment order on the reduced set category based on the narrowed data category and the vital sign data, and use the improvement effect to generate a reward model for the spliced ​​data.

[0084] The treatment recommendation module is used to construct a state set corresponding to the narrowed data category and the vital sign data, calculate the cumulative reward corresponding to the numerical adjustment order based on the reward model and the state set, use the cumulative reward to filter the target adjustment order from the numerical adjustment order, and determine the treatment recommendation for the ventilator through the target adjustment order.

[0085] Compared to the problems described in the background art, this embodiment of the invention calculates the curve similarity between each curve in the curve data to calculate the similarity between each curve based on the relevant factors of common respiratory problems. This embodiment of the invention constructs a curve-problem relationship between the curve data and the set of respiratory problems to utilize intelligent technologies such as neural networks and deep learning to identify the input-output relationship between the curve data and the set of respiratory problems, thereby improving the intelligence of ventilator data analysis. This embodiment of the invention performs problem self-detection on the ventilator to monitor whether the ventilator's state changes when conditions change; if it does change, then... If the problem set category indicates an instrument malfunction, and the malfunction does not change, it means the instrument malfunction in the problem set category has been repaired. Furthermore, this embodiment of the invention analyzes the improvement effect of the values ​​in the numerical adjustment order on the narrowed set category based on the narrowed data category and the vital sign data. This is used to analyze whether the curve data monitored by the ventilator will still be abnormal after adjusting the values ​​in the numerical adjustment order of the instrument data. It should be noted that the intelligent analysis performed in this invention is based on the condition that the curve data will be abnormal; it does not consider the case where the curve data is normal. That is, when the curve data is normal, there is no need to adjust the instrument parameters of the ventilator. Therefore, the intelligent analysis and treatment suggestion method based on ventilator data proposed in this invention can improve the intelligence of ventilator data analysis and treatment suggestions. Attached Figure Description

[0086] Figure 1 This is a flowchart illustrating an embodiment of the intelligent analysis and treatment suggestion method based on ventilator data provided by the present invention.

[0087] Figure 2 This is a schematic diagram of a module for implementing the intelligent analysis and treatment suggestion method based on ventilator data, according to an embodiment of the present invention.

[0088] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0089] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0090] This application provides a method for intelligent analysis and treatment recommendations based on ventilator data. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for intelligent analysis and treatment recommendations based on ventilator data can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0091] Example 1:

[0092] Reference Figure 1 The diagram shown is a flowchart illustrating an intelligent analysis and treatment recommendation method based on ventilator data according to an embodiment of the present invention. In this embodiment, the intelligent analysis and treatment recommendation method based on ventilator data includes:

[0093] S1. Collect common respiratory problems of ventilators, simulate the curve data of the common respiratory problems, calculate the curve similarity between each curve data in the curve data, and use the curve similarity to aggregate the common respiratory problems to obtain a set of respiratory problems. The curve data includes respiratory rate curve, tidal volume curve, airway pressure curve and respiratory ratio curve.

[0094] In this embodiment of the invention, the common respiratory problems refer to the reasons why the patient's data monitored by the ventilator is abnormal. These include reasons related to the patient's physical condition and reasons related to instrument malfunction. Reasons related to instrument malfunction include malfunctions of the instruments that monitor the respiratory rate curve, tidal volume curve, airway pressure curve, and respiratory ratio curve. For example, air leakage in the breathing tubing can cause abnormalities in the respiratory rate curve, and airway obstruction can cause abnormalities in the airway pressure curve. Reasons related to the patient's physical condition include incomplete exhalation and increased residual volume in the lungs, which can cause abnormalities in the airway pressure curve.

[0095] Optionally, the process of simulating the curve data of the common respiratory problem refers to using a constructed physical patient model and a ventilator to simulate the conditions under which the common respiratory problem occurs, thereby measuring the curve data when the common respiratory problem occurs, wherein the curve data refers to continuous values ​​within a fixed time interval.

[0096] Furthermore, embodiments of the present invention calculate the curve similarity between each curve in the curve data to calculate the similarity between each curve based on relevant factors of common respiratory problems.

