Multi-channel medicine collaborative intelligent infusion system for department of cardiology

Abstract

The invention relates to the technical field of data processing, in particular to a multichannel medicine collaborative intelligent infusion system for the department of cardiology. The system determines the clinical similarity according to the clinical feature index of the current object and the clinical feature index of each historical object; determining the historical cases with the clinical similarity exceeding a first threshold value as a reference historical case set; aiming at each historical case in the reference historical case set; according to the real-time physiological monitoring data in the current time period, the physiological monitoring data of the historical cases in the historical similar time period, and the fluctuation difference between the physiological monitoring data of the historical cases in the historical similar time period and the physiological monitoring data of the historical cases in the next adjacent historical time period; according to the difference between the real-time infusion parameter of the current time period and the infusion parameter of the historical case in the historical similar time period, determining an adjustment weight; weighted fusion is carried out on infusion parameters in historical similar time periods by adjusting weights, target infusion parameters are determined, and therefore it is ensured that the target infusion parameters are fused with the quintessence of historical optimal cases.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically to a multi-channel intelligent drug infusion system for cardiology. Background Technology

[0002] In the clinical treatment of critically ill patients in the Department of Cardiology, it is often necessary to administer multiple drugs simultaneously via a multi-channel fluid infusion system, such as anticoagulants, cardiotonics, and vasoactive drugs. However, in scenarios where precise control of the synergistic delivery of multiple drugs is required, although the system contains multiple independent delivery channels, the drugs delivered by each channel may mix upstream of the system, resulting in physical or chemical interference, which may lead to denaturation of the drug medium, blockage of pipelines, or damage to equipment.

[0003] In the field of multi-channel drug infusion, it is often necessary to simultaneously and independently control the infusion flow rate of multiple drug media and ensure that the drug media are combined in a precise ratio at the terminal. Existing multi-channel control systems usually operate based on preset fixed infusion parameters. However, due to the physical or chemical interference between drug media, temperature, pH value, or the complex physiological environment in the patient's body during the infusion process, the state of the infusion system itself is dynamic. As a result, this method makes it impossible for existing control systems to maintain a stable and optimal infusion state throughout the entire delivery cycle. At the same time, this also requires operators to frequently monitor the system output status and manually adjust the control parameters of multiple channels, making it difficult to cope with rapid changes in the system, which may lead to control failure or the system state exceeding the safe range. Summary of the Invention

[0004] To address the technical problem that existing multi-channel control systems typically operate based on preset fixed infusion parameters, resulting in their inability to maintain stable and optimal infusion status throughout the entire delivery cycle, this invention provides a multi-channel intelligent drug-assisted infusion system for cardiology. The specific technical solution adopted is as follows: This invention proposes a multi-channel intelligent drug infusion system for cardiology, the system comprising: The acquisition module is used to acquire the current subject's clinical characteristic indicators, real-time physiological monitoring data, and real-time infusion parameters of each channel after the infusion process begins; and to acquire historical case data containing the clinical characteristic indicators, physiological monitoring data, and infusion parameters of historical subjects. The first determination module is used to determine the clinical similarity based on the clinical characteristic indicators of the current object and the clinical characteristic indicators of each historical object; and to determine historical cases with clinical similarity exceeding a first threshold as a set of reference historical cases. The weight adjustment module is used to determine the real-time similarity of each historical case in the reference historical case set based on the real-time physiological monitoring data in the current time period and the physiological monitoring data of historical cases in similar historical time periods; and to determine the adjustment weight for the infusion parameters of historical cases based on the fluctuation differences between the physiological monitoring data of historical cases in similar historical time periods and the next adjacent historical time period, the differences between the real-time infusion parameters of the current time period and the infusion parameters of historical cases in similar historical time periods, and the real-time similarity. The second determination module is used to perform weighted fusion of infusion parameters in similar historical time periods by adjusting the weights, and to determine the target infusion parameters for the next adjacent time period of the current object; The verification module is used to verify whether the target infusion parameters are within the preset safety boundary. If so, it controls the multi-channel infusion system to run according to the target infusion parameters in the next adjacent time period.

[0005] Furthermore, the clinical similarity determination process includes: Numerical preprocessing is performed on the clinical characteristic indicators of the current and historical subjects to form their respective feature vectors; Calculate the distance between the feature vector of the current object and the feature vector of each historical object; Clinical similarity is determined based on distance, where distance is negatively correlated with clinical similarity.

[0006] Furthermore, the clinical feature indicators include numerical indicators, textual indicators, and image / curve indicators; the numerical preprocessing of the clinical feature indicators of the current object and historical objects includes: For numerical indicators, normalization is performed; For text-based metrics, a pre-defined bag-of-words model is used to convert the text-based metrics into numerical vectors. For image and curve indicators, feature vectors of image and curve indicators are extracted through a preset neural network.

[0007] Furthermore, the real-time similarity determination process includes: Real-time physiological monitoring data and historical physiological monitoring data are divided into corresponding data sequences based on the same time period. For various physiological monitoring data items, the data sequence of the current time period is matched with the data sequence of similar historical time periods to identify two target data sequences containing the same physiological monitoring data item; the dynamic time warping algorithm is used to calculate the warping distance between the two target data sequences, and the warping distance is mapped to the sub-similarity of the corresponding physiological monitoring data item; By using the preset weights of each physiological monitoring data item, the sub-similarity of each physiological monitoring data item is calculated by weighted average to obtain the real-time similarity.

