Insulin pen infusion control system and method

Abstract

The invention relates to the technical field of medical instruments, and discloses an insulin pen infusion control system which comprises a processor module used for executing a system control strategy; the dose acquisition module is connected with the processor module and is used for acquiring a dose setting signal; the time measuring module is connected with the processor module; the injection resistance sensing module is arranged on the liquid medicine pushing path and used for detecting injection resistance parameters; the audio broadcasting module is connected with the processor module; triggering a first audio broadcast according to the output signal of the dose acquisition module and the timing parameter of the time measurement module; activating a second audio broadcast according to an initial trigger signal of the injection resistance sensing module; and triggering the third audio broadcast based on the preset duration parameter. The invention aims to solve the problems that an existing insulin injection device lacks a key operation node intelligent identification and active prompt mechanism, the abnormal state judgment accuracy is low, and accurate diagnosis of complex conditions cannot be realized.

Description

Technical Field

[0001] This invention relates to the field of medical device technology, specifically to an insulin pen infusion control system and method. Background Technology

[0002] Diabetic patients rely on insulin injection devices for long-term blood sugar control. Insulin pumps, as precision infusion devices, mimic the function of the pancreas to provide continuous subcutaneous insulin infusion. Traditional insulin injection devices use mechanical dosage adjustment mechanisms to set the injection dose, and a manual push button drives the piston to deliver the medication. Due to the high proportion of elderly and visually impaired individuals among diabetic patients, difficulties in identifying dosage windows and maintaining consistent injection technique have long been problems. Furthermore, abnormal conditions such as needle blockage or bending cannot be detected promptly, easily leading to injection failures or even device damage. To address these pain points, some insulin injection devices have recently begun integrating electronic control modules and sensor detection units, attempting to improve operational safety and user experience through technological means.

[0003] However, existing control systems lack intelligent identification and proactive prompting mechanisms for key operational nodes. Users cannot obtain real-time confirmation feedback during dosage setting, injection initiation, and injection completion, leading to increased operational uncertainty. On the other hand, although some solutions have introduced resistance monitoring functions, they generally use a single threshold judgment method and fail to establish a deep analysis model for resistance data. This results in low accuracy in identifying abnormal states and a delayed response, making it impossible to accurately diagnose complex situations such as needle blockage, changes in drug viscosity, and abnormal injection patterns. The intelligence and reliability of the system urgently need to be improved. Summary of the Invention

[0004] The purpose of this invention is to solve the problems of existing insulin injection devices lacking intelligent identification and active prompting mechanisms for key operation nodes, low accuracy in judging abnormal states, and inability to accurately diagnose complex situations. Therefore, an insulin pen infusion control system and method are proposed.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows:

[0006] An insulin pen infusion control system includes:

[0007] The processor module is used to execute system control strategies;

[0008] A dose acquisition module, connected to the processor module, is used to acquire dose setting signals;

[0009] The time measurement module is connected to the processor module;

[0010] The injection resistance sensing module is located on the drug delivery path and is used to detect injection resistance parameters.

[0011] An audio playback module is connected to the processor module;

[0012] The first audio broadcast is triggered based on the output signal of the dose acquisition module and the timing parameters of the time measurement module; the second audio broadcast is activated based on the initial trigger signal of the injection resistance sensing module; the third audio broadcast is triggered based on the preset duration parameter; and the fourth audio broadcast is triggered when the injection resistance parameter meets the abnormal judgment condition.

[0013] The processor module also integrates a resistance feature extraction algorithm, which constructs a resistance-time mapping relationship by analyzing the time-domain features of the injection resistance parameters. ,when At that time, among them This is a dynamic threshold that triggers an exception handling mechanism.

[0014] Based on the above technical solution, the present invention can be further improved as follows.

[0015] Furthermore, it also includes a biometric identification module, which is connected to the processor module and is used to collect and verify user identity features;

[0016] The biometric recognition module retrieves a corresponding personalized parameter set based on the biometric recognition results. The personalized parameter set includes a baseline value for the injection rate. Resistance threshold coefficient and time compensation factor ;

[0017] The dynamic threshold Calculate using the following formula:

[0018]

[0019] in, The standard threshold, This is due to the deviation in ambient temperature.