[0097] In one embodiment of the present invention, calculating the curve similarity between each curve data in the curve data includes: obtaining the respiratory rate curve, tidal volume curve, airway pressure curve, and respiratory ratio curve in the curve data; and calculating the first curve similarity corresponding to the respiratory rate curve using the following formula:

[0098]

[0099] Among them, DKC(x i ,x j ) represents the similarity of the first curve, x i This represents the respiratory rate curve for the i-th common breathing problem, x j This represents the respiratory rate curve for the j-th common breathing problem, where Φ() represents the dynamic kernel-related kernel function, Φ(x i ) represents x i The kernel function, Φ(x) j ) represents x j kernel function, R i R represents the set of factors related to the i-th common breathing problem. j Let represent the set of factors related to the j-th common breathing problem, and α represent the preset similarity threshold;

[0100] Calculate the second curve similarity corresponding to the tidal volume curve; calculate the third curve similarity corresponding to the airway pressure curve; calculate the fourth curve similarity corresponding to the respiratory ratio curve; and use the first curve similarity, the second curve similarity, the third curve similarity, and the fourth curve similarity as the curve similarity between each curve data in the curve data.

[0101] The set of relevant factors refers to factors that are associated with the common respiratory problems based on grey relational analysis, such as respiratory rate, tidal volume, airway pressure, and respiratory ratio. The preset similarity threshold is a value set according to the actual scenario to filter similarity, which can be set to 0.9 or can be set according to the actual scenario.

[0102] It should be noted that the principles for calculating the similarity of the second curve corresponding to the tidal volume curve, the third curve corresponding to the airway pressure curve, and the fourth curve corresponding to the respiratory ratio curve are similar to those for calculating the similarity of the first curve corresponding to the respiratory rate curve, and will not be elaborated further here. Furthermore, when... When x is an irrelevant factor for both the i-th and j-th common breathing problems, to ensure the smooth execution of the subsequent process of aggregating the common breathing problems using the curve similarity, DKC(x) is defined as follows: i ,x j) = α.

[0103] In one embodiment of the present invention, the step of aggregating common breathing problems using the curve similarity to obtain a set of breathing problems includes: obtaining a first curve similarity, a second curve similarity, a third curve similarity, and a fourth curve similarity from the curve similarity; determining whether the first curve similarity, the second curve similarity, the third curve similarity, and the fourth curve similarity are all not less than a preset similarity threshold; when the first curve similarity, the second curve similarity, the third curve similarity, and the fourth curve similarity are all not less than the preset similarity threshold, obtaining a first target common problem corresponding to the first curve similarity, the second curve similarity, the third curve similarity, and the fourth curve similarity from the common breathing problems; aggregating the first target common problem into a set of breathing problems; when the first curve similarity, the second curve similarity, the third curve similarity, and the fourth curve similarity are not all not less than the preset similarity threshold, obtaining a second target common problem corresponding to the first curve similarity, the second curve similarity, the third curve similarity, and the fourth curve similarity from the common breathing problems; and using the second target common problem as a set of breathing problems.

[0104] Optionally, the process of aggregating the first target common problems into a breathing problem set refers to aggregating multiple first target common problems into a breathing problem set, while the process of using the second target common problems as a breathing problem set refers to using multiple second target common problems as multiple breathing problem sets respectively.

[0105] S2. Construct a curve-problem relationship between the curve data and the set of respiratory problems, collect monitoring data from the ventilator, input the monitoring data into the curve-problem relationship, and output the problem set category corresponding to the monitoring data through the curve-problem relationship. The monitoring data includes respiratory rate, tidal volume, airway pressure, and respiratory ratio.

[0106] This invention constructs a curve-problem relationship between the curve data and the set of breathing problems, so as to use intelligent technologies such as neural networks and deep learning to identify the input-output relationship between the curve data and the set of breathing problems, thereby improving the intelligence of ventilator data analysis.

[0107] The curve-problem relationship refers to a neural network model consisting of gated recurrent units, convolutional neural networks (CNNs), and attention mechanisms.

[0108] In one embodiment of the present invention, constructing the curve-problem relationship between the curve data and the set of breathing problems includes: obtaining the respiratory rate curve, tidal volume curve, airway pressure curve, and respiratory ratio curve corresponding to the curve data; converting the respiratory rate curve, tidal volume curve, airway pressure curve, and respiratory ratio curve into a vector matrix; extracting the temporal features of each row in the vector matrix using a gated recurrent unit; extracting the spatial features of each column in the vector matrix using a convolutional neural network; performing spatiotemporal fusion of the temporal features and the spatial features using an attention mechanism to obtain fused features; analyzing the classification probability value of the fused features with respect to the set of breathing problems; and using the classification probability value to identify the curve-problem relationship between the curve data and the set of breathing problems.