[0008] Furthermore, before determining the two target data sequences, the method includes: Extract complete physiological monitoring data items that are common to the current object and historical cases to form a set of valid data items; The determination of two target data sequences containing the same physiological monitoring data item is limited to the determination of physiological monitoring data items in the set of valid data items.

[0009] Furthermore, the physiological monitoring data includes at least one physiological monitoring data item, and the weight adjustment determination process includes: Obtain the preset safe value range for various physiological monitoring data items in the valid data item set; For various physiological monitoring data items in the valid data item set, a consistency score is determined based on the similarity between the physiological monitoring data items of historical cases in the next adjacent historical period and their preset safe value range; Based on the fluctuation differences between the physiological monitoring data of historical cases in similar historical periods and in the next adjacent historical period, and combined with the consistency score, individual high-quality indicators are determined; using the preset first weight coefficient, the individual quality of all physiological monitoring data items in the effective data item set is weighted and summed to obtain the historical validity score. The degree of parameter difference is determined based on the differences between the real-time infusion parameters of the current period and the infusion parameters of historical cases in similar periods in the past. The product of the historical validity score and the real-time similarity is calculated as the first intermediate value; the sum of the positive integer 1 and the parameter difference is calculated as the second intermediate value; the ratio of the first intermediate value to the second intermediate value is normalized to obtain the adjusted weight.

[0010] Furthermore, the determination of a consistency score based on the similarity between physiological monitoring data items of historical cases in the next adjacent historical period and their preset safety value range includes: Obtain all physiological monitoring data items in the next adjacent historical period in a time-series order, and count the total number of data items for all physiological monitoring data items; Determine whether the value of each physiological monitoring data item is within the preset safe value range for the physiological monitoring data item; Count the target number of all physiological monitoring data items that are within the preset safety value range; The ratio of the number of targets to the total number of data items is used as the consistency score.

[0011] Furthermore, based on the fluctuation differences in physiological monitoring data of historical cases in similar historical periods and in the next adjacent historical period, and combined with a consistency score, individual high-quality indicators are determined, including: The standard deviation of all physiological monitoring data items in the valid data set within a similar historical time period is calculated as the first degree of fluctuation; the standard deviation of all physiological monitoring data items in the next adjacent historical time period is calculated as the second degree of fluctuation. The absolute difference between the second fluctuation level and the first fluctuation level is normalized to obtain the fluctuation level increase. Calculate the difference between the positive integer 1 and the increase in volatility, and use it as the first difference; calculate the product of the first difference and the consistency score, and use it as a single quality indicator.

[0012] Furthermore, determining the parameter difference degree based on the differences between the real-time infusion parameters of the current time period and the infusion parameters of historical cases in similar historical time periods includes: The real-time infusion parameters of the current time period are constructed as the first parameter vector, and the infusion parameters of historical cases in similar historical time periods are constructed as the second parameter vector. Each dimension in the vector represents the parameter value of an infusion channel. Calculate the distance between the first parameter vector and the second parameter vector, which is used as the parameter difference.

[0013] Furthermore, after verification and before controlling the operation of the multi-channel infusion system, the method further includes: Based on the target infusion parameters and the pre-set drug concentration in the infusion system, the target infusion flow rate to be issued is determined; Control the multi-channel infusion system to operate at the target infusion flow rate.

[0014] The present invention has the following beneficial effects: This invention, based on real-time physiological monitoring data and historical case data, automatically determines and sets the target infusion parameters for the next time period, significantly reducing reliance on frequent manual intervention by operators, achieving a higher degree of automation, and effectively saving labor costs. By analyzing real-time similarity, the fluctuation differences between physiological monitoring data of historical cases in similar historical time periods and in the next adjacent historical time period, and the differences between the real-time infusion parameters of the current time period and the infusion parameters of historical cases in similar historical time periods, the system identifies cases in historical cases that achieved successful control effects under similar dynamic states. This allows the system to learn the dynamic law of "what combination of infusion parameters, under what system state, can lead to a good subsequent state of the system," thereby generating coordinated, historically validated target infusion parameters. By dynamically assigning adjustment weights to different historical cases, the system achieves refined knowledge extraction and quantification of the degree of improvement of the system state by infusion parameters from massive amounts of historical similar time periods, ensuring that the final target infusion parameters incorporate the essence of the best historical cases while taking into account the matching degree with the current system state and the safety of the adjustment process. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of a multi-channel intelligent drug synergistic infusion system for cardiology, provided in one embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the weight adjustment determination process provided in one embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a multi-channel intelligent drug infusion system for cardiology proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details a specific solution for a multi-channel intelligent drug infusion system for cardiology provided by the present invention.

[0020] Please see Figure 1 The diagram illustrates a schematic of a multi-channel intelligent drug synergistic infusion system for cardiology, according to an embodiment of the present invention. The system includes: The acquisition module 101 is used to acquire the current subject's clinical characteristic indicators, real-time physiological monitoring data, and real-time infusion parameters of each channel after the infusion process begins; and to acquire historical case data containing historical subjects' clinical characteristic indicators, physiological monitoring data, and infusion parameters.

[0021] Clinical characteristic indicators refer to the static characteristics of the current subject that are relevant to the infusion decision and are obtained through the electronic medical record system before or at the start of infusion (i.e., intravenous infusion). Clinical characteristic indicators are not real-time changes. For example, clinical characteristic indicators include, but are not limited to, the subject's demographic information (such as age and weight), primary diagnostic information (standardized coding, such as ICD-10), disease severity classification labels (such as Killip classification, NYHA functional classification), and key laboratory test results (such as troponin and B-type natriuretic peptide levels).