[0020] Furthermore, the injection resistance sensing module includes a main sensor and an auxiliary sensor, which are respectively set at different detection sites;

[0021] The processor module executes a dual-channel resistance fusion algorithm:

[0022]

[0023] in, , The outputs are from the main and auxiliary sensors, respectively. , For adaptive weighting coefficients, As compensation;

[0024] when When this happens, the system triggers the sensor self-test program;

[0025] It also includes a push-in pattern recognition module, which is connected to the processor module;

[0026] The injection pattern recognition module analyzes the first derivative of the injection force curve. and second derivative It identifies injection operation modes, including constant injection, accelerated injection, and intermittent injection;

[0027] The processor module dynamically adjusts the audio playback strategy and anomaly detection parameters based on the identified injection pattern.

[0028] Furthermore, it also includes a drug solution characteristic analysis module;

[0029] The drug solution characteristic analysis module detects the viscosity parameter of the drug solution using impedance spectroscopy analysis technology. and transmit it to the processor module;

[0030] The processor module corrects the injection resistance reference value in real time based on the viscosity parameter.

[0031]

[0032] in, This is the standard resistance value. This is the viscosity influence coefficient. Standard viscosity;

[0033] It also includes a needle status assessment module;

[0034] The needle condition assessment module constructs a resistance spectrum feature vector. ,in For the first The energy distribution of each frequency band is used to determine the needle state category using a support vector machine classifier.

[0035] The needle status includes four categories: normal, partially blocked, completely blocked, and bent / deformed. Each status corresponds to a different audio prompt.

[0036] Furthermore, it also includes an injection trajectory recording module and a predictive analysis module;

[0037] The injection trajectory recording module records the displacement-time series of the injection process. ;

[0038] The predictive analysis module uses historical trajectory data and a long short-term memory network to predict the remaining injection time. It will trigger a correction prompt when the deviation between the predicted value and the actual value exceeds the allowable range.

[0039] The audio broadcasting module includes a speech synthesis unit and a sound effect generation unit;

[0040] The speech synthesis unit uses parametric speech synthesis technology, which adjusts speech rate, pitch and volume parameters according to user preferences.

[0041] The sound effect generation unit generates different frequency combinations of prompt sounds, and transmits different levels of prompt information through frequency encoding.

[0042] Furthermore, it also includes an environmental sensing module for detecting ambient noise levels. ;

[0043] The processor module adaptively adjusts the audio output parameters according to the ambient noise level.

[0044]

[0045] in, This refers to the actual output volume. Based on the basic volume, This is the volume adjustment factor. This is the noise threshold.

[0046] An insulin pen infusion control method includes:

[0047] S1: Initialize system parameters and establish a baseline resistance model. ;

[0048] S2: Monitors dose adjustment status; upon detecting an adjustment operation, a delay judgment is initiated, with the delay duration specified. Dynamically determined based on historical operational practices:

[0049]

[0050] S3: After the delay condition is met, a dose confirmation broadcast is triggered, and the broadcast content includes the dose value and the recommended injection duration;

[0051] S4: Detect injection initiation characteristics, including resistance mutation rate. and duration ;

[0052] S5: Perform multi-dimensional monitoring during the injection process:

[0053] Real-time calculation of resistance change rate and comparison with historical models

[0054] Evaluation of injection uniformity index

[0055] Predicting the probability of anomalies

[0056] S6: Dynamically adjust control strategies and broadcast content based on monitoring results.

[0057] Based on the above technical solution, the present invention can be further improved as follows.

[0058] Furthermore, the probability of an anomaly occurring as described in S5 Calculations using Bayesian inference:

[0059]

[0060] in, For the current detection dataset, This is an abnormal event. This is a priori probability, derived from historical data statistics;

[0061] when When this happens, the system enters an early warning state and increases the monitoring frequency.

[0062] Furthermore, it also includes intelligent learning optimization steps:

[0063] S7: Collect the complete dataset for each bet. ,in For the evaluation of the inference results;

[0064] S8: The control parameters are optimized using a reinforcement learning algorithm, and the reward function is defined as follows:

[0065]

[0066] in, Rewards for time accuracy, To promote stability rewards Penalties for false alarms;

[0067] S9: Update control policy This maximizes the expected cumulative reward.