[0109] In this vector matrix, each row represents continuous data from a curve of a certain category, such as continuous data from a respiratory rate curve, and the attention mechanism is a self-attention layer.

[0110] Optionally, the analysis of the classification probability value of the fused features with respect to the set of breathing problems refers to the process of outputting the classification probability using an activation function. The process of identifying the curve-problem relationship between the curve data and the set of breathing problems using the classification probability value refers to the process of obtaining a trained neural network model when the set corresponding to the classification probability value is consistent with the actual set. Here, each classification probability value corresponds to each set of breathing problems, rather than the problems in the set.

[0111] S3. Query whether there is a ventilator malfunction in the problem set category.

[0112] The ventilator malfunction refers to the cause of ventilator instrument failure.

[0113] Optionally, the process of querying whether there is a ventilator malfunction in the problem set category refers to querying whether the set (problem set category) contains either a patient's physical cause or a ventilator instrument malfunction.

[0114] S4. When a ventilator malfunction exists in the problem set category, perform a self-detection of the ventilator to obtain the self-detection result, and determine whether the self-detection result indicates that the ventilator malfunction exists.

[0115] This invention embodiment performs a self-detection of the ventilator to monitor whether the ventilator's status changes when conditions change. If it changes, it indicates that an instrument malfunction exists in the problem set category; if it does not change, it indicates that the instrument malfunction in the problem set category has been repaired.

[0116] In one embodiment of the present invention, the step of performing a self-detection of the ventilator to obtain a self-detection result includes: acquiring a ventilator malfunction; taking the ventilator malfunction as the current state of the ventilator; querying the transferable states and transferable conditions of the current state within a historical period; performing conditional transfer on the ventilator based on the transferable conditions to obtain transfer conditions; identifying the transfer state of the ventilator under the transfer conditions; determining whether the transfer state is consistent with the transferable state; when the transfer state is consistent with the transferable state, taking the existence of the ventilator malfunction as the self-detection result; when the transfer state is inconsistent with the transferable state, taking the absence of the ventilator malfunction as the self-detection result.

[0117] The transferable state refers to the state that can change from the current state to another state, such as another fault or normal state. The transferable condition refers to the condition that allows the transition from the current state to the transferable state, such as the parameters of the ventilator.

[0118] S5. When the self-detection result indicates a ventilator malfunction, after repairing the ventilator malfunction, the current data of the ventilator is collected again, the current problem category corresponding to the current data is identified using the curve-problem relationship, and the process returns to step S3 above.

[0119] Optionally, the process of using the curve-problem relationship to identify the current problem category corresponding to the current data refers to the process of using the data collected by the repaired ventilator again to analyze whether there is a ventilator malfunction, until there is no ventilator malfunction, and then proceeding to step S6.

[0120] S6. When the self-detection result does not show any ventilator malfunction, reduce the number of problems in the problem set category to obtain a reduced set category, and based on the reduced set category, reduce the data category of the monitoring data to obtain a reduced data category, and construct the numerical adjustment order of the reduced data category.

[0121] In this embodiment of the invention, the reduction of the set category refers to the removal of the problem set category after removing the ventilator malfunction.

[0122] In one embodiment of the present invention, the step of narrowing down the data categories of the monitoring data based on the narrowed set categories to obtain narrowed data categories includes: obtaining the ventilator faults corresponding to the narrowed set categories; and identifying the categories to be removed from the monitoring data based on the ventilator faults and the narrowed set categories using the following method:

[0123]

[0124] Where z represents the category to be removed, Rk R represents the set of factors related to ventilator malfunction k. k′ This represents the set of relevant factors that narrow down the category k′.

[0125] The categories to be removed are removed from the monitoring data to obtain a narrowed data category.

[0126] The numerical adjustment order refers to a randomly set sequence of ventilator instrument parameters. For example, the sequence of ventilator instrument parameters includes sequence A, sequence B, and sequence C. A represents Class A parameters of the ventilator instrument, B represents Class A parameters of the ventilator instrument, and C represents Class C parameters of the ventilator instrument. The sequence A is 1, 2, 3, 4, which means that the Class A parameters are adjusted from the current value to 1, then to 2, then to 3, and so on until the adjustment ends at 4.

