A controlled drug injection monitoring system and armband

Abstract

This invention relates to the field of medical monitoring information processing technology, and in particular to a controlled drug injection monitoring system and armband. The system includes a controlled drug injection monitoring armband and a cloud processing module. The controlled drug injection monitoring armband includes a monitoring module, an information transmission module, and a drug injection module. The monitoring module collects the user's vital signs information. The cloud processing module constructs a deep learning model to determine the corresponding pain level. The drug injection module uses an extended unscented Kalman filter system to estimate the pain level based on a pharmacokinetic model and the pain level, establishes a control model for optimization, establishes a real-time drug injection plan that minimizes the cost, and dynamically administers the drug. This invention achieves automatic analysis of the user's pain status through multimodal analysis and dynamically adjusts the medication plan in real time based on the user's state during injection, realizing personalized analgesia optimization within safety boundaries.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring information processing technology, and in particular to a controlled drug injection monitoring system and an armband. Background Technology

[0002] Current controlled substance injection monitoring technologies suffer from several serious shortcomings, failing to meet the actual needs of end-stage patients for safe medication use. First, most medical institutions still rely on manual recording of injections, resulting in outdated information that is prone to omissions, errors, or alterations, lacking real-time accuracy. Second, existing systems lack automatic identification and real-time monitoring of injection activities, making it impossible to determine whether an injection actually occurred, whether the dosage was appropriate, or to identify the injector, posing risks of proxy injections, misuse, or abuse. Furthermore, most systems lack abnormal behavior warning mechanisms, failing to promptly detect frequent medication use, overdoses, or illegal use, increasing the risk of drug abuse.

[0003] Therefore, how to conduct closed-loop monitoring of controlled drug use in a real-time, accurate, and safe manner is an urgent problem to be solved. Summary of the Invention

[0004] To address one of the aforementioned problems in the prior art, the present invention provides a controlled drug injection monitoring system and an armband.

[0005] To achieve the above objectives, the present invention provides a controlled drug injection monitoring system, comprising: a controlled drug injection monitoring armband and a cloud processing module; the controlled drug injection monitoring armband comprises: a monitoring module, an information transmission module and a drug injection module;

[0006] The monitoring module is used to collect vital sign information of the user wearing the controlled drug injection monitoring armband. The vital sign information refers to multiple sets of data used to characterize the degree of pain in the user's body.

[0007] The information transmission module is used to send the vital signs information to the cloud processing module;

[0008] The cloud processing module is used to construct a deep learning model, pre-train the deep learning model using the vital sign information to obtain a trained deep learning model, and acquire the user's current vital sign information data sequence in real time. The trained deep learning model is used to determine the pain level corresponding to the user's current vital sign information data sequence, and the pain level is transmitted to the information transmission module. The current vital sign information data sequence includes the vital sign information collected at a preset period.

[0009] The information transmission module is also used to receive the pain level and transmit the pain level to the drug injection module;

[0010] The drug injection module is used to acquire the pain level, estimate the user's injection status in real time based on the pharmacokinetic model and the pain level using an extended unscented Kalman filter system, establish a control model using the user's injection status in real time for optimization to determine the minimum cost drug injection real-time plan, and dynamically inject drugs according to the minimum cost drug injection real-time plan. The user's injection status in real time includes: predicted blood drug concentration, predicted effect-site concentration, and predicted pain level.

[0011] In another aspect, the present invention provides a controlled drug injection monitoring armband, comprising: an armband body, a monitoring module, an analysis module, an information transmission module, and a drug injection module;

[0012] The monitoring module is used to collect vital sign information of the user wearing the controlled drug injection monitoring armband through the armband body. The vital sign information refers to multiple sets of data used to characterize the degree of pain in the user's body.

[0013] The analysis module is used to construct a deep learning model, pre-train the deep learning model using the vital sign information to obtain a trained deep learning model, and acquire the user's current vital sign information data sequence in real time. The trained deep learning model is used to determine the pain level corresponding to the user's current vital sign information data sequence, and the pain level is transmitted to the information transmission module. The current vital sign information data sequence includes the vital sign information collected at a preset period.

[0014] The information transmission module is also used to receive the pain level, upload the pain level to the background processing center, and receive the drug injection authorization information returned by the background processing center, and unlock the use permission of the drug injection module according to the drug injection authorization information.

[0015] The drug injection module is used to acquire the pain level, estimate the user's injection status in real time based on the pharmacokinetic model and the pain level using an extended unscented Kalman filter system, establish a control model using the user's injection status in real time for optimization to determine the minimum cost drug injection real-time plan, and dynamically inject drugs according to the minimum cost drug injection real-time plan. The user's injection status in real time includes: predicted blood drug concentration, predicted effect-site concentration, and predicted pain level.

[0016] The beneficial effects of this invention are reflected in the following: it proposes a controlled drug injection monitoring system and armband, which analyzes multimodal vital sign data and achieves automatic analysis of the user's pain status through multimodal analysis methods. It enables precise intervention for users wearing the armband, timely improvement of the user's vital status, and precise analysis of the user's pain status, thereby accurately setting the required medication plan. It also dynamically adjusts the medication plan in real time according to the user's status during the injection, thereby standardizing the use of controlled drugs, preventing drug "misuse" or "abuse", preventing mechanical injury and acute overdose injection, avoiding fatal risks such as respiratory depression, and ultimately achieving personalized analgesia optimization within the safety boundary. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the controlled drug injection monitoring arm ring structure provided in Embodiment 1 of the present invention;

[0018] Figure 2 This is an example of the structural design of a controlled drug injection monitoring armband provided in Embodiment 1 of the present invention;

[0019] Figure 3 This is a schematic diagram of the controlled drug injection monitoring armband system provided in Embodiment 2 of the present invention.