[0068] Furthermore, the baseline resistance model described in S1 is constructed using piecewise linear interpolation:

[0069]

[0070] in, , , These serve as baseline values ​​for each stage. , The rate of change is obtained by fitting historical data using the least squares method.

[0071] Compared with the prior art, the technical solution of this application has the following beneficial technical effects:

[0072] This invention integrates a multi-stage intelligent triggering mechanism into the processor module, setting audio broadcast responses for four key operation nodes: dose confirmation, injection initiation, injection completion, and anomaly detection. This allows users to receive real-time confirmation feedback at each crucial stage. Furthermore, it introduces a resistance feature extraction algorithm, performing time-domain feature analysis on injection resistance parameters and constructing a resistance-time mapping relationship. The traditional instantaneous threshold determination method is upgraded to one based on integral calculation. The dynamic judgment mode of this technology can comprehensively consider the changing trend and cumulative effect of the resistance value over time, effectively avoiding the risk of misjudgment caused by single-point sampling, and greatly improving the accuracy and robustness of abnormal state identification. Dynamic threshold The introduction of this technology enables the system to adaptively adjust the judgment criteria according to actual working conditions, overcoming the contradiction between sensitivity and reliability that cannot be balanced by fixed thresholds. By setting the injection resistance sensing module on the drug delivery path to achieve real-time monitoring throughout the process, and combined with the deep analysis capabilities of the processor module, the system can accurately identify various complex abnormalities such as needle blockage, abnormal drug viscosity, and improper injection force, and trigger corresponding abnormality handling mechanisms and audio warnings. This achieves a technological leap from passive detection to active diagnosis, and the overall intelligence level and system reliability are qualitatively improved, providing diabetic patients with a safer, more convenient, and intelligent insulin infusion solution. Attached Figure Description

[0073] Figure 1 This is a flowchart of the overall control system structure of the present invention;

[0074] Figure 2 This is a flowchart of the biometric identification and personalized parameter process of the present invention;

[0075] Figure 3 This is a flowchart of the injection resistance sensing and pattern recognition process of the present invention;

[0076] Figure 4 This is a flowchart for evaluating the drug solution properties and needle condition of the present invention. Detailed Implementation

[0078] 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.

[0079] like Figure 1 As shown, an insulin pen infusion control system of the present invention includes:

[0080] The processor module is used to execute system control strategies;

[0081] A dose acquisition module, connected to the processor module, is used to acquire dose setting signals;

[0082] The time measurement module is connected to the processor module;

[0083] The injection resistance sensing module is located on the drug delivery path and is used to detect injection resistance parameters.

[0084] An audio playback module is connected to the processor module;

[0085] The first audio broadcast is triggered based on the output signal of the dose acquisition module and the timing parameters of the time measurement module; the second audio broadcast is activated based on the initial trigger signal of the injection resistance sensing module; the third audio broadcast is triggered based on the preset duration parameter; and the fourth audio broadcast is triggered when the injection resistance parameter meets the abnormal judgment condition.

[0086] The processor module also integrates a resistance feature extraction algorithm, which constructs a resistance-time mapping relationship by analyzing the time-domain features of the injection resistance parameters. ,when At that time, among them This is a dynamic threshold that triggers an exception handling mechanism.

[0087] like Figure 2 As shown, it also includes a biometric identification module, which is connected to the processor module and is used to collect and verify user identity features;

[0088] The biometric recognition module retrieves a corresponding personalized parameter set based on the biometric recognition results. The personalized parameter set includes a baseline value for the injection rate. Resistance threshold coefficient and time compensation factor ;

[0089] The dynamic threshold Calculate using the following formula:

[0090]

[0091] in, The standard threshold, To accommodate ambient temperature variations, the system integrates a biometric recognition module, which can be a fingerprint sensor or an iris recognition device, located in the grip area of ​​the pen. Users register their fingerprint or iris information upon first use, and the system creates a unique set of personalized parameters for each user.

[0092] The personalized parameter set includes three core parameters: the baseline value of injection rate. The injection rate is set according to the user's injection habits, typically ranging from 0.5 to 2.0 units per second; resistance threshold coefficient. Adjustments are made to suit different users' hand strength, with typical values ​​between 0.8 and 1.2; time compensation factor. This setting is used to adapt to changes in the viscosity of the drug solution under different ambient temperatures, and is generally set at 0.01-0.05℃. -1 .