[0127] S7. Collect the vital signs data of the patient on the ventilator, and based on the narrowed data category and the vital signs data, analyze the improvement effect of the values ​​in the numerical adjustment order on the narrowed set category, and use the improvement effect to generate a reward model for the spliced ​​data.

[0128] The vital signs data refer to the physical condition data of patients undergoing ventilator treatment, such as blood pressure, body temperature, and blood oxygen saturation.

[0129] Furthermore, this embodiment of the invention analyzes the improvement effect of the values ​​in the numerical adjustment order on the reduced set category based on the reduced data category and the vital sign data. This is used to analyze whether the curve data monitored by the ventilator will still be abnormal after adjusting the values ​​in the numerical adjustment order of the instrument data. It should be noted that the intelligent analysis performed by this invention is based on the condition that the curve data will be abnormal, and does not consider the case where the curve data is normal. That is to say, when the curve data is normal, there is no need to adjust the instrument parameters of the ventilator.

[0130] In one embodiment of the present invention, the step of analyzing the improvement effect of the numerical values ​​in the numerical adjustment order on the reduced set category based on the reduced data category and the vital sign data includes: assigning category values ​​to the reduced set category to obtain initial category values; and constructing input data between the numerical adjustment order, the initial category values, the reduced data category, and the vital sign data using the following formula:

[0131]

[0132] Where X represents the input data, This represents the value in the numerical adjustment order at round t. This represents the set of values ​​whose categories are reduced in round t, where t=1. X3 represents the initial value of the category, X4 represents the set of narrowed data categories, and X5 represents the set of vital sign data.

[0133] Feature extraction is performed on the input data to obtain extracted features; the latent vectors corresponding to the extracted features are identified; the next-round distribution of the initial category values ​​corresponding to the latent vectors, the next-round categories of the data categories, and the next-round vital signs of the vital sign data are output; it is determined whether there is a next-round value in the numerical adjustment order; if there is a next-round value in the numerical adjustment order, a next-round value is selected from the numerical adjustment order; the next-round data between the next-round value, the next-round distribution, the next-round categories, and the next-round vital signs is constructed, and the steps of feature extraction from the input data to obtain extracted features are returned; if there is no next-round value in the numerical adjustment order, the next-round distribution is weighted and summed to obtain a weighted sum value; the weighted sum value is used as the improvement effect.

[0134] The initial value of the category is 0. For example, the value of each question in the reduced set category is set to 0. If there are 3 questions in the reduced set category, then 0, 0, 0 are set as the initial values ​​of the category, because the reduced set category is originally due to the curve anomaly, that is, there are no values.

[0135] Optionally, the process of extracting features from the input data to obtain extracted features; identifying the hidden vectors corresponding to the extracted features; and outputting the next-round distribution of the initial class values ​​corresponding to the hidden vectors, the next-round categories of the data, and the next-round vital signs of the vital signs data refers to the process of classifying using a multi-classifier in the input layer, hidden layer, and output layer of a BP neural network. The output next-round distribution is, for example, 1, -2, 2, etc., where positive numbers indicate an improvement of 1 or 2, and negative numbers indicate a decrease of 2. The next-round value refers to the value in the order of adjustment. The next consecutive value.

[0136] The reward model refers to the improvement effect.

[0137] S8. Construct a state set corresponding to the narrowed data category and the vital sign data. Calculate the cumulative reward corresponding to the numerical adjustment order based on the reward model and the state set. Use the cumulative reward to filter the target adjustment order from the numerical adjustment order and determine the treatment recommendation for the ventilator based on the target adjustment order.

[0138] In one embodiment of the present invention, constructing the state set corresponding to the reduced data category and the vital sign data includes: constructing the state set corresponding to the reduced data category and the vital sign data using the following formula:

[0139] S = {s1, s2, ..., s} M}

[0140] s m ={s u ,s v}

[0141] Where S represents the set of states, s1, s2, ..., s M This represents the M states corresponding to the narrowed data categories and the vital sign data, s m Represents s1, s2, ..., s M The m-th state in the sequence, s u s represents any set of values ​​corresponding to a narrower data category. v It represents any set of values ​​corresponding to vital signs data.