[0020] Reference numerals: 11-Armband body; 12-Monitoring module; 13-Analysis module; 14-Information transmission module; 15-Drug injection module; 151-Replaceable drug stick; 152-Indwelling needle interface; 16-Button; 31-Controlled drug injection monitoring armband; 32-Cloud processing module. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1

[0023] This embodiment provides a controlled drug injection monitoring armband, the structural schematic of which is shown below. Figure 1As shown, the arm ring includes: a main body 11, a monitoring module 12, an analysis module 13, an information transmission module 14, and a drug injection module 15. The main body 11 can be made into a rollable shape to wrap around a human arm or the limbs of other animals, and serves as a carrier for the other modules. The monitoring module 12, the analysis module 13, and the information transmission module 14 can be implemented using electronic components such as sensors, chips, and circuits installed within the arm ring. These electronic components can be located on the surface or inside the structure of the main body 11.

[0024] In one alternative implementation, each armband is uniquely bound to a user ID, thus avoiding the risk of misuse.

[0025] The armband body 11 is designed to make wearable contact with the user; specifically, the armband body 11 may be made of flexible wearable material and may include components for connecting to the human body (such as wristbands, watch straps, etc.) and components for carrying other modules (such as watch faces, watch bodies, displays, etc.).

[0026] like Figure 2 The figure shows a possible structural design implementation of the armband in this embodiment. The armband body 11 mainly includes a wristband and a watch body with a display screen. The watch body is equipped with electronic components (not shown in the figure) that carry and realize the monitoring module 12, the analysis module 13 and the information transmission module 14. The drug injection module 15 is connected to the watch body and is controlled by other modules.

[0027] The monitoring module 12 is used to collect vital sign information of the user wearing the controlled drug injection monitoring armband via the armband body 11. Vital sign information refers to multiple sets of data used to characterize the user's pain level; specifically, when a terminally ill cancer patient experiences pain, the patient may exhibit elevated blood pressure, rapid heartbeat, and trembling. Therefore, the monitoring module 12 needs to monitor the user's vital signs to determine the patient's pain status. Vital sign information can include at least heart rate, blood oxygen saturation, body temperature, blood pressure, and movement status—characteristics reflecting the user's vital state. The monitoring module 12 may include detection units such as a heart rate monitoring unit, a blood oxygen saturation monitoring unit, a body temperature monitoring unit, a blood pressure estimation unit, and a movement status detection unit, each used to monitor and collect various vital signs of the user. For example, the heart rate monitoring unit can use a PPG (Photoplethysmography) sensor to acquire the average heart rate data sequence within a preset period (e.g., a period of 1 minute). ,in This represents the average heart rate data for the first cycle. N The integer represents the number of data cycles to be retrieved; the blood oxygen monitoring unit can use dual-wavelength PPG to monitor and obtain a blood oxygen data sequence. ,in This represents the average blood oxygen saturation data for the first cycle; the body temperature monitoring unit can use a negative temperature coefficient thermistor (NTC thermistor) to obtain a body temperature data sequence. ,in This represents the average body temperature data for the first cycle; the blood pressure monitoring unit still uses PPG to acquire it. Since PPG cannot directly obtain the user's blood pressure data, the raw PPG wave signal can be directly acquired for further processing; the motion detection unit can determine the user's motion frequency data sequence using a triaxial accelerometer. ,in This represents the average exercise frequency during the first cycle. Except for blood pressure data, the cycles for all other data are the same as those for heart rate data.

[0028] In a specific implementation, the method for obtaining blood pressure data from the raw PPG waveform signal is as follows: First, parameters including rise time, fall time, waveform area, and dicrotic wave characteristics are extracted from the raw PPG waveform; then, support vector regression is used to provide a predicted blood pressure data sequence based on the aforementioned parameters. The period it uses can be the same as the heart rate data (e.g., 1 minute), where This represents the average blood pressure data for the first cycle.

[0029] The analysis module 13 is used to construct a deep learning model. It pre-trains the deep learning model using vital sign information to obtain a trained model. It also acquires the user's current vital sign information data sequence in real time. Using the trained deep learning model, it determines the pain level corresponding to the user's current vital sign information data sequence and transmits the pain level to the information transmission module 14. The current vital sign information data sequence includes vital sign information collected at preset intervals. Specifically, the analysis module 13 is used to analyze the user's pain tolerance and can be embedded within the armband body. Before performing pain analysis, the analysis module 13 needs to pre-train the deep learning model to improve the accuracy of pain level calculation based on vital sign information.

[0030] In an optional implementation, the analysis module 13 uses a trained deep learning model to determine the pain level corresponding to the user's current vital signs data sequence based on the current vital signs data sequence. Specifically, this includes: aligning the current vital signs data sequence by grouping it according to the time dimension and constructing a first multimodal vector; feeding the first multimodal vector into the trained deep learning model for computation to obtain first calculated data; and comparing the first calculated data with preset reference data to determine the pain level corresponding to the current vital signs data sequence. Specifically, the current vital signs information can characterize the user's physical state, and the current vital signs information can contain the aforementioned feature data sequence, such as... , , , , Data sequences, which may contain N One data point, N The value can be determined as needed. For example, if the data sequence acquisition period is 1 minute, the value can be set to... N A value of 10 can reflect the user's physical condition over the past 10 minutes. N A value of 30 reflects the user's physical condition over the past 30 minutes. The first calculation data is obtained by processing the current vital sign information sequence through a trained deep learning model. The preset reference data is based on relevant medical definitions of pain levels, which can be implemented using a stepped threshold. By comparing the first calculation data with the values ​​of each step threshold, the closest pain level is determined as the final result. Euclidean distance can be used for this comparison.