[0093] Dynamic threshold The calculation takes into account individual differences and environmental factors, and the calculation formula is as follows:

[0094]

[0095] in The system standard threshold (e.g., set to 200 N·s), This represents the deviation of the current ambient temperature from the standard temperature (25°C). For example, when the ambient temperature is 15°C, =−10℃, if the user's =1.1, =0.03, then =200×1.1×(1−0.3)=154 N·s.

[0096] like Figure 3 As shown, the injection resistance sensing module includes a main sensor and an auxiliary sensor, which are respectively set at different detection sites;

[0097] The processor module executes a dual-channel resistance fusion algorithm:

[0098]

[0099] in, , The outputs are from the main and auxiliary sensors, respectively. , For adaptive weighting coefficients, As compensation;

[0100] when When this happens, the system triggers the sensor self-test program;

[0101] It also includes a push-in pattern recognition module, which is connected to the processor module;

[0102] The injection pattern recognition module analyzes the first derivative of the injection force curve. and second derivative It identifies injection operation modes, including constant injection, accelerated injection, and intermittent injection;

[0103] The processor module dynamically adjusts the audio playback strategy and anomaly detection parameters based on the identified injection pattern. The injection resistance sensing module adopts a dual-sensor configuration. The main sensor is a piezoresistive micro-force sensor, integrated on the push rod directly below the injection button, with a measurement range of 0-50N and an accuracy of ±0.1N. The auxiliary sensor is a thin-film strain gauge, attached to the inner wall of the cartridge holder, to sense the deformation of the cartridge under force.

[0104] The processor module executes a dual-channel resistance fusion algorithm to calculate the fused resistance value in real time.

[0105]

[0106] Weighting coefficient and The system adaptively adjusts based on the sensor signal quality; when the main sensor signal is stable... =0.7, =0.3; When main sensor drift is detected, the weight is automatically adjusted to... =0.5, =0.5. Compensation Item To correct the effect of temperature on the sensor's zero point, it is usually achieved by measuring the temperature in real time using a temperature sensor and then compensating by looking up a table.

[0107] System set tolerance threshold ,when If the readings of the two sensors differ too much, it indicates a possible sensor malfunction. In this case, the sensor self-test program is triggered, the audio broadcast is paused, and the LED indicator prompts the user to check the device.

[0108] The injection pattern recognition function is achieved by analyzing the injection force curve. The system samples at a frequency of 100Hz and calculates the injection force in real time. The first and second derivatives. When And if it lasts for more than 3 seconds, it is judged as a uniform injection mode; when and It was determined to be an accelerated betting mode; when detected If the value drops to near zero multiple times within 0.5 seconds and then rises again, it is determined to be an intermittent injection pattern.

[0109] For different injection modes, the processor dynamically adjusts the broadcast strategy: in the constant speed injection mode, it broadcasts "Injection complete" every 20 seconds as standard; in the accelerated injection mode, it broadcasts "Please maintain constant speed" at 15 seconds; and in the intermittent injection mode, it broadcasts "Please continue injection" at 25 seconds.

[0110] like Figure 4 As shown, it also includes a drug solution characteristic analysis module;

[0111] The drug solution characteristic analysis module detects the viscosity parameter of the drug solution using impedance spectroscopy analysis technology. and transmit it to the processor module;

[0112] The processor module corrects the injection resistance reference value in real time based on the viscosity parameter.

[0113]

[0114] in, This is the standard resistance value. This is the viscosity influence coefficient. Standard viscosity;

[0115] It also includes a needle status assessment module;

[0116] The needle condition assessment module constructs a resistance spectrum feature vector. ,in For the first The energy distribution of each frequency band is used to determine the needle state category using a support vector machine classifier.

[0117] The needle status is categorized into four types: normal, partially blocked, completely blocked, and bent / deformed. Each status corresponds to a different audio prompt. The drug solution characteristic analysis module uses impedance spectroscopy technology, integrating a micro-electrode pair at the bottom of the cartridge. By applying a weak AC signal at a frequency of 1kHz, the complex impedance of the drug solution is measured, thereby calculating the viscosity parameters. The viscosity of standard insulin (U-100) at 25°C ≈2.0 mPa·s.