[0142] Among them, s u This refers to any set of values ​​corresponding to a narrowed data category. It does not refer to the set within the narrowed data category itself, but rather to any set of values ​​from all possible sets for each narrowed data category. For example, if the narrowed data categories include category A and category B, then s... u This represents the set of any numerical distribution of category A and any numerical distribution of category B, where the numerical distribution refers to continuous data within a fixed time interval, that is, the values ​​on the curve that change over time.

[0143] In one embodiment of the present invention, calculating the cumulative reward corresponding to the numerical adjustment order based on the reward model and the state set includes: calculating the action value corresponding to the numerical adjustment order based on the reward model and the state set; calculating the state value corresponding to the numerical adjustment order based on the reward model and the state set; and determining the cumulative reward corresponding to the numerical adjustment order using the action value and the state value.

[0144] Optionally, the process of determining the cumulative reward corresponding to the numerical adjustment order using the action value and the state value refers to the process of calculating the cumulative reward using the state value function and the action value function, and the reward model acts as the reward function when calculating the state value function and the action value function.

[0145] Optionally, the process of using the cumulative reward to filter the target adjustment order from the numerical adjustment order and determining the treatment recommendation of the ventilator through the target adjustment order refers to selecting the maximum cumulative reward after calculating the cumulative reward corresponding to each numerical adjustment order, and taking the numerical adjustment order corresponding to the maximum cumulative reward as the target adjustment order.

[0146] The treatment recommendations refer to the order in which the parameters of the ventilator are adjusted.

[0147] Compared to the problems described in the background art, this embodiment of the invention calculates the curve similarity between each curve in the curve data to calculate the similarity between each curve based on the relevant factors of common respiratory problems. This embodiment of the invention constructs a curve-problem relationship between the curve data and the set of respiratory problems to utilize intelligent technologies such as neural networks and deep learning to identify the input-output relationship between the curve data and the set of respiratory problems, thereby improving the intelligence of ventilator data analysis. This embodiment of the invention performs problem self-detection on the ventilator to monitor whether the ventilator's state changes when conditions change; if it does change, then... If the problem set category indicates an instrument malfunction, and the malfunction does not change, it means the instrument malfunction in the problem set category has been repaired. Furthermore, this embodiment of the invention analyzes the improvement effect of the values ​​in the numerical adjustment order on the narrowed set category based on the narrowed data category and the vital sign data. This is used to analyze whether the curve data monitored by the ventilator will still be abnormal after adjusting the values ​​in the numerical adjustment order of the instrument data. It should be noted that the intelligent analysis performed in this invention is based on the condition that the curve data will be abnormal; it does not consider the case where the curve data is normal. That is, when the curve data is normal, there is no need to adjust the instrument parameters of the ventilator. Therefore, the intelligent analysis and treatment suggestion method based on ventilator data proposed in this invention can improve the intelligence of ventilator data analysis and treatment suggestions.

[0148] Example 2:

[0149] like Figure 2 The diagram shown is a functional block diagram of an intelligent analysis and treatment suggestion system based on ventilator data according to the present invention.

[0150] The intelligent analysis and treatment suggestion system 200 based on ventilator data described in this invention can be installed in an electronic device. Depending on the functions implemented, the intelligent analysis and treatment suggestion system based on ventilator data may include a problem aggregation module 201, a category output module 202, a fault query module 203, a fault judgment module 204, a category identification module 205, a sequence construction module 206, a reward generation module 207, and a treatment suggestion module 208. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.

[0151] In this embodiment of the invention, the functions of each module / unit are as follows:

[0152] The problem aggregation module 201 is used to collect common respiratory problems of ventilators, simulate curve data of the common respiratory problems, calculate the curve similarity between each curve data in the curve data, and aggregate the common respiratory problems using the curve similarity to obtain a set of respiratory problems. The curve data includes respiratory rate curve, tidal volume curve, airway pressure curve and respiratory ratio curve.

[0153] The category output module 202 is used to construct a curve-problem relationship between the curve data and the set of respiratory problems, collect monitoring data from the ventilator, input the monitoring data into the curve-problem relationship, and output the problem set category corresponding to the monitoring data through the curve-problem relationship. The monitoring data includes respiratory rate, tidal volume, airway pressure, and respiratory ratio.