[0022] The electronic medical record system is a digital, highly detailed personal health information archive. It consists of electronic medical information generated and recorded by medical institutions in accordance with the law during the provision of compliant medical services. The data transmission, storage, and access of the system comply with national regulations on health and medical data security and take necessary encryption and security measures to prevent data leakage.

[0023] The static characteristics of an object refer to a "snapshot" of the object's individual condition at the point in time when the infusion begins.

[0024] Physiological monitoring data refers to time-series signal data that reflects the real-time physiological state of the subject, continuously or frequently collected by medical monitoring devices connected to the subject during the infusion process. Examples include core vital signs such as arterial blood pressure, heart rate, blood oxygen saturation, and body temperature.

[0025] Infusion parameters are a higher-level functional concept, referring to the core variables used by the system to control the execution of infusion equipment and ultimately affect the delivery intensity of the infusion medium (i.e., the drug medium). It should be noted that in practical implementation, infusion parameters can be reflected in the following two levels of data: infusion flow rate: referring to the instructions directly received and executed by the infusion pump, i.e., the volume of fluid delivered per unit time, usually measured in milliliters per hour (mL / h); and drug delivery rate: referring to the mass of the active pharmaceutical ingredient delivered per unit time, usually measured in micrograms per minute (mcg / min), which is the core decision target for clinically characterizing treatment intensity.

[0026] It is important to understand that the drug concentration may vary in different cases. Directly comparing infusion rates is like comparing only the "flow rate" while ignoring the "components of drug concentration," which cannot truly reflect the therapeutic intensity of the drug. To ensure the fairness of the comparison and the clinical relevance of the decision, it can be uniformly converted to dosing rate. This is equivalent to placing all historical cases under the same "pharmacological ruler" for measurement. Therefore, to ensure the accuracy and comparability of intelligent decision-making, the system uniformly uses the dosing rate in subsequent internal core calculations, and the infusion rate is only used as the final instruction executed by the device.

[0027] The historical case database stores complete records of a large number of past infusion processes. Each historical case record contains three types of data corresponding to the aforementioned "current real-time data": clinical characteristic indicators of the historical subject, physiological monitoring data sequence of the entire infusion process, and records of infusion parameters of each channel changing over time during the entire infusion process.

[0028] The first determining module 102 is used to determine the clinical similarity based on the clinical characteristic indicators of the current object and the clinical characteristic indicators of each historical object; and to determine historical cases with clinical similarity exceeding a first threshold as a set of reference historical cases.

[0029] It's important to understand that, for example, the infusion parameters for a patient who needs antibiotics for "pneumonia" are incomparable to those for a patient who needs vasoactive drugs for "acute myocardial infarction." Moreover, the historical case data can be extremely vast. Therefore, in order to immediately eliminate a large number of irrelevant cases, the subsequent complex real-time computing resources are concentrated on a few high-value reference cases. This ensures that the historical cases referenced by the system are similar to those of the current patient, with similar severity. By comparing clinical characteristic indicators, the clinical rationality and relevance of subsequent parameter references are fundamentally guaranteed.

[0030] In this embodiment, the clinical feature indicators of the current object and historical objects are preprocessed numerically to form their respective feature vectors; the distance between the feature vector of the current object and the feature vector of each historical object is calculated; and the clinical similarity is determined based on the distance, wherein the distance is negatively correlated with the clinical similarity.

[0031] It should be noted that clinical feature indicators include numerical indicators, textual indicators, and image / curve indicators. Therefore, in order to convert the clinical feature indicators of each object into a numerical feature vector, numerical preprocessing is performed on the clinical feature indicators of the current object and historical objects: for numerical indicators, normalization is performed; for textual indicators, textual indicators are converted into numerical vectors using a pre-set bag-of-words model; for image and curve indicators, feature vectors of image and curve indicators are extracted using a pre-set neural network.

[0032] For example, for numerical indicators (such as age, weight, and troponin levels), the original values ​​can be scaled to a uniform range (e.g., [0, 1]) to ensure that all numerical features have equal importance in clinical similarity assessment. For textual indicators (such as the diagnostic information "acute anterior wall myocardial infarction"), bag-of-words model natural language processing technology can be used to convert the text description into a numerical vector, so that qualitative medical diagnoses can be transformed into mathematical expressions that computers can understand and compare. For image / curve indicators (such as electrocardiograms), a pre-set neural network can be used to extract high-level features and output a fixed-length feature vector, so as to extract the essential features most relevant to clinical judgment from complex image or signal data.

[0033] It should be noted that the bag-of-words model is a simple and intuitive method for text feature representation. Its core idea is to treat text as a collection of words, ignoring word order and grammatical structure, focusing only on the frequency or absence of words. The specific process of converting text descriptions into numerical vectors using the bag-of-words model is a well-known technique and will not be elaborated upon in this embodiment. For example, suppose a standard vocabulary is pre-defined as ["myocardial infarction", "infarction", "acute", "anterior wall", "heart failure", "hypertension", ...]. This standard vocabulary is the bag of words. Suppose there is a text-based indicator for "acute anterior wall myocardial infarction". Then, the indicator "acute anterior wall myocardial infarction" is compared word-by-word in the standard vocabulary. If the word "acute anterior wall myocardial infarction" appears in the diagnosis, it is marked as 1 at the corresponding position; otherwise, it is marked as 0.