[0118] The injection resistance baseline value is corrected in real time based on viscosity, using the following formula:

[0119]

[0120] in =15 N is the reference resistance under standard conditions, viscosity influence coefficient. =0.25. For example, if detected =3.0 mPa·s (low temperature or high concentration insulin), then =15×(1+0.25×ln1.5)≈16.5 N, and the system adjusts the resistance anomaly judgment threshold accordingly to reduce false alarms.

[0121] The needle condition assessment module utilizes spectrum analysis technology. The system performs a Fast Fourier Transform on the resistance signal, dividing it into 10 frequency bands (0-5Hz, 5-10Hz, ..., 45-50Hz), and calculates the energy distribution of each band to form a feature vector. .

[0122] The trained support vector machine classifier classifies the needle state into four categories:

[0123] Normal state: Low-frequency energy dominates. >0.6, standard broadcast prompt

[0124] Partial blockage: Increased mid-frequency energy. When the value exceeds 0.4, the system will announce, "Slight resistance detected, please observe carefully."

[0125] Complete blockage: high-frequency thrusting, >0.3, announces "Needle clogged, please replace immediately"

[0126] Bending deformation: characterized by intermittent high resistance. and Alternating peak values ​​are displayed, accompanied by a message stating "The needle may be bent; replacement is recommended."

[0127] It also includes an injection trajectory recording module and a predictive analysis module;

[0128] The injection trajectory recording module records the displacement-time series of the injection process. ;

[0129] The predictive analysis module uses historical trajectory data and a long short-term memory network to predict the remaining injection time. The module triggers a correction prompt when the deviation between the predicted and actual values ​​exceeds the allowable range. The push rod trajectory recording module records the push rod displacement through a rotary encoder, with a sampling interval of 100ms, generating a displacement-time series. Each injection generates approximately 200 data points.

[0130] The predictive analysis module uses a three-layer LSTM network (64 input units, 32 hidden units, and 1 output unit), taking displacement and resistance data from the last 10 time steps as input, to predict the remaining injection time. For example, after 10 seconds of betting, if the current status is entered, the network prediction will take another 9 seconds to complete; if it is not completed after 15 seconds, the deviation exceeds the allowable range (set to ±3 seconds), and the system triggers a correction prompt "Bet speed is too slow, please increase the intensity slightly".

[0131] The audio broadcasting module includes a speech synthesis unit and a sound effect generation unit;

[0132] The speech synthesis unit uses parametric speech synthesis technology, which adjusts speech rate, pitch and volume parameters according to user preferences.

[0133] The sound effect generation unit generates different frequency combinations of prompts, transmitting different levels of prompt information through frequency encoding. The audio broadcast module comprises two functional units. The speech synthesis unit employs parametric TTS technology, allowing users to set the speech rate (0.8-1.5x speed), tone (male / female voice), and volume (30-90dB) in the accompanying app. The sound effect generation unit generates different frequency combinations based on the prompt level: ordinary prompts use a single audio frequency (1000Hz), important prompts use dual audio frequencies (800Hz + 1200Hz), and emergency alarms use a rapid switching of three audio frequencies (500Hz, 1000Hz, and 1500Hz each sounding for 200ms).

[0134] It also includes an environmental sensing module for detecting ambient noise levels. ;

[0135] The processor module adaptively adjusts the audio output parameters according to the ambient noise level.

[0136]

[0137] in, This refers to the actual output volume. Based on the basic volume, This is the volume adjustment factor. To set a noise threshold, the environmental sensing module integrates a MEMS microphone to continuously monitor the ambient noise level. (Unit: dB). The system adaptively adjusts audio output parameters based on noise levels.

[0138]

[0139] Basic volume =60 dB, suitable for quiet environments; volume adjustment coefficient =0.5; Noise threshold =50 dB. When the user uses it outdoors in a noise level of 70 dB, =60 + 0.5 × (70 − 50) = 70 dB, ensuring the prompt tone can be clearly heard.

[0140] An insulin pen infusion control method includes:

[0141] S1: Initialize system parameters and establish a baseline resistance model. ;

[0142] S2: Monitors dose adjustment status; upon detecting an adjustment operation, a delay judgment is initiated, with the delay duration specified. Dynamically determined based on historical operational practices:

[0143]

[0144] S3: After the delay condition is met, a dose confirmation broadcast is triggered, and the broadcast content includes the dose value and the recommended injection duration;

[0145] S4: Detect injection initiation characteristics, including resistance mutation rate. and duration ;

[0146] S5: Perform multi-dimensional monitoring during the injection process:

[0147] Real-time calculation of resistance change rate and comparison with historical models

[0148] Evaluation of injection uniformity index

[0149] Predicting the probability of anomalies

[0150] S6: Dynamically adjust control strategies and broadcast content based on monitoring results.