[0154] The fault query module 203 is used to query whether there is a ventilator fault in the problem set category;

[0155] The fault judgment module 204 is used to perform a self-detection of the ventilator when there is a ventilator fault in the problem set category, obtain a self-detection result, and determine whether the self-detection result indicates that there is a ventilator fault.

[0156] The category identification module 205 is used to, when the self-detection result shows that the ventilator has a fault, after repairing the ventilator fault, collect the current data of the ventilator again, use the curve-problem relationship to identify the current problem category corresponding to the current data, and return to the above S3 step;

[0157] The sequence construction module 206 is used to reduce the number of problems in the problem set category when the self-detection result does not show the ventilator fault, to obtain a reduced set category, and based on the reduced set category, reduce the data category of the monitoring data to obtain a reduced data category, and construct the numerical adjustment order of the reduced data category;

[0158] The reward generation module 207 is used to collect the vital signs data of the patient on the ventilator, analyze the improvement effect of the values ​​in the numerical adjustment order on the reduced set category based on the narrowed data category and the vital signs data, and generate a reward model for the spliced ​​data using the improvement effect.

[0159] The treatment recommendation module 208 is used to construct a state set corresponding to the narrowed data category and the vital sign data, calculate the cumulative reward corresponding to the numerical adjustment order based on the reward model and the state set, use the cumulative reward to filter the target adjustment order from the numerical adjustment order, and determine the treatment recommendation for the ventilator through the target adjustment order.

[0160] In detail, the modules in the intelligent analysis and treatment suggestion system 200 based on ventilator data described in this embodiment of the invention employ the same methods as described above. Figure 1 The method used is the same as the intelligent analysis and treatment recommendation method based on ventilator data described in the previous article, and can produce the same technical effect, so it will not be elaborated here.

[0161] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for intelligent analysis and treatment recommendations based on ventilator data, characterized in that, The method includes: S1. Collect common respiratory problems of ventilators, simulate the curve data of the common respiratory problems, calculate the curve similarity between each curve data in the curve data, and use the curve similarity to aggregate the common respiratory problems to obtain a set of respiratory problems. The curve data includes respiratory rate curve, tidal volume curve, airway pressure curve and respiratory ratio curve. S2. Construct a curve-problem relationship between the curve data and the set of respiratory problems, collect monitoring data from the ventilator, input the monitoring data into the curve-problem relationship, and output the problem set category corresponding to the monitoring data through the curve-problem relationship. The monitoring data includes respiratory rate, tidal volume, airway pressure, and respiratory ratio. S3. Query whether there is a ventilator malfunction in the aforementioned problem set category; S4. When a ventilator malfunction exists in the problem set category, perform a problem self-detection on the ventilator, obtain the self-detection result, and determine whether the self-detection result indicates that the ventilator malfunction exists. S5. When the self-detection result indicates a ventilator malfunction, after repairing the ventilator malfunction, collect the current data of the ventilator again, use the curve-problem relationship to identify the current problem category corresponding to the current data, and return to step S3 above. S6. When the self-test result does not show any ventilator malfunction, reduce the number of problems in the problem set category to obtain a reduced set category, and based on the reduced set category, reduce the data category of the monitoring data to obtain a reduced data category. Construct the numerical adjustment order of the reduced data category, where the numerical adjustment order refers to a sequence of randomly set ventilator instrument parameters. S7. Collect the vital signs data of the patients on the ventilator, and based on the reduced data category and the vital signs data, analyze the improvement effect of the values ​​in the numerical adjustment order on the reduced set category, and use the improvement effect to generate a reward model; S8. Construct a state set corresponding to the narrowed data category and the vital sign data. Calculate the cumulative reward corresponding to the numerical adjustment order based on the reward model and the state set. Use the cumulative reward to filter the target adjustment order from the numerical adjustment order and determine the treatment recommendation for the ventilator based on the target adjustment order.