[0034] It should be noted that the process of using a preset neural network to extract high-level features and output feature vectors is a well-known technique in the art, and will not be described in detail in this embodiment.

[0035] It should be noted that Manhattan distance or Euclidean distance can be used to calculate the distance between the feature vector of the current object and the feature vector of each historical object. The method of using Manhattan distance or Euclidean distance to calculate the distance is a common technical means, and will not be described in detail in this embodiment.

[0036] Preferably, as one possible implementation, clinical similarity can be expressed by the following formula: Clinical similarity = 1 - Normalized (distance).

[0037] It should be noted that the distance is normalized using a minimum-maximum normalization method. For example, the system calculates the original distances between the current object and 5 historical cases, which are [2, 8, 15, 5, 10], where the minimum distance is 2 and the maximum distance is 15. The normalized distance is normalized using the core formula: (original distance - current minimum distance) ÷ (current maximum distance - current minimum distance).

[0038] It is important to understand that the smaller the distance value, the closer the two objects are in the feature space, meaning that their clinical characteristics are more similar.

[0039] The weight adjustment module 103 is used to determine the real-time similarity of each historical case in the reference historical case set based on the real-time physiological monitoring data in the current period and the physiological monitoring data of historical cases in similar historical periods; and to determine the adjustment weight for the infusion parameters of historical cases based on the fluctuation differences between the physiological monitoring data of historical cases in similar historical periods and the next adjacent historical period, the differences between the real-time infusion parameters of the current period and the infusion parameters of historical cases in similar historical periods, and the real-time similarity.

[0040] It is important to understand that initial static feature similarity alone is not enough. It is also necessary to ensure that the current object remains in a good state during the infusion process. Therefore, real-time similarity can enable the system to find historical cases that are not only similar at the starting point, but also highly synchronized with the current object in terms of current physiological response, ensuring that subsequent reference data is more accurate and timely.

[0041] In this embodiment, real-time physiological monitoring data and historical physiological monitoring data are divided into corresponding data sequences based on the same time period. For various physiological monitoring data items, the data sequence of the current time period is matched with the data sequences of similar historical time periods to determine two target data sequences containing the same physiological monitoring data item. The dynamic time warping algorithm is used to calculate the warping distance between the two target data sequences, and the warping distance is mapped to the sub-similarity of the corresponding physiological monitoring data item. The sub-similarity of each physiological monitoring data item is calculated by weighted averaging using the preset weights of each physiological monitoring data item to obtain the real-time similarity.

[0042] It should be noted that sub-similarity = 1 - normalization (normalization distance), which maps the normalization distance to the interval [0, 1]. The smaller the normalization distance, the closer the sub-similarity is to 1, and the more similar the data sequences are.

[0043] Among them, the normalization process for the regular distance can be performed by minimum-maximum normalization.

[0044] In order to discretize continuous data into independently analyzable time series units, i.e. data sequences, and lay the foundation for subsequent segment-by-segment comparisons, as a possible implementation method, continuous, real-time physiological monitoring data, as well as physiological data from historical cases corresponding to the current time period, are cut into analysis windows of the same duration to form corresponding data sequences.

[0045] It should be noted that the specific value of the time period should be determined based on the actual situation, taking into account both the system response speed and the delay in physiological response. For example, from the intravenous infusion of drugs to the time it takes for the drugs to circulate in the blood and act on the target organs, triggering observable physiological effects (such as changes in blood pressure and heart rate), this process takes a certain amount of time. A time window much shorter than this process (such as 10 seconds) cannot capture meaningful physiological changes; while an excessively long window (such as 10 minutes) will lead to an overly sluggish system response. A duration of 2 minutes is a reasonable choice in clinical experience that can balance signal effectiveness and system responsiveness.

[0046] Physiological monitoring data items refer to different types of physical or chemical signal parameters that are continuously or frequently collected by medical monitoring equipment and can quantify the real-time functional state of a life system, such as heart rate. Physiological monitoring data includes at least one physiological monitoring data item.

[0047] It is important to understand that since each physiological monitoring data item is physically measurable and has different clinical interpretations, an independent and precise similarity analysis can be performed on each physiological monitoring data item. For example, the heart rate data sequence of the current time period and the heart rate data sequence of historical cases at a similar time period can be established as "two target data sequences" to be compared.

[0048] A historically similar time period refers to a phase in the input process of historical cases that is comparable to the current time period in terms of input progress. The determination of a historically similar time period is mainly based on the relative time of input start. For example, if the current object is in the 25th minute after input starts, then in the selected "reference historical case set", the system will prioritize the time period of 20-30 minutes after input starts for each historical case and preliminarily define it as a "historically similar" time period to the current time period.

[0049] It should be noted that dynamic time warping is a common technique, and will not be described in detail in this embodiment.

[0050] It should be noted that different physiological monitoring data items are assigned different preset weights. The weighting is based on the degree of correlation with the core treatment goal. The preset weights are determined according to the clinical importance of the corresponding physiological monitoring data item in cardiology infusion therapy decisions. This embodiment does not impose specific limitations. For example, in the "cardiology infusion" scenario addressed by this invention, since the direct treatment goal of most cardiology vasoactive drugs (such as vasopressors, antihypertensives, and cardiotonics) is to regulate blood pressure and heart rate, the high-weighted items are blood pressure and heart rate, which are the most direct and critical indicators for assessing efficacy and safety. The second highest weighted item is blood oxygen saturation, which reflects the comprehensive efficiency of circulation and respiration and is an important indicator for assessing tissue infusion. Finally, the basal / low-weighted item is body temperature. Body temperature is usually not the primary target of cardiology drug regulation, but rather serves as a warning indicator for basic vital signs and complications such as infection.