[0151] The probability of an anomaly occurring as described in S5 Calculations using Bayesian inference:

[0152]

[0153] in, For the current detection dataset, This is an abnormal event. This is a priori probability, derived from historical data statistics;

[0154] when When this happens, the system enters an early warning state and increases the monitoring frequency.

[0155] It also includes intelligent learning optimization steps:

[0156] S7: Collect the complete dataset for each bet. ,in For the evaluation of the inference results;

[0157] S8: The control parameters are optimized using a reinforcement learning algorithm, and the reward function is defined as follows:

[0158]

[0159] in, Rewards for time accuracy, To promote stability rewards Penalties for false alarms;

[0160] S9: Update control policy To maximize the expected cumulative reward, the S1 initialization phase involves loading the user-specific benchmark resistance model after the system powers on. The model is constructed using piecewise linear interpolation:

[0161]

[0162] The parameters were obtained by fitting data from the user's first five uses using the least squares method. Typical values ​​are: =5 N (initial starting resistance), =1.5 N / s (slope of acceleration segment), =2 s, =8 N, =0.5 N / s (slope of the uniform velocity segment), =18s, =16 N (steady-state resistance).

[0163] S2 Dosage Monitoring: After the dose acquisition module (rotary encoder) detects that the knob has stopped rotating, a delay judgment is initiated. Delay duration. It is not a fixed value, but rather it adjusts adaptively based on user operating habits:

[0164]

[0165] in =2 s as the base delay, =0.5 is the adjustment coefficient. This represents the duration of the previous n pauses (usually n=10). This is the average value. If users prefer quick settings, Reduced to 2.5 seconds; if user operation is slow, Extended to 4 seconds to avoid premature false alarms.

[0166] S3 Dosage Broadcast: After the delay is met, a broadcast will be triggered stating "XX units have been set, recommended injection duration is XX seconds". The recommended duration is calculated at 2 seconds per unit. For example, if 12 units are set, the broadcast will state "12 units have been set, recommended injection duration is 24 seconds".

[0167] S4 Injection Startup Detection: The system simultaneously determines two conditions: resistance mutation rate. =10 N / s, and duration =0.3 s. This design can filter out accidental triggering by users tapping the button.

[0168] S5 Multidimensional Monitoring: During the injection process, the system performs three parallel monitoring functions:

[0169] Resistance Comparison: Calculates the deviation between the current resistance and the model resistance every second. ,like Continue for 2 seconds to trigger a resistance alarm.

[0170] Uniformity assessment: Calculate the coefficient of variation of injection rate. ,in For the speed standard deviation, This represents the average speed. This indicates that the injection is uniform. The system will broadcast the message "Please maintain a constant speed".

[0171] Anomaly probability prediction: Calculating the probability of anomalies occurring through Bayesian inference.

[0172]

[0173] The current dataset Parameters including resistance, speed, and time, and prior probabilities Based on historical statistical settings (e.g., a prior probability of needle blockage is 0.05). When When this happens, the system enters an early warning state, and the monitoring frequency is increased from 1Hz to 5Hz.

[0174] S6 Strategy Adjustment: Based on the monitoring results of S5, dynamically optimize the broadcast content and timing. For example, if it is detected that a user's betting speed is too slow, an additional broadcast will be made at the 10-second mark: "The current speed is too slow, please increase the speed appropriately."

[0175] S7-S9 Intelligent Learning: The system collects complete datasets. ,in To evaluate the injection results (complete / incomplete), the control parameters were optimized using a reinforcement learning algorithm, and the reward function was designed as follows:

[0176]

[0177] Time accuracy reward A perfect score of 1.0 is awarded if the actual duration deviates from the recommended duration by within ±2 seconds; 0.2 points are deducted for each second of deviation. Stability is rewarded. Uniformity 1.0 point is awarded, with points decreasing proportionally; false alarms will be penalized. 0.5 points will be deducted for each false alarm. The weighting is set as follows: , .