[0031] In one optional implementation, the deep learning model includes a Transformer encoder (a deep learning architecture) and a Multi-layer Perceptron (MLP). Feeding the first multimodal vector into the trained deep learning model for computation includes: using the Transformer encoder with a multi-head attention mechanism to compute the first multimodal vector, performing residual connections and normalization calculations to obtain the final layer output; and then using the MLP to reconstruct the final layer output. Specifically, the Transformer encoder can have multiple layers, and residual connections and normalization calculations are performed on the vector data at each layer. The final layer output refers to the computation result output after the first multimodal vector data has passed through the last layer of the Transformer encoder. Residual connections make deep networks easier to train; normalization makes the data distribution more stable, accelerating training. This embodiment uses the multi-head attention mechanism of the Transformer encoder, allowing the model to simultaneously focus on information from different locations, while the MLP can integrate and transform this information, improving the model's generalization ability. By combining this with a reasonable parameter design of the MLP, computational overhead can be reduced while maintaining performance.

[0032] In an optional implementation, the analysis module 13 trains the deep learning model in the following manner: The analysis module 13 receives a sequence of vital sign information samples collected by the monitoring module 12 from the user's vital sign information; the vital sign information sample sequence is grouped and aligned according to the time dimension, and a second multimodal vector is constructed; the second multimodal vector is processed using a deep learning model to obtain a second reconstructed vector; a third and a fourth reconstructed vector are obtained, where the third reconstructed vector is obtained by processing the third multimodal vector using a deep learning model, and the fourth reconstructed vector is obtained by processing the fourth multimodal vector using a deep learning model. The third multimodal vector is constructed based on the user's vital sign information in a previous non-painful state, and the fourth multimodal vector is constructed based on the user's vital sign information in a previous painful state; a loss function is constructed based on the second, third, and fourth reconstructed vectors; the parameters of the deep learning model are updated through iterative training until the loss function converges, resulting in a trained deep learning model. Specifically, the vital sign information sample sequence may contain the user's real vital sign information over a certain period of time. By incorporating vital sign information from both the user's previous non-painful state and the user's previous painful state into the training of the deep learning model, better model parameters can be provided for subsequent specific task calculations, thereby improving model performance, reducing data requirements, accelerating training speed, and enhancing generalization ability.

[0033] In one specific implementation, the analysis module 13 can complete the pre-training of the deep learning model and determine the user's current pain level based on the current vital sign information data sequence through the following specific steps:

[0034] (1) Obtain the vital signs information of users in a certain time period as training samples, including: average heart rate data sequence Blood oxygen data sequence Body temperature data sequence Motion frequency data sequence and blood pressure prediction data series .

[0035] (2) The training samples , , , , Aligned according to the time dimension, and constructed into a multimodal vector. ,in Represents the real number field; Indicates the length of the time window ( T This can indicate how long the data period is for analysis; for example, when the period of the data sequence is 1 minute, T=N ); The modality number refers to the number of types of vital sign information. In this embodiment, the modality number is 5.

[0036] (3) Sent by The Transformer encoder, composed of layers, calculates the output of each layer as follows:

[0037] ;

[0038] ;

[0039] in, Represents the first... of the Transformer encoder layer, This indicates normalized calculation. Indicates the first l The intermediate quantities of the layer after residual connection and normalization Indicates the first l The output of the layer, Indicates the first The output of the layer, This indicates the use of a multi-head attention mechanism for processing. This indicates processing is performed using a feedforward neural network. During computation, a multi-head attention mechanism is first applied... ) Applied to the previous layer The output is then subjected to residual connection and layer normalization to produce... .Then, Through feedforward neural network ( ), and through another residual connection and normalization process, the first is obtained. The final representation of the layer The output of the last layer of the Transformer (represented as...) The data is then reconstructed again using a multilayer perceptron to obtain... .

[0040] (4) Use the user's vital signs information when there was no previous pain. Similarly, after the calculation in step (3), we obtain Specifically, the user's vital signs data under pain-free conditions are used to form a data sequence using the same sequence format described above. The data for the user in a pain-free situation can be historical vital sign data manually collected from the user, or historical data collected by the armband of the present invention before that, as long as the statistical caliber of the data is the same.

[0041] (5) Obtain the user's vital signs information under previous illness conditions. Similarly, after the calculation in step (3), we obtain Similarly, the user's vital signs data under previous medical conditions are used to form a data sequence using the same sequence format described above. The data on the user's past medical conditions can be historical vital sign data collected manually from the user, or historical data collected by the armband of this invention before that, as long as the statistical standards of the data are the same.

[0042] (6) Construct a variable loss function based on the results of steps (3), (4) and (5). :

[0043] ;

[0044] in, and These are binary tags (one is 1, and the other is 0). and These represent the labels for different inputs. During training, the input data is... , and One value is 0 and the other is 1. Data with a value of 0 is set to a random value within the possible range of human vital signs, while data with a value of 1 is retained as the actual data. Specifically, when... A value of (1,0) indicates that vital sign data under pain-free conditions are retained. ;when A value of (0,1) indicates that vital sign data under illness conditions are retained. Through binary tags and The use of 0 and 1 is mainly to distinguish whether the historical data is in a painful state or a painless state, ensuring that the historical data of the input data during training can only be in a painful state or a painless state.

[0045] (7) Iteratively train and update the parameters of the Transformer encoder and the multilayer perceptron until... convergence.

[0046] The above completes the pre-training of the model. When actually using the trained model to calculate the user's current pain level, the following steps are also included:

[0047] (8) Obtain the current vital signs information data sequence, calculate the current vital signs information data sequence through the trained model to obtain the calculation result, evaluate the Euclidean distance between the calculation result and the reference data to determine the pain level (in the form of a stepped threshold).

[0048] (9) Transmit the analysis results, including the degree of pain, to the information transmission module 14.

[0049] The information transmission module 14 receives the pain level, uploads it to the backend processing center, and receives the medication injection authorization information returned by the backend processing center. Based on the medication injection authorization information, it unlocks the access permission for the medication injection module 15. Specifically, the backend processing center can be a medical institution processing center with medical qualifications. The information transmission module 14 uploads the feedback from the analysis module 13 to the medical institution processing center for analysis and processing. Optionally, the medical institution processing center can determine whether to grant the user injection permission based on the pain analysis results. Users with injection permission can press the injection button to inject controlled medication to relieve pain. In a home setting, it can also be set so that once pain reaches a certain level, the device can directly grant injection permission without backend processing.

