Sinking compensation method and system for CGM sensor

Abstract

The invention relates to the technical field of blood glucose detection, and discloses a sag compensation method and system for a CGM sensor, and the method comprises the steps: carrying out the time domain feature and physiological correlation feature extraction of a collected blood glucose related current signal, generating fusion feature data, firstly judging whether the sag is a short-term sag or not based on a preset time domain dimension rule, and carrying out the recognition of the short-term sag; the method can accurately lock the time domain characteristics of the non-physiological sag, and can achieve the effective recognition of the non-physiological signal sag at night. Precise distinguishing of signal causes is realized through a two-dimensional judgment rule, verification is performed through a physiological dimension rule when short-term subsidence is judged in a time domain dimension, the double-layer identification logic of time domain feature preliminary screening and physiological feature verification avoids signal confusion from the aspect of mechanism, the misjudgment rate can be reduced, and the accuracy of signal identification is improved. The risk that hypoglycemia is missed to be reported or wrong intervention is caused is prevented, and the technical problem that real hypoglycemia signals are easily corrected by mistake or non-physiologically sunken into hypoglycemia is misjudged in night monitoring is solved.

Description

Technical Field

[0001] This invention relates to the field of blood glucose detection technology, and in particular to a method and system for indentation compensation of CGM sensors. Background Technology

[0002] Continuous glucose monitoring (CGM) systems are key devices for daily blood glucose management in diabetic patients. They generate continuous blood glucose monitoring results by collecting glucose concentration-related current signals from subcutaneous interstitial fluid in real time, providing patients with a basis for avoiding the risks of hypoglycemia and hyperglycemia and adjusting treatment plans. Especially during the nighttime sleep period, accurate monitoring is crucial for avoiding serious consequences such as hypoglycemic coma.

[0003] However, changes in sleep posture during the night can easily cause pressure or slight displacement of the sensor, resulting in a short-term dip in the blood glucose-related electrical signal. Furthermore, when a subject is experiencing true hypoglycemia, this non-physiological signal dip is difficult to distinguish from the decrease in electrical signal corresponding to physiological hypoglycemia, easily leading to confusion between physiological dips and true hypoglycemia signals. If a true hypoglycemia signal is mistakenly identified as a dip caused by sensor pressure and corrected accordingly, the risk of true hypoglycemia will be masked, leading to missed detections and delayed interventions. Conversely, if a non-physiological dip is mistakenly identified as true hypoglycemia, false alarms and unnecessary blood glucose-lowering interventions will occur, seriously threatening the patient's health.

[0004] In summary, existing technologies cannot effectively identify non-physiological signal depression caused by nighttime sleep posture, nor can they distinguish the cause of signal depression when the subject is experiencing true hypoglycemia. This makes it easy to miscorrect true hypoglycemia signals or misjudge non-physiological depression as hypoglycemia, resulting in low reliability and safety of nighttime monitoring. Summary of the Invention

[0005] This invention provides a depression compensation system for CGM sensors, which solves the technical problem that nighttime monitoring is prone to miscorrecting the true hypoglycemia signal or misjudging non-physiological depression as hypoglycemia.

[0006] To solve the above-mentioned technical problems, this application provides the following technical solution: A depression compensation system for a CGM sensor, comprising: The signal acquisition module is used to acquire the blood glucose-related current signal of the subject at night and transmit the blood glucose-related current signal to the signal processing module in real time. The signal processing module is used to receive the blood glucose-related current signal, extract its time-domain features and physiological correlation features to obtain fused feature data of blood glucose change rate, and then transmit the fused feature data to the signal identification module. The signal identification module is used to receive the fused feature data and analyze it based on a preset two-dimensional judgment rule: first, it determines whether it is a short-term depression through the time domain dimension rule, and then verifies whether it is a real hypoglycemia through the physiological dimension rule, so as to obtain the identification result and the corresponding target signal segment, and transmit it to the depression compensation module. The depression compensation module is used to receive the identification result and the target signal segment. If the identification result is a non-physiological depression, the baseline data of the signal before the depression in the target signal segment is extracted. Combined with the dynamic trend data of the overall blood glucose signal, the target signal segment is corrected by a linear extrapolation algorithm to obtain the compensated current signal. If it is a true hypoglycemia, the original target signal segment is retained as the compensated current signal, and the compensated current signal is transmitted to the blood glucose estimation module. The blood glucose estimation module is used to receive the compensated current signal, process it based on a preset blood glucose value calculation model, obtain accurate nighttime blood glucose monitoring results, and output them.