[0051] For example, suppose there are four physiological monitoring data items: blood pressure, heart rate, blood oxygen saturation, and body temperature. The preset weights for blood pressure, heart rate, blood oxygen saturation, and body temperature are 0.35, 0.35, 0.20, and 0.10, respectively. The sum of all preset weights is 1. Then, the real-time similarity is calculated as: sub-similarity of blood pressure × 0.35 + sub-similarity of heart rate × 0.35 + sub-similarity of blood oxygen saturation × 0.20 + sub-similarity of body temperature × 0.10.

[0052] It's important to understand that core real-time similarity calculation algorithms (such as Dynamic Time Warping (DTW)) require that both data sequences being compared must exist and be complete. However, forcibly requiring all data items to be complete is not feasible in current common medical scenarios. After all, different medical units (such as ICUs and general wards) may be equipped with different monitoring equipment. In order to adapt to different data environments and ensure that the system can always continue to work based on the "best available subset of data" under any imperfect data conditions, avoiding system paralysis due to single point of failure, a data availability check is performed before calculating real-time similarity. This involves comparing the list of physiological monitoring data items actually successfully collected by the current object in the current time period with the list of data items with complete records in similar historical time periods of the historical cases to be compared, thus obtaining a set of valid data items.

[0053] In this embodiment, physiological monitoring data items that are common to the current object and historical cases and are complete are extracted to form a set of valid data items; two target data sequences containing the same physiological monitoring data item are identified, but this is limited to physiological monitoring data items in the set of valid data items.

[0054] It should be noted that the physiological monitoring data items of the current object in the current time period can be taken as the intersection of the physiological monitoring data items in similar historical time periods. The physiological monitoring data items contained in this intersection are the set of valid data items.

[0055] It's important to understand that a historical case similar to the current subject may not be the best approach for the current subject. Generally, a plan that guides a smooth transition in physiological state (with minimal fluctuations) is a high-quality, low-risk plan and should receive a higher reference weight. Therefore, analyzing "fluctuation differences" can help assess the effectiveness and safety of historical plans. Furthermore, if a historically effective plan differs significantly from the currently implemented plan, switching directly may be risky. Therefore, analyzing "infusion parameter differences" can quantify the potential risks of adjusting the current plan based on historical reference plans. The smaller the infusion parameter differences, the smoother the adjustment, the lower the risk, and the higher the reference weight of the historical plan, thus avoiding drastic fluctuations in control.

[0056] The process of adjusting weights is as follows: Figure 2 As shown, it includes: S101-1, Obtain the preset safe value range of various physiological monitoring data items in the set of valid data items.

[0057] The preset safety value range refers to a numerical interval set for each physiological monitoring data item, used to quantify the degree of "normal", "abnormal", or "ideal" physiological state. The specific values ​​are determined based on authoritative medical guidelines and clinical expert experience; this embodiment does not impose specific limitations. For example, there is generally broad medical consensus on routine vital signs: for heart rate, the target range for adult resting heart rate is typically set at 60-100 beats / min; for systolic blood pressure, the target range may be strictly set depending on individual physiological conditions. For instance, for specific heart failure patients, the target range may be 90-110 mmHg to avoid excessive cardiac load.

[0058] S101-2, for various physiological monitoring data items in the valid data item set, a consistency score is determined based on the similarity between the physiological monitoring data items of historical cases in the next adjacent historical period and their preset safe value range.

[0059] The consistency score quantifies the overall quality of the physiological state of historical subjects within the evaluation period, i.e., the next adjacent historical period of historical cases. The higher the consistency score, the better the corresponding historical plan maintains physiological stability.

[0060] In this embodiment, all physiological monitoring data items in the next adjacent historical period are obtained in a time sequence, and the total number of data items of all physiological monitoring data items is counted; it is determined whether the value of each physiological monitoring data item is within the preset safe value range for physiological monitoring data items; the target number of all physiological monitoring data items within the preset safe value range is counted; the ratio of the target number to the total number of data items is calculated as a consistency score.

[0061] It should be noted that if the total number of data items is zero, it means that in the next adjacent historical period, all physiological monitoring devices used in the set of valid data items have malfunctioned and have not recorded any valid data points, or the data has been completely lost during storage or transmission. This is usually an extreme case, and such extreme historical cases will be excluded from the weight calculation. In normal clinical scenarios, medical monitoring devices will continuously collect data, so the total number of data items will not be zero.

[0062] The target number is the number of physiological monitoring data items that are in an ideal safety state after the corresponding infusion protocol has been used in historical cases.

[0063] It is important to understand that, since the target number cannot be negative or exceed the total number of data items, if a historical case has a consistency score of zero in the next adjacent historical period, it means that no physiological monitoring data item is within the preset safety range, indicating that the infusion protocol used in that historical case was extremely ineffective. Conversely, if a historical case has a consistency score of 1 in the next adjacent historical period, it means that all physiological monitoring data items are perfectly within the preset safety range, indicating that the infusion protocol used in that historical case was extremely effective.

[0064] S101-3. Based on the fluctuation differences between the physiological monitoring data of historical cases in similar historical periods and in the next adjacent historical period, and combined with the consistency score, determine the single quality indicator; using the preset first weight coefficient, the single quality of all physiological monitoring data items in the effective data item set is weighted and summed to obtain the historical validity score.