[0050] The drug injection module 15 is used to obtain the pain level. It estimates the user's injection status in real time using an extended unscented Kalman filter (UKF) based on the pharmacokinetic model and pain level. A control model is then built using this real-time prediction information to optimize and establish a cost-minimizing real-time drug injection plan. Drug injection is then dynamically performed according to this plan. The real-time prediction information includes the predicted blood drug concentration, the predicted effect-site concentration, and the predicted pain level. Specifically, the extended unscented Kalman filter (UKF) is a nonlinear filtering estimation algorithm with high computational accuracy. The pharmacokinetic model primarily focuses on blood drug concentration and effect-site concentration. Blood drug concentration is related to the dosage and metabolic capacity; excessively high blood drug concentration can lead to drug poisoning or other side effects, while excessively low concentrations can reduce the effect-site concentration. The effect-site is the theoretical compartment where the drug exerts its effect, and its concentration is related to the blood drug concentration through the blood-effect-site transport constant. During drug injection, to balance safety and effectiveness, it is necessary to ensure drug safety while effectively relieving patient pain. Therefore, the drug injection module 15 needs to first determine the safe drug injection plan based on the patient's pain condition, and estimate the patient's status information (blood drug concentration, effect-site concentration, and pain level, etc.) in real time, and adjust the medication according to the patient's real-time status.

[0051] In one alternative implementation: the predicted blood drug concentration is determined based on the drug clearance rate constant and the drug distribution volume; the predicted effect-site concentration is determined based on the predicted blood drug concentration and the blood-effect-site transport constant; the predicted pain level is determined at least based on the baseline pain level and the maximum analgesic effect the drug can produce. In a specific example, this can be implemented by first constructing the state variables using an extended unscented Kalman filter:

[0052] ;

[0053] in, Indicates the system in time t The state vector; Indicates time t The pain level (pain level can be expressed as a score of 1, 2...10, etc.); This indicates the blood drug concentration predicted by the filter. , by the clearance rate constant With the distribution volume of the drug drive; This represents the predicted effect-site concentration, expressed by the blood-effect-site transport constant. Connected to Predictive correction equations were established using a pharmacokinetic model, and a blood drug concentration dynamic equation was used to describe the accumulation and clearance dynamics of the drug in the blood, as shown below:

[0054] ;

[0055] in, Indicates the rate of change of blood drug concentration over time. , Represents the drug clearance rate constant , Indicates drug infusion rate , This represents the volume of distribution (L) of the drug. The effect-site concentration dynamic equation is shown below. The effect-site is the theoretical compartment in which the drug takes effect (such as the central nervous system). This equation quantifies the lag in drug efficacy.

[0056] ;

[0057] in, Indicates the rate of change of concentration in the effect room over time. , This represents the blood-effect-compartment transport constant, reflecting the rate at which a drug enters its target site. This represents the concentration gradient between the blood and the effect site. The pain rating-pharmacological relationship equation (also known as the Hill equation) is shown below, used to describe the process of converting blood drug concentration into analgesic effect (VAS decrease value);

[0058] ;

[0059] in, This indicates the pain level predicted by the model. This indicates the baseline pain level measured when no medication was used. This indicates the maximum analgesic effect of the drug (usually expressed as a reduction in pain score). Indicates half-effect concentration That is, the effect-site concentration at which 50% of the maximum analgesic effect is achieved. This indicates the steepness of the Hill curve (indicated by a pharmacology database or entered by a doctor).

[0060] After establishing the prediction and correction equations, the residuals can also be observed in real time. Used for updating Considering individual patient differences, it is expected that the patient will be able to complete the procedure within 2-3 hours. Error converges to The residuals reflect the difference between the actual pain level and the model's predicted register, and are used to correct for errors in UKF state estimation.

[0061] In an optional implementation, a control model is established using real-time prediction information of the user's injection status to optimize and establish a cost-minimizing real-time drug injection plan. Specifically, this includes: obtaining drug injection constraint expressions, which include a maximum infusion rate constraint and a 24-hour maximum infusion dose constraint; establishing an optimization function by performing a weighted summation operation based on the predicted blood drug concentration, predicted pain level, and drug injection constraint expressions as cost terms; and establishing a cost-minimizing real-time drug injection plan with the optimization objective of minimizing the weighted sum of the three cost terms. Specifically, excessively high blood drug concentrations may pose a risk of toxicity to the patient, a reduced pain level indicates analgesic effect, and the drug injection constraint expressions, including the maximum infusion rate constraint and the 24-hour maximum infusion dose constraint, can prevent mechanical injury and acute drug overdose. Using these three as cost terms, the control module can calculate the optimal infusion rate sequence and dynamically adjust the infusion rate to achieve the target analgesic effect under safety constraints.

[0062] The following is a specific example. After estimating the real-time prediction information of the user's injection status using the pharmacokinetic model, the information is fed back to the control model for iterative optimization. Specifically, the optimization can be performed at k time steps after time t. The optimization function is as follows:

[0063] ;

[0064] The optimization objective of the above function is to minimize the weighted sum of the three costs. Wherein, The pain control target set for doctors (usually a pain score of ≤2). This indicates the predicted pain level at time t+k; This represents the squared term predicting the blood drug concentration at time t+k. This represents the weighting coefficient for balancing analgesia and toxicity. This represents the weighting coefficient for the soft penalty over-threshold. Let represent the drug injection rate at time t+k, subject to the following constraints:

[0065] ;

[0066] ;

[0067] in, This refers to the peak flow rate of the syringe pump. This is the upper limit of the cumulative dose over 24 hours. By using constraints, the infusion rate at each moment can be controlled, and the upper limit of the cumulative dose over 24 hours can also be controlled (to prevent chronic poisoning), thus achieving dual safety protection.