[0178] After approximately 30 rounds of learning through push betting, the system can implement user-specific control strategies. (Given state) Select action The probability of accumulating rewards is optimized to the best, increasing the expected cumulative reward by approximately 25%.

[0179] The baseline resistance model described in S1 is constructed using piecewise linear interpolation:

[0180]

[0181] in, , , These serve as baseline values ​​for each stage. , The rate of change was obtained by fitting historical data using the least squares method. The three-stage design of the benchmark resistance model is based on the mechanical characteristics of the actual injection process: the initial stage (0-2s) is the stage of overcoming static friction and initial compression of the drug, during which the resistance increases linearly; the middle stage (2-18s) is the stable injection stage, during which the resistance increases slowly; and the final stage (after 18s) is when the drug is about to be exhausted, the plunger is close to the bottom, and the resistance tends to be constant. The parameters for each stage were obtained by collecting injection data from 50 users for 10 injections each, and piecewise least squares fitting was used. The model fit goodness of fit was... .

[0182] Intelligent voice-guided injection is achieved through multi-module collaborative operation. Upon system power-up, the biometric recognition module verifies the user's identity and loads a personalized parameter set, including a baseline injection rate, resistance threshold coefficient, and time compensation factor. Simultaneously, the processor module initializes a user-specific baseline resistance model. When the user rotates the dosage knob to set the injection dose, the dosage acquisition module monitors the knob's rotation angle in real time via a rotary encoder, converting it into drug units. After the knob pauses for an adaptive delay, the audio playback module provides a voice prompt indicating the set dose and suggested injection duration. When the user presses the injection button to begin injection, the injection resistance sensing module's dual sensors simultaneously collect data on the push rod resistance and the force on the cartridge. The processor module executes a fusion algorithm to calculate the overall resistance value and compares it in real time with the baseline resistance model to determine if the deviation exceeds the dynamic threshold range. The time measurement module simultaneously starts timing, recording the injection displacement-time trajectory, and the predictive analysis module predicts the remaining injection time based on an LSTM network. During injection, the drug characteristic analysis module measures the drug viscosity using impedance spectroscopy and corrects the resistance baseline value, while the needle status assessment module performs spectral analysis on the resistance signal to identify whether the needle is blocked or deformed. The processor module integrates multi-dimensional monitoring data to assess the uniformity of the injection speed and calculate the probability of anomalies. When slow injection speed, abnormal resistance, or needle blockage is detected, the audio broadcast module adaptively adjusts the volume based on the ambient noise level, promptly broadcasting corresponding prompts or warnings. When the injection time approaches the recommended value, a reminder of the remaining time is broadcast, and upon completion, "Injection complete" is announced. After each injection, the system collects a complete dataset and optimizes control parameters through reinforcement learning algorithms, continuously adjusting personalized thresholds and broadcast strategies to achieve adaptive optimization. The entire process, from dosage setting to injection completion, provides full voice guidance and intelligent monitoring, ensuring accurate, safe, and convenient injection.

[0183] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0184] 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. An insulin pen infusion control system, comprising: Comprise: A processor module for executing system control strategy; A dose acquisition module connected with the processor module for collecting dose setting signal; A time measurement module connected with the processor module; A bolus resistance sensing module arranged on the drug liquid pushing path for detecting bolus resistance parameter; An audio broadcast module connected with the processor module; Trigger the first audio broadcast according to the output signal of the dose acquisition module and the timing parameter of the time measurement module; activate the second audio broadcast according to the initial trigger signal of the bolus resistance sensing module; trigger the third audio broadcast based on the preset time length parameter; trigger the fourth audio broadcast when the bolus resistance parameter meets the abnormal judgment condition; The processor module is integrated with a resistance feature extraction algorithm, and a resistance-time mapping relationship is constructed through time domain feature analysis of the bolus injection resistance parameter When , wherein is a dynamic threshold value, triggering an abnormal processing mechanism.

2. The insulin pen infusion control system of claim 1, wherein, Further comprising a biometric identification module connected with the processor module for collecting and verifying user identity characteristics; The biometric recognition module retrieves a corresponding set of personalized parameters based on the biometric recognition result, the set of personalized parameters including a bolus rate reference value , a resistance threshold coefficient , and a time compensation factor ; The dynamic threshold is calculated according to the following formula: wherein, is a standard threshold value, is an ambient temperature deviation.