[0065] In this embodiment, the standard deviation of all physiological monitoring data items in the effective data item set within a similar historical time period is calculated as the first degree of fluctuation; the standard deviation of all physiological monitoring data items in the effective data item set within the next adjacent historical time period is calculated as the second degree of fluctuation; the absolute difference between the second degree of fluctuation and the first degree of fluctuation is normalized to obtain the fluctuation increase; the difference between the positive integer 1 and the fluctuation increase is calculated as the first difference; the product of the first difference and the consistency score is calculated as the single quality indicator.

[0066] The first level of fluctuation represents the original level of fluctuation in the physiological monitoring data items before the adoption of the historical protocol.

[0067] The second level of fluctuation represents the new level of fluctuation in physiological monitoring data items after adopting historical protocols.

[0068] It should be noted that the specific method for calculating the standard deviation is a well-known technique in the art, and will not be described in detail in this embodiment.

[0069] The increase in volatility quantifies the relative magnitude of the change in volatility of physiological monitoring data items from a similar historical period to the next adjacent historical period. The closer the increase in volatility is to zero, the less the volatility of physiological monitoring data items changes, the smoother the transition is, and the lower the risk of using the historical scheme to switch to the next adjacent historical period.

[0070] It should be noted that the absolute difference between the second fluctuation level and the first fluctuation level can be normalized using the minimum-maximum normalization method.

[0071] A single high-quality indicator quantifies the extent to which a particular historical infusion regimen improves the performance of a specific physiological monitoring data item (such as heart rate or blood pressure).

[0072] Since a higher consistency score for a historical case in the next adjacent historical period indicates a better effect of the historical plan in maintaining physiological stability, and a smaller increase in the degree of fluctuation of a historical case from a similar historical period to the next adjacent historical period indicates a smaller fluctuation in the physiological monitoring data and a smoother transition, the risk of switching the historical plan used for the next adjacent historical period is lower. Correspondingly, a larger first difference value indicates a better adjustment effect for the historical case. Therefore, a single high-quality indicator can be expressed by the following formula: in, This indicates the individual quality of each physiological monitoring data item in the set of valid data items; Represents the normalization function; Indicates the second degree of fluctuation; Indicates the first degree of fluctuation; L represents the consistency score.

[0073] The specific value of the preset first weight coefficient is determined based on authoritative treatment guidelines and rich clinical experience. This embodiment does not impose specific limitations. For example, the higher the weight of a physiological monitoring data item that has a greater impact on the prognosis of the current physiological condition of the subject, the higher the corresponding weight. Assuming there are three physiological monitoring data items: diastolic blood pressure, heart rate, and central venous pressure, for example, diastolic blood pressure is the core target of most vasoactive drugs in cardiology, so the weight of diastolic blood pressure is higher and can be set to 0.45; heart rate is a key indicator of cardiac function and myocardial oxygen consumption, and many drugs can significantly affect heart rate, so its weight is also relatively high and can be set to 0.4; central venous pressure reflects volume status, is important but is usually not used as the primary real-time control target, so its weight is relatively smaller and can be set to 0.15.

[0074] Historical validity score is used to evaluate the overall quality and reference value of a historical betting case. It reflects the entire betting scheme experienced by a historical object and whether, from an overall perspective, it is a successful, high-quality scheme that is worth referencing for the current object.

[0075] S101-4, Determine the parameter difference degree based on the difference between the real-time infusion parameters of the current time period and the infusion parameters of historical cases in similar historical time periods.

[0076] It should be noted that, as explained in the relevant description of the acquisition module 101, when calculating the "parameter difference", the difference between the "current dosing rate" and the "historical dosing rate of historical cases in similar time periods" is compared, in order to assess the risk of adjusting the treatment intensity and make the calculated parameter difference more reasonable in clinical sense.

[0077] In this embodiment, the real-time infusion parameters of the current time period are constructed as a first parameter vector, and the infusion parameters of historical cases in similar historical time periods are constructed as a second parameter vector, wherein each dimension of the vector represents the parameter value of an infusion channel; the distance between the first parameter vector and the second parameter vector is calculated as the parameter difference degree.

[0078] The first parameter vector is composed of the current dosing rate of all channels (e.g., channel 1, channel 2, ... channel N) in the current time period, where N represents the total number of channels in the infusion system.

[0079] The second parameter vector is composed of the historical dosing rates of all channels in similar historical time periods.

[0080] It should be noted that Manhattan distance or Euclidean distance can be used to calculate the distance between the first parameter vector and the second parameter vector. Using Manhattan distance or Euclidean distance to calculate distance is a common technique, and will not be elaborated on in this embodiment.

[0081] It is important to understand that large and rapid switching of infusion parameters in a system is usually accompanied by a higher risk of instability. Therefore, the parameter difference can be used to quantify the overall difference between the currently executed infusion parameters and the infusion parameters executed in reference historical cases. The larger the parameter difference value, the more drastic the adjustment of the infusion parameters, and the higher the possibility of introducing instability into the system.

[0082] S101-5, calculate the product of historical validity score and real-time similarity as the first intermediate value; calculate the sum of positive integer 1 and parameter difference as the second intermediate value; normalize the ratio of the first intermediate value to the second intermediate value to obtain the adjusted weight.