[0068] The objective function described above contains three costs. Using the squared deviation between the predicted VAS score and the target VAS score as the cost can ensure analgesic efficacy. Multiplying the square of the blood drug concentration by a weighting factor can penalize high blood drug concentrations (because high concentrations may lead to poisoning), and minimizing this term can reduce the risk of poisoning. Penalize the portion exceeding the maximum permissible infusion rate. When, the value of this item is 0; when When the excess portion is multiplied by the weighting factor, This is to prevent excessively fast infusion rates, protect equipment safety, and reduce the risk of side effects.

[0069] Furthermore, if the armband is offline, it may not be able to obtain the latest user status information. To monitor the injection of the drug injection module 15, reinforcement learning methods can be used during the offline phase. These methods utilize state-action-reward samples generated from historical data, eliminating the need for actual interaction and thus enhancing safety. Therefore, in an optional implementation, when the drug injection module 15 is offline, it is also used to establish a negative reward function based on the deviation of the pain level from the pain control target, whether the historical drug infusion volume exceeds the maximum threshold, and whether respiratory side effects occur. The drug injection protocol is then adjusted according to the negative reward function. Specifically, during the offline phase, historical trajectories and control modules drive the simulation to generate... Samples were collected to observe key information such as analgesic effect, real-time drug concentration, and safety margin. s For the set of observation vectors, a This refers to the set of actions (i.e., the drug infusion rate). r For the reward set. s The observation vectors are as follows:

[0070] ;

[0071] in, Representing action history, that is, recording the current moment t Previous L The historical drug infusion rates are arranged in reverse chronological order into a vector; This indicates the dosage to be used within 24 hours. s The observation vector comprehensively covers pain feedback, movement history, drug metabolism status, and safe drug dosage. A reward function is constructed. r for:

[0072] ;

[0073] in, , and Indicates the weight of each item; The trigger condition for this item is that the pain score deviates from the target value, indicating insufficient analgesia. >2) or excessive sedation ( All deviations from 0 are penalized, with the greater the deviation, the greater the penalty. This item is an overdose indicator, representing the risk of acute poisoning. The trigger condition is or Record 1 if the threshold is exceeded, otherwise record 0; This item is a respiratory side effect indicator, representing respiratory depression side effects. Blood oxygen saturation ( ) triggers respiratory side effects such as decreased breathing or sleep apnea, when respiratory side effects are detected (e.g. When the number of sleep apnea events decreases or an apnea event occurs. It is 1 if it is true, otherwise it is 0. The reward function is negative, which means that it is regarded as a cost (minimizing the cost). When the reward function is maximized, the cost is minimized.

[0074] The above learning system uses state variables s Full observation, covering all safety-critical variables (pain, drug concentration, total dose), through action a Achieving dual constraints of physical rate limit and cumulative dose, and through rewards r By imposing extreme penalties on safety incidents (such as drug overdose or respiratory depression), this design ensures that the reinforcement learning strategy prioritizes patient safety above all else while pursuing analgesia, thus complying with medical ethics requirements. Therefore, offline training further avoids high-risk actions through conservative strategy updates, ultimately achieving personalized analgesia optimization within safety boundaries.

[0075] Furthermore, the drug injection module 15 can be implemented through a combination of software and hardware, as in one optional implementation, such as... Figure 2 The drug injection module 15 shown may include: a replaceable drug stick 151 for storing the medication to be injected by the user; and an indwelling needle interface 152 for injecting the medication into the user through the indwelling needle. Specifically, when a drug injection is required, the drug injection module 15 injects the medication into the user's body through the indwelling needle via a built-in piezoelectric pump, and tags the vital signs data after the injection for monitoring by medical staff (a copy of the data is tagged for a period of time after the injection for monitoring by medical staff).

[0076] This embodiment proposes a controlled drug injection monitoring armband, which analyzes multimodal vital sign data and automatically analyzes the user's pain status through multimodal analysis methods. It enables precise intervention for users wearing the armband, timely improvement of their vital signs, and accurate analysis of their pain status, thereby accurately setting the required medication regimen. The medication regimen is dynamically adjusted in real time according to the user's status during the injection, thus standardizing the use of controlled drugs, preventing drug "misuse" or "abuse," preventing mechanical injury and acute overdose injection, avoiding fatal risks such as respiratory depression, and ultimately achieving personalized analgesia optimization within safety boundaries.

[0077] In an optional implementation, the controlled drug injection monitoring armband of this embodiment further includes: a button 16; the button 16 is used to receive injection instructions; the drug injection module 15 is also used to complete drug injection according to the injection instructions. Specifically, the controlled drug injection monitoring armband of this embodiment can receive injection instructions through the button 16, or automatically generate injection instructions according to the usage permissions of the drug injection module 15 to automatically inject drugs. By setting the button 16, users can easily express their injection intentions, ensuring the user's true intentions are expressed, and facilitating user operation. By automatically generating injection instructions through the drug injection module 15, injections can be automatically performed on users when they cannot express their intentions. The combination of the two ensures both the user's true intentions are expressed and helps users who cannot express their intentions to complete automatic injections.

[0078] In an optional implementation, after the drug injection is completed, the information transmission module 14 also uploads the injection information to the back-end processing center. Specifically, the information transmission module 14 feeds back the actual operation information of the drug injection module 15 (such as injection time, dosage, and post-injection vital signs data) to the back-end processing center so that the back-end processing center can promptly obtain the user's medical status and prevent duplicate injections. Simultaneously uploading vital signs data and injection records to the back-end processing center enhances the safety of drug use.