3. The insulin pen infusion control system of claim 1, wherein, The bolus resistance sensing module comprises a main sensor and an auxiliary sensor arranged at different detection sites respectively; The processor module executes double channel resistance fusion algorithm: wherein, , are the main and auxiliary sensor outputs, respectively, , is an adaptive weight coefficient, is a compensation term; When the system triggers a sensor self-test procedure; Further comprising a bolus mode identification module connected with the processor module; The bolus mode recognition module recognizes the bolus operation mode by analyzing the first derivative and the second derivative of the bolus force curve, including uniform bolus, accelerated bolus and intermittent bolus. The processor module dynamically adjusts the audio broadcast strategy and abnormal judgment parameter according to the identified bolus mode.

4. The insulin pen infusion control system of claim 3, wherein, Further comprising a drug liquid characteristic analysis module; The drug solution characteristic analysis module detects the viscosity parameter of the drug solution using impedance spectroscopy analysis technology. and transmit it to the processor module; The processor module corrects the bolus resistance reference value in real time according to the viscosity parameter: wherein, R is the standard resistance value, is the viscosity influence coefficient, is the standard viscosity; Further comprising a needle state evaluation module; The needle state evaluation module constructs a resistance spectrum feature vector wherein is the energy distribution of the th frequency band, and a support vector machine classifier is used to determine the needle state category. The needle state includes four categories of normal, partial blockage, complete blockage and bending deformation, each state corresponds to different audio prompt content.

5. The insulin pen infusion control system of claim 1, wherein, Further comprising a bolus trajectory recording module and a prediction analysis module; The bolus trajectory recording module records a displacement-time sequence of the bolus procedure ; The prediction analysis module predicts the remaining bolus time based on historical trajectory data using a long short-term memory network and triggers a correction prompt when the predicted value deviates from the actual value beyond an allowable range. The audio broadcast module comprises a speech synthesis unit and an audio effect generation unit; The speech synthesis unit adopts parameterized speech synthesis technology to adjust the parameters of speech rate, tone and volume according to user preference settings; The audio effect generation unit generates prompt sound of different frequency combinations to deliver different levels of prompt information through frequency coding.

6. The insulin pen infusion control system of claim 5, wherein, Also included is an environmental awareness module for detecting environmental noise levels ; The processor module adjusts the audio output parameter adaptively according to the environmental noise level: wherein, is the actual output volume, is the base volume, is the volume adjustment coefficient, is the noise threshold.

7. An insulin pen infusion control method, characterized by, Applied to the system of any one of claims 1 to 9, comprising: S1 : initialize system parameters, establish a reference resistance model ; S2: monitor the dose adjustment state, start the delay determination after detecting the adjustment operation, the delay duration Dynamically determine according to historical operation habits: S3: after meeting the delay condition, trigger dose confirmation broadcast, the broadcast content includes dose value and recommended bolus time; S4: detecting a bolus initiation feature, the feature comprising a rate of change of resistance and duration ; S5: execute multi-dimensional monitoring during bolus process: Real-time calculation of resistance change rate and comparison with historical model Assessing bolus uniformity metrics Probability of abnormal occurrence prediction S6: dynamically adjust control strategy and broadcast content based on monitoring result.

8. An insulin pen infusion control system and method according to claim 7, wherein, In S5 the probability of the anomaly occurring By Bayesian inference calculation: wherein, is the current detection data set, is the abnormal event, is the prior probability, which is obtained based on historical data statistics; When the system enters a pre-alarm state and increases the monitoring frequency.

9. An insulin pen infusion control system and method according to claim 7, wherein, Further comprising intelligent learning optimization steps: S7: Collecting the complete data set for each bolus wherein is the bolus result evaluation; S8: optimize control parameters by using reinforcement learning algorithm, and the reward function is defined as: wherein, is a time accuracy reward, is a bolus smoothness reward is a false positive penalty; S9: update control policy maximize the cumulative reward expectation.

10. The insulin pen infusion control system and method thereof according to claim 7, wherein, The reference resistance model in S1 is constructed by piecewise linear interpolation: wherein, , , is the reference value for each stage, , is the rate of change, obtained by least square fitting of historical data.

Images