[0007] The basic principle and beneficial effects of the solution are as follows: Temporal and physiological features are extracted from the collected blood glucose-related current signals to generate fused feature data. Based on preset temporal dimension rules, it is first determined whether the depression is short-term, which can accurately identify the temporal features of non-physiological depressions. This enables effective identification of non-physiological signal depressions at night. The dual-dimensional judgment rules achieve accurate differentiation of signal causes. When a short-term depression is determined in the temporal dimension, it is verified by physiological dimension rules. This dual-layer identification logic of initial screening of temporal features and verification of physiological features avoids signal confusion, reduces the false positive rate, and prevents missed hypoglycemia risks or erroneous interventions.

[0008] Due to the low reliability and safety of nighttime monitoring, for non-physiological dips, the baseline signal data before the dip is extracted and the target signal segment is corrected by combining it with the overall dynamic trend of blood glucose signals to ensure that the compensated signal closely matches the actual blood glucose change trend. For true hypoglycemia signals, the original segment is directly retained. This differentiated processing and precise compensation design achieves the dual goals of accurately compensating for non-physiological signals and completely preserving true hypoglycemia signals. It avoids interference from non-physiological dips on the blood glucose curve, ensuring the reliability of monitoring results, and prevents true hypoglycemia signals from being incorrectly corrected, thus avoiding the risk of misjudgment and missed judgment.

[0009] Furthermore, the signal acquisition module is also used to collect the subject's physiological characteristic data at night and transmit the physiological characteristic data to the signal processing module in real time; After receiving the physiological feature data, the signal processing module performs feature quantization processing on it, and integrates the quantized physiological feature data with the fused feature data to generate joint feature data containing time-domain features, physiological correlation features and quantized physiological features, which is then transmitted to the signal identification module. The signal identification module analyzes the joint feature data based on a preset two-dimensional judgment rule: based on the time domain dimension rule determining it as a short-term depression, it uses quantified physiological feature data to corroborate whether it is a non-physiological depression.

[0010] The beneficial effects are as follows: Since it is difficult to accurately distinguish between non-physiological depression and true hypoglycemia signals based solely on blood glucose-related current signals, after determining short-term depression in the time domain, quantitative physiological characteristic data is used for corroboration. This dual-data-dimensional identification mode of blood glucose signals and physiological characteristics makes up for the blind spots of identification based on a single blood glucose signal, reduces the confusion rate between non-physiological depression and true hypoglycemia, thereby reducing the risk of misjudgment, making the identification results more physiologically based, and improving the accuracy and safety of nighttime monitoring.

[0011] Furthermore, the signal processing module is also used to receive the acquisition timestamps of blood glucose-related current signals and physiological characteristic data, and calculate the time difference Δt between the two: If the time difference Δt ≤ preset synchronization threshold, then the blood glucose related current signal and physiological characteristic data are determined to be synchronous data collected at the same time. The signal processing module transmits the integrated joint feature data to the signal identification module. If the time difference Δt is greater than the preset synchronization threshold, then the two are determined to be asynchronous data, and the signal processing module only transmits the fusion feature data extracted based on the blood glucose-related current signal to the signal identification module.

[0012] The beneficial effect is that, since non-physiological depressions are easily confused with true hypoglycemia signals, if there is a time delay between the collection of physiological characteristic data and blood glucose signals, data misalignment will occur, which will increase the probability of misjudgment. By calculating the relationship between the time difference between the two and a preset synchronization threshold, data validity screening is achieved. This time synchronization verification mechanism eliminates misalignment interference from the data source, prevents time misalignment, and reduces the misjudgment rate caused by data asynchrony.

[0013] Furthermore, the signal processing module is also used to obtain the acquisition timestamp T1 of the blood glucose-related current signal transmitted by the signal acquisition module and the acquisition timestamp T2 of the physiological characteristic data, and calculate the time delay between the two ΔT=|T1-T2|. If ΔT=0, it is determined to be synchronous data without delay, and the data is integrated to generate joint feature data and transmitted to the signal identification module. If ΔT > 0, it is determined to be asynchronous data with a time delay. The signal processing module calls the historical data cache library and performs time calibration on the two types of signals based on the changing trends of the historical data: For blood glucose-related current signals, a corrected current signal at time (T1+T2) / 2 is generated using a linear extrapolation method. For the physiological characteristic data, the moving average method was used to generate corrected physiological characteristic data at time (T1+T2) / 2. The corrected time-matched synchronization data is integrated into a joint feature data transmission and sent to the signal identification module.

[0014] The beneficial effects are as follows: relying on the collaborative corroboration of multi-source data will discard some physiological feature data. Therefore, by calculating the time delay, the asynchronous data can be calibrated in a targeted manner to ensure that the time of the two types of data is matched. This avoids the limitations of data discarding and avoids the misalignment interference of asynchronous data, improves the utilization rate of physiological feature data, and reduces the misjudgment rate of the calibrated data.