[0083] It's important to understand that a higher historical effectiveness score for a given historical solution indicates a higher quality solution. Furthermore, a higher real-time similarity between the current object and that historical case in terms of real-time physiological state suggests a greater relevance between the two solutions. This reflects the fact that higher historical effectiveness scores and higher real-time similarity are necessary to generate higher expected returns, meaning a larger first median value. Conversely, a greater difference between current and historical input parameters indicates a greater overall difference, representing a larger adjustment range and higher potential risk. Therefore, a smaller second median value suggests a higher likelihood of instability in the system due to the use of historical input parameters, requiring severe penalties for large adjustments.

[0084] The second determining module 104 is used to perform weighted fusion of infusion parameters for similar historical time periods by adjusting the weights, and to determine the target infusion parameters for the next adjacent time period of the current object.

[0085] It should be noted that, as explained in the relevant description of the acquisition module 101, in the step of "weighted fusion using adjusted weights", the object of fusion is the "historical dosing rate" of historical cases, and the result obtained after fusion is the "target dosing rate".

[0086] The adjustment weights are determined for each historical case. To ensure that the sum of all weights is 1, the sum of the adjustment weights of all historical cases in the reference historical case set can be calculated first as the total weight. Then, the ratio of the adjustment weight of each historical case to the total weight can be calculated as the normalized weight of each historical case. Finally, the product of the historical dosing rate and the corresponding normalized weight for each infusion channel with similar historical time periods can be calculated as the target dosing rate.

[0087] The verification module 105 is used to verify whether the target infusion parameters are within the preset safety boundary. If so, it controls the multi-channel infusion system to run according to the target infusion parameters in the next adjacent time period.

[0088] It is important to understand that, based on the explanation of the acquisition module 101, the system makes intelligent decisions based on the "drug administration rate". Similarly, the obtained target infusion parameter is actually a "target drug administration rate". Subsequently, after the safety boundary verification is passed, it is converted into a specific "target infusion flow rate" based on the current "preset drug concentration" and sent to the device. This is to decouple the high-level decision-making logic from the low-level device control throughout the process, so that the core algorithm is not affected by changes in the drug in the specific preparation scheme.

[0089] In this embodiment, the target infusion rate to be delivered is determined based on the target infusion parameters and the drug concentration preset in the infusion system; the multi-channel infusion system is controlled to operate at the target infusion rate.

[0090] It should be noted that the drug concentration is the original concentration of the drug solution in the infusion bag / bottle, which is a fixed value prepared by the pharmacist before the infusion begins. The unit is mass / volume, such as milligrams per milliliter (mg / mL). The infusion flow rate is the command directly received and executed by the infusion pump, which refers to the volume of drug solution pumped per unit time. The unit is volume / time, such as milliliters per hour (mL / h). The administration rate is the mass of the active ingredient of the drug entering the patient's body per unit time. The unit is mass / time, such as milligrams per hour (mg / h). Therefore, the administration rate = drug concentration × infusion flow rate. Thus, the target infusion flow rate = target administration rate ÷ drug concentration.

[0091] It should be noted that the drug media pre-set in different channels are not the same, and correspondingly, the concentration of the pre-set drug solution is also different.

[0092] Preset safety boundaries are upper and lower limits set for the drug delivery rate of each infusion channel during system initialization. They define the absolute safety range for personalized treatment. The preset safety boundaries configured for each channel are not the same. The specific value of the preset safety boundary is determined based on the drug medium infused in the channel and the individual information of the current subject. For example, assuming that the drug infused in channel 1 is nitroglycerin, for channel 1, the drug delivery rate of channel 1 is calculated by combining the maximum clinical safe dose of the drug in channel 1 (calculated by weight, unit such as μg / kg / min), the current subject's weight, age, and underlying diseases (the lower limit can be 0, and the upper limit = maximum safe dose of drug × current subject's weight). The target infusion parameter (drug delivery rate) must simultaneously meet the personalized safety boundaries of all channels. After verification, it is converted into the infusion flow rate based on the drug concentration (target infusion flow rate = target drug delivery rate ÷ drug concentration). Here, 1mg / h≈16.7mcg / min.

[0093] It should be noted that if the target infusion parameters are not within the preset safety boundaries, a tiered alarm will be triggered (minor exceedances will prompt for adjustment, while severe exceedances will prohibit execution), and the current infusion parameters will be retained pending manual intervention.

[0094] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0095] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A multi-channel intelligent drug infusion system for cardiology, characterized in that, The system includes: The acquisition module is used to acquire the current subject's clinical characteristic indicators, real-time physiological monitoring data, and real-time infusion parameters of each channel after the infusion process begins; and to acquire historical case data containing the clinical characteristic indicators, physiological monitoring data, and infusion parameters of historical subjects. The first determination module is used to determine the clinical similarity based on the clinical characteristic indicators of the current object and the clinical characteristic indicators of each historical object; and to determine historical cases with clinical similarity exceeding a first threshold as a set of reference historical cases. The weight adjustment module is used to determine the real-time similarity of each historical case in the reference historical case set based on the real-time physiological monitoring data in the current time period and the physiological monitoring data of historical cases in similar historical time periods; and to determine the adjustment weight for the infusion parameters of historical cases based on the fluctuation differences between the physiological monitoring data of historical cases in similar historical time periods and the next adjacent historical time period, the differences between the real-time infusion parameters of the current time period and the infusion parameters of historical cases in similar historical time periods, and the real-time similarity. The second determination module is used to perform weighted fusion of infusion parameters in similar historical time periods by adjusting the weights, and to determine the target infusion parameters for the next adjacent time period of the current object; The verification module is used to verify whether the target infusion parameters are within the preset safety boundary. If so, it controls the multi-channel infusion system to run according to the target infusion parameters in the next adjacent time period.