2. The intelligent analysis and treatment recommendation method based on ventilator data as described in claim 1, characterized in that, The calculation of curve similarity between each curve data in the curve data includes: Obtain the respiratory rate curve, tidal volume curve, airway pressure curve, and respiratory ratio curve from the curve data; The similarity of the first curve corresponding to the respiratory rate curve is calculated using the following formula: in, Indicates the similarity of the first curve. Indicates the first Respiratory rate curves for common breathing problems Indicates the first Respiratory rate curves for common breathing problems This represents the kernel function related to the dynamic kernel. express kernel function, express kernel function, Indicates the first A collection of factors related to common breathing problems Indicates the first A collection of factors related to common breathing problems This indicates a preset similarity threshold; Calculate the similarity between the second curve and the tidal volume curve; Calculate the similarity of the third curve corresponding to the airway pressure curve; Calculate the similarity of the fourth curve corresponding to the respiratory ratio curve; The first curve similarity, the second curve similarity, the third curve similarity, and the fourth curve similarity are used as the curve similarity between each curve data in the curve data.

3. The intelligent analysis and treatment recommendation method based on ventilator data as described in claim 1, characterized in that, The method of aggregating common breathing problems using the curve similarity to obtain a set of breathing problems includes: Obtain the first curve similarity, the second curve similarity, the third curve similarity, and the fourth curve similarity from the curve similarity; Determine whether the similarity of the first curve, the second curve, the third curve, and the fourth curve is not less than a preset similarity threshold; When the similarity of the first curve, the similarity of the second curve, the similarity of the third curve, and the similarity of the fourth curve are all not less than a preset similarity threshold, the first target common problem corresponding to the similarity of the first curve, the similarity of the second curve, the similarity of the third curve, and the similarity of the fourth curve are obtained from the common breathing problems; The common problems of the first objective are aggregated into a set of breathing problems; When the similarity of the first curve, the similarity of the second curve, the similarity of the third curve, and the similarity of the fourth curve are not all less than a preset similarity threshold, the second target common problem corresponding to the similarity of the first curve, the similarity of the second curve, the similarity of the third curve, and the similarity of the fourth curve are obtained from the common breathing problems; The common problems of the second objective are categorized as a set of breathing problems.

4. The intelligent analysis and treatment recommendation method based on ventilator data as described in claim 1, characterized in that, The construction of the curve-problem relationship between the curve data and the set of breathing problems includes: Obtain the respiratory rate curve, tidal volume curve, airway pressure curve, and respiratory ratio curve corresponding to the curve data; The respiratory rate curve, the tidal volume curve, the airway pressure curve, and the respiratory ratio curve are converted into vector matrices. Temporal features of each row in the vector matrix are extracted using a gated loop unit; The spatial features of each column in the vector matrix are extracted using a convolutional neural network; The temporal features and spatial features are spatiotemporally fused using an attention mechanism to obtain fused features; Analyze the classification probability values ​​of the fused features with respect to the set of breathing problems; The classification probability values ​​are used to identify the curve-problem relationship between the curve data and the set of breathing problems.

5. The intelligent analysis and treatment recommendation method based on ventilator data as described in claim 1, characterized in that, The self-detection of the ventilator to obtain the self-detection results includes: To obtain the ventilator malfunction information; The ventilator malfunction is taken as the current state of the ventilator; Query the transition states and transition conditions of the current state within the historical time period; Based on the transferable conditions, the ventilator is conditionally transferred to obtain the transfer conditions; Identify the transfer status of the ventilator under the transfer conditions; Determine whether the transition state is consistent with the transferable state; When the transfer state is consistent with the transferable state, the existence of the ventilator malfunction will be taken as the self-detection result; When the transfer state is inconsistent with the transferable state, the absence of the ventilator malfunction will be taken as the self-detection result.

6. The intelligent analysis and treatment recommendation method based on ventilator data as described in claim 1, characterized in that, The process of narrowing down the data categories of the monitoring data based on the narrowed set categories to obtain narrowed data categories includes: Obtain the ventilator faults corresponding to the narrowed set categories; Based on the ventilator malfunction and the narrowed set categories, the categories to be removed from the monitoring data are identified using the following method: in, Indicates the category to be removed. Indicates ventilator malfunction The set of relevant factors This indicates a reduction in the size of the collection. The set of relevant factors; The categories to be removed are removed from the monitoring data to obtain a narrowed data category.