[0015] Furthermore, the signal processing module establishes a dynamic time synchronization mechanism through a reference time source. When the signal acquisition module acquires blood glucose-related current signals, it records a reference timestamp T1'. When acquiring physiological characteristic data, it synchronously records a reference timestamp T2'. After receiving T1' and T2', the signal processing module calculates the initial time delay ΔT'=|T1'-T2'| and starts dynamic monitoring: extracts a set of ΔT' values ​​of synchronization data at each preset time interval, constructs a delay monitoring sequence, and calculates the delay mean μ and standard deviation σ through a sliding window algorithm. When ΔT' is within the range of [μ-2σ, μ+2σ], it is determined to be a time delay within the normal range; when ΔT' exceeds this range, it is determined to be an abnormal time delay, triggering a synchronization calibration command.

[0016] The beneficial effects are as follows: Since signal acquisition is prone to drift due to local clock and crystal oscillator deviation, the consistency of the timing dimension of the timestamp is ensured by synchronizing with a unified reference time source; at the same time, the mean and standard deviation of the delay monitoring sequence are calculated to dynamically divide the normal and abnormal delay intervals to trigger recalibration, eliminate the cumulative error of clock drift, reduce the timing error of the timestamp, and improve the accuracy of delay determination.

[0017] Furthermore, the dual-dimensional judgment rule achieves the merging and integration of the temporal and physiological dimensions through feature weight allocation: A fusion judgment value S is set, and the calculation formula is S = α×A+β×B +γ×C +δ×D, where A is the time-domain feature parameter, B is the duration parameter, C is the recovery trend parameter, D is the physiological feature parameter, and α, β, γ, and δ are weighting coefficients. When S≤ the preset fusion judgment value, it is judged as non-physiological depression; when S> the preset fusion judgment value, it is judged as true hypoglycemia.

[0018] The beneficial effects are as follows: Since staged identification is prone to misjudgment due to the rigidity of single feature thresholds, the quantitative integration of time domain and physiological dimensions is achieved through feature weight allocation, which upgrades the identification logic from threshold judgment to multi-feature comprehensive scoring, thereby improving the accuracy and flexibility of signal identification.

[0019] Furthermore, when the depression compensation module extracts the signal baseline data before the depression in the target signal segment, it pre-determines the depression time point: High-frequency sampling analysis is performed on the blood glucose-related current signal in the fused feature data transmitted by the signal processing module to generate a continuous waveform sequence of time signal intensity; Set the sinking trigger threshold: When the signal strength of M consecutive sampling points decreases by more than the previous sampling point, and the signal strength of the Mth sampling point decreases by more than the first preset percentage compared with the average value of the stable segment before the sinking, mark the time corresponding to the Mth sampling point as the sinking start time point t0. When the cumulative rise of the signal strength at N consecutive sampling points exceeds the second preset percentage of the total dip amplitude, and the signal fluctuation amplitude at the subsequent K sampling points is less than the third percentage, the time corresponding to the Nth sampling point is marked as the dip end time point t1, and the dip time point range is [t0, t1]. Signal data within a preset time period before the sinking start time t0 is extracted, and the baseline signal data before the sinking is calculated using a weighted average algorithm. The sampling points closer to t0 have higher weights, and the weight coefficients are distributed according to a linear increasing law.

[0020] The beneficial effects are: by using high-frequency sampling and multi-condition triggering mechanisms, a precise benchmark is provided for depression compensation, avoiding over-correction or under-correction caused by baseline distortion, and improving the authenticity of the corrected signal.

[0021] A method for indentation compensation for a CGM sensor includes the following steps: Step 1: Collect the subject's blood glucose-related current signal at night through the signal acquisition module, and transmit the blood glucose-related current signal to the signal processing module in real time; Step 2: The signal processing module receives the blood glucose-related current signal, extracts its time-domain features and physiological correlation features to obtain fused feature data containing the rate of blood glucose change, and then transmits the fused feature data to the signal identification module. Step 3: The signal identification module receives the fused feature data and analyzes it based on the preset two-dimensional judgment rules: first, it determines whether it is a short-term depression through the time domain dimension rules, and then verifies whether it is a real hypoglycemia through the physiological dimension rules, and obtains the identification result and the corresponding target signal segment, which is then transmitted to the depression compensation module. Step 4: The depression compensation module receives the identification result and the target signal segment. If the identification result is a non-physiological depression, the baseline signal data before the depression in the target signal segment is extracted. Combined with the overall blood glucose signal dynamic trend data, the target signal segment is corrected using a linear extrapolation algorithm to obtain the compensated current signal. If it is a true hypoglycemia, the original target signal segment is retained as the compensated current signal, and the compensated current signal is transmitted to the blood glucose estimation module. Step 5: The blood glucose estimation module receives the compensated current signal, processes it based on the preset blood glucose value calculation model, and obtains and outputs accurate nighttime blood glucose monitoring results.