2. The multi-channel intelligent drug infusion system for cardiology as described in claim 1, characterized in that, The clinical similarity determination process includes: Numerical preprocessing is performed on the clinical characteristic indicators of the current and historical subjects to form their respective feature vectors; Calculate the distance between the feature vector of the current object and the feature vector of each historical object; Clinical similarity is determined based on distance, where distance is negatively correlated with clinical similarity.

3. The multi-channel intelligent drug infusion system for cardiology as described in claim 2, characterized in that, The clinical feature indicators include numerical indicators, textual indicators, and image / curve indicators; the numerical preprocessing of the clinical feature indicators of the current object and historical objects includes: For numerical indicators, normalization is performed; For text-based metrics, a pre-defined bag-of-words model is used to convert the text-based metrics into numerical vectors. For image and curve indicators, feature vectors of image and curve indicators are extracted through a preset neural network.

4. The multi-channel intelligent drug infusion system for cardiology as described in claim 1, characterized in that, The real-time similarity determination process includes: Real-time physiological monitoring data and historical physiological monitoring data are divided into corresponding data sequences based on the same time period. For various physiological monitoring data items, the data sequence of the current time period is matched with the data sequence of similar historical time periods to identify two target data sequences containing the same physiological monitoring data item; the dynamic time warping algorithm is used to calculate the warping distance between the two target data sequences, and the warping distance is mapped to the sub-similarity of the corresponding physiological monitoring data item; By using the preset weights of each physiological monitoring data item, the sub-similarity of each physiological monitoring data item is calculated by weighted average to obtain the real-time similarity.

5. A multi-channel intelligent drug delivery system for cardiology as described in claim 4, characterized in that, Before determining the two target data sequences, the method includes: Extract complete physiological monitoring data items that are common to the current object and historical cases to form a set of valid data items; The determination of two target data sequences containing the same physiological monitoring data item is limited to the determination of physiological monitoring data items in the set of valid data items.

6. A multi-channel intelligent drug infusion system for cardiology according to claim 5, characterized in that, The physiological monitoring data includes at least one physiological monitoring data item, and the weighting determination process includes: Obtain the preset safe value range for various physiological monitoring data items in the valid data item set; For various physiological monitoring data items in the valid data item set, a consistency score is determined based on the similarity between the physiological monitoring data items of historical cases in the next adjacent historical period and their preset safe value range; Based on the fluctuation differences between the physiological monitoring data of historical cases in similar historical periods and in the next adjacent historical period, and combined with the consistency score, individual high-quality indicators are determined; using the preset first weight coefficient, the individual quality of all physiological monitoring data items in the effective data item set is weighted and summed to obtain the historical validity score. The degree of parameter difference is determined based on the differences between the real-time infusion parameters of the current period and the infusion parameters of historical cases in similar periods in the past. The product of the historical validity score and the real-time similarity is calculated as the first intermediate value; the sum of the positive integer 1 and the parameter difference is calculated as the second intermediate value; the ratio of the first intermediate value to the second intermediate value is normalized to obtain the adjusted weight.

7. A multi-channel intelligent drug delivery system for cardiology as described in claim 6, characterized in that, The consistency score is determined based on the similarity between physiological monitoring data items from historical cases in the next adjacent historical period and their preset safety value range, including: Obtain all physiological monitoring data items in the next adjacent historical period in a time-series order, and count the total number of data items for all physiological monitoring data items; Determine whether the value of each physiological monitoring data item is within the preset safe value range for the physiological monitoring data item; Count the target number of all physiological monitoring data items that are within the preset safety value range; The ratio of the number of targets to the total number of data items is used as the consistency score.

8. A multi-channel intelligent drug delivery system for cardiology as described in claim 7, characterized in that, The method, based on the fluctuation differences in physiological monitoring data of historical cases within similar historical periods and in the next adjacent historical period, and combined with a consistency score, determines individual high-quality indicators, including: The standard deviation of all physiological monitoring data items in the valid data set within a similar historical time period is calculated as the first degree of fluctuation; the standard deviation of all physiological monitoring data items in the next adjacent historical time period is calculated as the second degree of fluctuation. The absolute difference between the second fluctuation level and the first fluctuation level is normalized to obtain the fluctuation level increase. Calculate the difference between the positive integer 1 and the increase in volatility, and use it as the first difference; calculate the product of the first difference and the consistency score, and use it as a single quality indicator.

9. A multi-channel intelligent drug infusion system for cardiology according to claim 6, characterized in that, The determination of parameter difference based on the differences between real-time infusion parameters in the current time period and infusion parameters in similar historical time periods includes: The real-time infusion parameters of the current time period are constructed as the first parameter vector, and the infusion parameters of historical cases in similar historical time periods are constructed as the second parameter vector. Each dimension in the vector represents the parameter value of an infusion channel. Calculate the distance between the first parameter vector and the second parameter vector, which is used as the parameter difference.

10. A multi-channel intelligent drug infusion system for cardiology according to claim 1, characterized in that, After verification and before controlling the operation of the multi-channel infusion system, the method further includes: Based on the target infusion parameters and the pre-set drug concentration in the infusion system, the target infusion flow rate to be issued is determined; Control the multi-channel infusion system to operate at the target infusion flow rate.

Images