7. The intelligent analysis and treatment recommendation method based on ventilator data as described in claim 1, characterized in that, The step of analyzing the improvement effect of the numerical values ​​in the numerical adjustment order on the reduced set categories based on the reduced data categories and the vital sign data includes: The reduced set categories are assigned category values ​​to obtain initial category values; The following formula is used to construct the input data between the numerical adjustment order, the initial value of the category, the narrowed data category, and the vital sign data: in, Indicates input data, This represents the value in the numerical adjustment order at round t. This represents the set of values ​​whose categories are reduced in round t. hour, Indicates the initial value of the category. This represents a set that narrows down the data categories. This represents a set of vital signs data; Feature extraction is performed on the input data to obtain the extracted features; Identify the hidden vector corresponding to the extracted features; Output the next round distribution of the initial values ​​of the categories corresponding to the latent vector, the next round of categories for narrowing down the data categories, and the next round of vital signs for the vital sign data; Determine whether there is a next round of values ​​in the numerical adjustment order; If a next round value exists in the value adjustment order, then the next round value is selected from the value adjustment order. Construct the next round of data between the next round of values, the next round of distribution, the next round of categories, and the next round of vital signs, and return to the above steps of extracting features from the input data to obtain the extracted features; If there is no next round value in the numerical adjustment order, the next round distribution is weighted and summed to obtain a weighted sum value; The weighted summation value is used as the improvement effect.

8. The intelligent analysis and treatment recommendation method based on ventilator data as described in claim 1, characterized in that, The construction of the state set corresponding to the narrowed data category and the vital sign data includes: The following formula is used to construct the state set corresponding to the narrowed data categories and the vital sign data: in, Represents a set of states. This indicates that the narrowed data category corresponds to the vital sign data. a state, express The first in A state, This represents any set of values ​​corresponding to a narrower data category. It represents any set of values ​​corresponding to vital signs data.

9. The intelligent analysis and treatment recommendation method based on ventilator data as described in claim 1, characterized in that, The step of calculating the cumulative reward corresponding to the numerical adjustment order based on the reward model and the state set includes: Based on the reward model and the state set, calculate the action value corresponding to the numerical adjustment order; Calculate the state value corresponding to the numerical adjustment order based on the reward model and the state set; The cumulative reward corresponding to the numerical adjustment order is determined by using the action value and the state value.

10. A smart analysis and treatment recommendation system based on ventilator data, characterized in that, The system includes: The problem aggregation module is used to collect common respiratory problems of ventilators, simulate curve data of the common respiratory problems, calculate the curve similarity between each curve data in the curve data, and aggregate the common respiratory problems using the curve similarity to obtain a set of respiratory problems. The curve data includes respiratory rate curve, tidal volume curve, airway pressure curve and respiratory ratio curve. The category output module is used to construct the curve-problem relationship between the curve data and the set of respiratory problems, collect monitoring data from the ventilator, input the monitoring data into the curve-problem relationship, and output the problem set category corresponding to the monitoring data through the curve-problem relationship. The monitoring data includes respiratory rate, tidal volume, airway pressure, and respiratory ratio. The fault query module is used to query whether there are any ventilator faults in the problem set categories; The fault diagnosis module is used to perform a self-detection of the ventilator when a ventilator fault exists in the problem set category, obtain a self-detection result, and determine whether the self-detection result indicates that the ventilator fault exists. The category identification module is used to, when the self-detection result indicates that the ventilator is faulty, after repairing the ventilator, collect the current data of the ventilator again, identify the current problem category corresponding to the current data using the curve-problem relationship, and return to the above step of querying whether there is a ventilator fault in the problem set category; The sequence construction module is used to reduce the number of problems in the problem set category when the self-test result does not show the ventilator fault, to obtain a reduced set category, and based on the reduced set category, reduce the data category of the monitoring data to obtain a reduced data category, and construct the numerical adjustment order of the reduced data category, wherein the numerical adjustment order refers to the sequence of randomly set ventilator instrument parameters; The reward generation module is used to collect the vital sign data of the patients on the ventilator, analyze the improvement effect of the values ​​in the numerical adjustment order on the reduced set category based on the narrowed data category and the vital sign data, and generate a reward model using the improvement effect; The treatment recommendation module is used to construct a state set corresponding to the narrowed data category and the vital sign data, calculate the cumulative reward corresponding to the numerical adjustment order based on the reward model and the state set, use the cumulative reward to filter the target adjustment order from the numerical adjustment order, and determine the treatment recommendation for the ventilator through the target adjustment order.

Images