[0079] Example 2

[0080] Unlike Embodiment 1, in this embodiment, to save computing power and reduce the size of the controlled substance injection monitoring armband in Embodiment 1, the analysis data and processing procedures completed by the analysis module 13 in Embodiment 1 are processed by the cloud processing module. After acquiring vital sign information, the controlled substance injection monitoring armband sends it to the cloud processing module and receives the pain level determined by the cloud processing module. The content already explained in the controlled substance injection monitoring armband of Embodiment 1 will not be repeated here; only a brief introduction to the system architecture is given. Figure 3As shown, the controlled drug injection monitoring system of this embodiment includes: a controlled drug injection monitoring armband 31 and a cloud processing module 32; the controlled drug injection monitoring armband includes: a monitoring module, an information transmission module and a drug injection module.

[0081] The monitoring module collects vital sign information from users wearing a controlled substance injection monitoring armband. Vital sign information refers to multiple sets of data characterizing the user's pain level; specifically, when terminally ill cancer patients experience pain, they may exhibit elevated blood pressure, rapid heartbeat, and trembling. Therefore, the monitoring module needs to monitor the user's vital signs to assess the patient's pain level. Vital sign information can include at least heart rate, blood oxygen saturation, body temperature, blood pressure, and activity level—characteristics reflecting the user's vital state. The monitoring module may include detection units such as a heart rate monitoring unit, blood oxygen saturation monitoring unit, body temperature monitoring unit, blood pressure estimation unit, and activity level detection unit, each used to monitor and collect various vital signs of the user.

[0082] The information transmission module is used to send vital sign information to the cloud processing module 32. Specifically, in this embodiment, the information transmission module can be used to send and receive data with the cloud processing module 32, and the data can be sent and received using the wired or wireless data interface of the armband.

[0083] The cloud processing module 32 is used to build a deep learning model, pre-train the deep learning model using vital sign information, obtain a trained deep learning model, and acquire the user's current vital sign information data sequence in real time. Using the trained deep learning model, it determines the pain level corresponding to the user's current vital sign information data sequence and transmits the pain level to the information transmission module. The current vital sign information data sequence includes vital sign information collected at preset intervals. Specifically, the cloud processing module 32 is used to analyze the user's pain tolerance. Before performing pain analysis, the cloud processing module 32 also needs to pre-train the deep learning model to improve the accuracy of calculating the pain level based on vital sign information.

[0084] The information transmission module is also used to receive the pain level and transmit the pain level to the drug injection module. Specifically, in this embodiment, the information transmission module receives data from the cloud processing module 32 through an external transceiver interface and transmits the data to the drug injection module through an internal transmission path.

[0085] In one optional implementation, after receiving the pain level but before transmitting it to the drug injection module, the information transmission module also uploads the pain level to a backend processing center and receives drug injection authorization information returned by the backend processing center, unlocking access to the drug injection module based on this authorization information. Specifically, the backend processing center can be a medical institution processing center with medical qualifications. The information transmission module uploads the pain level to this center for analysis and processing. Optionally, the medical institution processing center can determine whether to grant injection permission to the user based on the pain analysis results. Users granted injection permission can then press the injection button to administer controlled medication to relieve pain. In a home setting, the device can be configured to directly grant injection permission upon detecting pain at a certain level, without requiring backend processing.

[0086] The drug injection module acquires pain levels and estimates the user's injection status in real time using an extended unscented Kalman filter system based on a pharmacokinetic model and pain levels. A control model is then built using this real-time user injection status prediction information to optimize and establish a cost-minimizing real-time drug injection plan. Drug injection is then dynamically performed according to this plan. The real-time user injection status prediction information includes predicted blood drug concentration, predicted effect-site concentration, and predicted pain level. During drug injection, to balance safety and effectiveness, it is necessary to ensure drug safety while effectively relieving patient pain. Therefore, the drug injection module first determines a safe injection plan based on the patient's pain level and then estimates the patient's status information (blood drug concentration, effect-site concentration, and pain level, etc.) in real time, adjusting the medication dosage according to the patient's real-time status.

[0087] In an optional implementation, the analysis function of the drug injection module can also be performed by the cloud processing module 32, which performs the corresponding calculation function, while the drug injection module only completes the physical process of drug injection according to the injection plan, so as to further save the computing power of the controlled drug injection monitoring arm ring 31 in this embodiment and reduce the size of the arm ring.

[0088] This embodiment proposes a controlled substance injection monitoring system. Based on multimodal vital sign data analysis, it automatically analyzes the user's pain status using multimodal analysis methods, enabling precise intervention for users wearing armbands, timely improvement of their vital signs, and accurate analysis of their pain levels. This allows for precise setting of the required medication regimen and real-time dynamic adjustment of the regimen based on the user's condition during injection. This standardizes the use of controlled substances, prevents misuse or abuse, avoids mechanical injury and acute overdose, and mitigates fatal risks such as respiratory depression, ultimately achieving personalized analgesia optimization within safety boundaries. Furthermore, this controlled substance injection monitoring system utilizes distributed computing via the armband and cloud, which improves computational efficiency, reduces armband costs, decreases armband size, and extends armband lifespan.

[0089] In one optional implementation, the cloud processing module 32 uses a trained deep learning model to determine the pain level corresponding to the user's current vital signs data sequence. Specifically, this includes: aligning the current vital signs data sequence by grouping it according to the time dimension and constructing a first multimodal vector; feeding the first multimodal vector into the trained deep learning model for computation to obtain first calculated data; and comparing the first calculated data with preset reference data to determine the pain level corresponding to the current vital signs data sequence. Specifically, the first calculated data is the result obtained by processing the current vital signs data sequence through the trained deep learning model. The preset reference data is based on relevant medically defined pain level reference data, which can be implemented using a stepped threshold. By comparing the first calculated data with the values ​​of each step of the stepped threshold, the closest pain level is determined as the final result.

[0090] In one optional implementation, the deep learning model includes a Transformer encoder and a multilayer perceptron (MLP). Feeding the first multimodal vector into the trained deep learning model for computation includes: using the Transformer encoder with a multi-head attention mechanism to compute the first multimodal vector, performing residual connections and normalization calculations to obtain the final layer output; and then using the MLP to reconstruct the final layer output. Specifically, the Transformer encoder can have multiple layers, and residual connections and normalization calculations are performed on the vector data at each layer. The final layer output refers to the computation result output after the first multimodal vector data has passed through the last layer of the Transformer encoder. Residual connections make deep networks easier to train; normalization makes the data distribution more stable, accelerating training. This embodiment uses the multi-head attention mechanism of the Transformer encoder, allowing the model to simultaneously focus on information from different locations, while the MLP can integrate and transform this information, improving the model's generalization ability. By combining this with a reasonable parameter design of the MLP, computational overhead can be reduced while maintaining performance.

[0091] In an optional implementation, the cloud processing module 32 pre-trains the deep learning model using vital sign information to obtain a trained deep learning model. Specifically, the cloud processing module 32 receives a sequence of vital sign information samples collected by the monitoring module from the user's vital sign information via the information transmission module; aligns the vital sign information sample sequences by grouping them according to the time dimension and constructs a second multimodal vector; operates the second multimodal vector using the deep learning model to obtain a second reconstructed vector; obtains a third and a fourth reconstructed vector, where the third reconstructed vector is obtained by operating the third multimodal vector using the deep learning model, and the fourth reconstructed vector is obtained by operating the fourth multimodal vector using the deep learning model. The third multimodal vector is constructed based on the user's vital sign information in a previous non-painful state, and the fourth multimodal vector is constructed based on the user's vital sign information in a previous painful state; constructs a loss function based on the second, third, and fourth reconstructed vectors; and updates the parameters of the deep learning model through iterative training until the loss function converges, thus obtaining the trained deep learning model. Specifically, the vital sign information sample sequence can contain the user's real vital sign information over a certain period of time. By introducing the user's vital sign information in a previous non-painful state and the user's vital sign information in a previous painful state to train the deep learning model, better model parameters can be provided for subsequent specific task calculations, thereby improving model performance, reducing data requirements, accelerating training speed, and enhancing generalization ability. The specific steps and methods for the cloud processing module 32 to complete the pre-training of the deep learning model and determine the user's current pain level based on the current vital sign information data sequence can be found in the specific steps of the analysis module in Example 1, and will not be repeated here.

[0092] In one alternative implementation, the predicted blood drug concentration is determined based on the drug clearance rate constant and the drug distribution volume; the predicted effect-site concentration is determined based on the predicted blood drug concentration and the blood-effect-site transport constant; and the predicted pain level is determined based at least on the baseline pain level and the maximum analgesic effect that the drug can produce.

[0093] In an optional implementation, a control model is established using real-time prediction information of the user's injection status to optimize and determine a real-time drug injection plan that minimizes costs. Specifically, this includes: obtaining drug injection constraint expressions, which include: maximum infusion rate constraint and 24-hour maximum infusion dose constraint; establishing an optimization function by performing a weighted summation operation based on the predicted blood drug concentration, predicted pain level, and drug injection constraint expressions as cost terms; and determining a real-time drug injection plan that minimizes costs by minimizing the weighted sum of the three cost terms as the optimization objective.

[0094] In one optional implementation, vital sign information includes at least: heart rate, blood oxygen saturation, body temperature, blood pressure, and activity level. By collecting multimodal vital sign data, the user's pain status can be more accurately assessed, preventing drug misuse and abuse.

[0095] In an optional implementation, the method of this embodiment further includes: receiving an injection command via a button or automatically generating an injection command based on the usage permissions of the drug injection module; the drug injection module then completes the drug injection according to the injection command. Setting up buttons allows users to easily express their injection intentions, ensuring the user's true intentions are conveyed and facilitating operation. Automatically generating injection commands through the drug injection module allows for automatic injection even when the user cannot express their intentions. The combination of these two methods ensures both the user's true intentions are expressed and enables automated injection for users unable to express their intentions.

[0096] In an optional implementation, the method of this embodiment further includes: uploading the injection information to the back-end processing center after the drug injection is completed. Specifically, the actual operation information of the drug injection module (such as injection time, injection dosage, post-injection vital signs data, etc.) is fed back to the back-end processing center so that the back-end processing center can obtain the user's medical status in a timely manner and prevent duplicate injections. Synchronously uploading vital signs data and injection records to the back-end processing center enhances the safety of drug use.

[0097] In the description of the embodiments of the present invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "center," "top," "bottom," "top," "bottom," "inner," "outer," "inner side," and "outer side," etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. "Inner side" refers to the interior or enclosed area or space. "Outer perimeter" refers to the area surrounding a specific component or specific area.

[0098] In the description of embodiments of the present invention, the terms "first," "second," "third," and "fourth" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first," "second," "third," or "fourth" may explicitly or implicitly include one or more of that feature. In the description of the present invention, unless otherwise stated, "a plurality of" means two or more.

[0099] In the description of the embodiments of the present invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," "joining," and "assembly" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0100] In the description of embodiments of the present invention, specific features, structures, materials or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0101] In the description of the embodiments of the present invention, it should be understood that "-" and "~" represent a range between two numerical values, and this range includes the endpoints. For example, "AB" represents a range greater than or equal to A and less than or equal to B. "A~B" represents a range greater than or equal to A and less than or equal to B.

[0102] In the description of embodiments of the present invention, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A controlled drug injection monitoring system, characterized in that, include: Controlled drug injection monitoring armband and cloud processing module; The controlled drug injection monitoring armband includes: a monitoring module, an information transmission module, and a drug injection module; The monitoring module is used to collect vital sign information of the user wearing the controlled drug injection monitoring armband. The vital sign information refers to multiple sets of data used to characterize the degree of pain in the user's body. The information transmission module is used to send the vital signs information to the cloud processing module; The cloud processing module is used to construct a deep learning model, pre-train the deep learning model using the vital sign information to obtain a trained deep learning model, and acquire the user's current vital sign information data sequence in real time. The trained deep learning model is used to determine the pain level corresponding to the user's current vital sign information data sequence, and the pain level is transmitted to the information transmission module. The current vital sign information data sequence includes the vital sign information collected at a preset period. The information transmission module is also used to receive the pain level and transmit the pain level to the drug injection module; The drug injection module is used to acquire the pain level, estimate the user's injection status in real time based on the pharmacokinetic model and the pain level using an extended unscented Kalman filter system, establish a control model using the user's injection status in real time for optimization to determine the minimum cost drug injection real-time plan, and dynamically inject drugs according to the minimum cost drug injection real-time plan. The user's injection status in real time includes: predicted blood drug concentration, predicted effect-site concentration, and predicted pain level.

2. The controlled drug injection monitoring system according to claim 1, characterized in that, The step of determining the pain level corresponding to the current vital signs data sequence based on the trained deep learning model includes: The current vital signs information data sequence is grouped and aligned according to the time dimension, and a first multimodal vector is constructed. The first multimodal vector is fed into the trained deep learning model for computation to obtain first computational data. The first computational data is compared with preset reference data to determine the pain level corresponding to the current vital sign information data sequence.

3. The controlled drug injection monitoring system according to claim 2, characterized in that, The deep learning model includes a Transformer encoder and a multilayer perceptron. The step of feeding the first multimodal vector into the trained deep learning model for computation includes: The Transformer encoder uses a multi-head attention mechanism to operate on the first multimodal vector, and performs residual connection and normalization calculations to obtain the final layer output. The output of the last layer is then reconstructed using the multilayer perceptron.

4. The controlled drug injection monitoring system according to any one of claims 1 to 3, characterized in that, The step of pre-training the deep learning model using the vital sign information to obtain a trained deep learning model specifically includes: The information transmission module receives a sequence of vital sign information samples obtained by the monitoring module from the user's vital sign information collection. The vital sign information sample sequences are grouped and aligned according to the time dimension, and a second multimodal vector is constructed. The deep learning model is used to operate on the second multimodal vector to obtain the second reconstructed vector; Obtain a third reconstruction vector and a fourth reconstruction vector. The third reconstruction vector is obtained by applying the deep learning model to the third multimodal vector, and the fourth reconstruction vector is obtained by applying the deep learning model to the fourth multimodal vector. The third multimodal vector is constructed based on the user's vital signs information in a previous non-painful state, and the fourth multimodal vector is constructed based on the user's vital signs information in a previous painful state. A loss function is constructed based on the second reconstruction vector, the third reconstruction vector, and the fourth reconstruction vector; The parameters of the deep learning model are updated through iterative training until the loss function converges, thereby obtaining the trained deep learning model.

5. The controlled drug injection monitoring system according to any one of claims 1 to 3, characterized in that, The information transmission module receives the pain level but transmits it to the drug injection module before doing so. The information transmission module is also used to upload the pain level to the background processing center, receive the drug injection authorization information returned by the background processing center, and unlock the use permission of the drug injection module according to the drug injection authorization information.

6. The controlled drug injection monitoring system according to any one of claims 1 to 3, characterized in that, The predicted blood drug concentration is determined based on the drug clearance rate constant and the drug distribution volume; The predicted effect-compartment concentration is determined based on the predicted blood drug concentration and the blood-effect-compartment transport constant. The predicted pain level is determined based at least on the baseline pain level and the maximum analgesic effect that the drug can produce.

7. The controlled drug injection monitoring system according to any one of claims 1 to 3, characterized in that, The step of establishing a control model using the real-time prediction information of the user's injection status to optimize and determine a real-time drug injection plan that minimizes costs specifically includes: Obtain the drug injection constraint formula, which includes: maximum infusion rate constraint and 24-hour maximum infusion dose constraint; An optimization function is established by performing a weighted summation operation based on the predicted blood drug concentration, the predicted pain level, and the drug injection constraint as cost terms. The cost-minimizing real-time drug injection protocol is established with the optimization objective of minimizing the weighted sum of the three cost terms.

8. The controlled drug injection monitoring system according to any one of claims 1 to 3, characterized in that, The vital signs information includes at least: heart rate, blood oxygen saturation, body temperature, blood pressure, and activity level.

9. A controlled drug injection monitoring armband, characterized in that, include: The armband consists of a main body, a monitoring module, an analysis module, an information transmission module, and a drug injection module. The monitoring module is used to collect vital sign information of the user wearing the controlled drug injection monitoring armband through the armband body. The vital sign information refers to multiple sets of data used to characterize the degree of pain in the user's body. The analysis module is used to construct a deep learning model, pre-train the deep learning model using the vital sign information to obtain a trained deep learning model, and acquire the user's current vital sign information data sequence in real time. The trained deep learning model is used to determine the pain level corresponding to the user's current vital sign information data sequence, and the pain level is transmitted to the information transmission module. The current vital sign information data sequence includes the vital sign information collected at a preset period. The information transmission module is also used to receive the pain level, upload the pain level to the background processing center, and receive the drug injection authorization information returned by the background processing center, and unlock the use permission of the drug injection module according to the drug injection authorization information. The drug injection module is used to acquire the pain level, estimate the user's injection status in real time based on the pharmacokinetic model and the pain level using an extended unscented Kalman filter system, establish a control model using the user's injection status in real time for optimization to determine the minimum cost drug injection real-time plan, and dynamically inject drugs according to the minimum cost drug injection real-time plan. The user's injection status in real time includes: predicted blood drug concentration, predicted effect-site concentration, and predicted pain level.

10. The controlled drug injection monitoring armband according to claim 9, characterized in that, The drug injection module also includes: Replaceable injection sticks for storing the medication that the user needs to inject; An indwelling needle interface is used to inject medication into the user through an indwelling needle.

Images