[0022] Furthermore, in step 1, the signal acquisition module also collects the subject's physiological characteristic data at night and transmits the physiological characteristic data to the signal processing module in real time; in step 2, after receiving the physiological characteristic data, the signal processing module performs feature quantization processing on it, and integrates the quantized physiological characteristic data with the fused characteristic data to generate joint feature data containing temporal features, physiological correlation features, and quantified physiological features, which is then transmitted to the signal identification module; in step 3, the signal identification module analyzes the joint feature data based on a preset two-dimensional judgment rule: based on the temporal dimension rule determining it as short-term depression, it uses the quantified physiological characteristic data to corroborate whether it is a non-physiological depression.

[0023] The basic principles and beneficial effects of the scheme are as follows: By generating fused feature data through feature extraction, multi-dimensional evidence is provided for subsequent identification, avoiding the limitations of single-signal judgment; a two-dimensional identification rule is adopted, first determining short-term depressions in the time domain and then physiologically verifying hypoglycemia, preventing signal confusion; differentiated processing is performed based on the identification results: baseline data is extracted and linearly extrapolated for non-physiological depressions, while the original signal is retained for true hypoglycemia, avoiding the risk of false positives and missed detections. This improves the accuracy of non-physiological depression correction, reduces the hypoglycemia false negative rate, and enhances the reliability and safety of nighttime monitoring. Attached Figure Description

[0024] Figure 1 This is a system structure block diagram of Embodiment 1 of a depression compensation system for a CGM sensor; Figure 2 This is a flowchart of Embodiment 8 of a depression compensation method for a CGM sensor. Detailed Implementation

[0025] The following detailed description illustrates the specific implementation method: Example 1 This embodiment discloses a depression compensation system for a CGM sensor, as shown in the attached figure. Figure 1As shown, the signal acquisition module uses an implantable CGM sensor that fits into the subcutaneous tissue of the subject's abdomen. It collects blood glucose-related current signals at night through the principle of electrochemical induction. The sampling frequency is set to 1Hz, which means that one set of current signal data is collected every second. After the acquisition is completed, it is transmitted to the signal processing module in real time via Bluetooth protocol.

[0026] The signal processing module uses a microcontroller. After receiving the blood glucose-related current signal transmitted from the signal acquisition module, it extracts its time-domain and physiological features. First, time-domain features are extracted using a sliding window algorithm (window size set to 5 seconds) to extract three time-domain features: signal drop amplitude, drop duration, and recovery trend. For example, if the current signal acquired at a certain moment drops sharply from 80nA to 50nA, the calculated drop amplitude is 30nA, the drop duration is 10 seconds, and the signal recovers to 75nA within the next 15 seconds, showing a rapid drop followed by a slow recovery. Then, physiological features are extracted. Based on a preset current signal-blood glucose concentration calibration curve, the current signal is converted into blood glucose concentration data through linear fitting, and the rate of blood glucose change is calculated using the difference formula. The blood glucose rate of change is calculated as v = (C2 - C1) / (t2 - t1), where C1 and C2 are the blood glucose concentrations at times t1 and t2, respectively. If the blood glucose concentration is 5.0 mmol / L at t1 = 0 seconds and 4.5 mmol / L at t2 = 10 seconds, then the blood glucose rate of change is calculated as v = -0.05 mmol / (L·s). Finally, fused feature data is generated by integrating the three extracted time-domain features with one physiological correlation feature (blood glucose rate of change) to generate fused feature data. The data format is decrease magnitude - duration - recovery trend - blood glucose rate of change, for example, "30nA - 10s - rapid decrease followed by slow recovery - 0.05 mmol / (L·s)", and is transmitted to the signal identification module through the interface.

[0027] The signal identification module uses an FPGA chip and pre-stores dual-dimensional judgment rules. After receiving and analyzing the fused feature data, the rules are as follows: Time domain dimension rules: The preset amplitude threshold is 1.5 times the maximum change value of the current signal corresponding to the normal blood glucose fluctuation of the subject in the past 3 days (assuming it is 20nA), i.e., 30nA; the duration of the decline is limited to 5-60 seconds; the signal recovery trend must meet the requirement that the signal recovers to more than 90% of the baseline value before the decline within 30 seconds after the decline. In this embodiment, the signal drop amplitude of 30 nA in the fused feature data is equal to the preset amplitude threshold, and the drop duration of 10 seconds is in the range of 5-60 seconds. However, the recovery trend is a rapid drop followed by a slow recovery, which does not meet the requirement of recovering to more than 90% of the baseline value within 30 seconds. Physiological dimension rule: The preset threshold for the rate of change of blood glucose in physiological hypoglycemia is -0.005 to -0.042 mmol / (L·s). In this embodiment, the rate of change of blood glucose is -0.05 mmol / (L·s), which exceeds the threshold range. Therefore, it is verified as a non-physiological depression. Identification result output: The identification result of "non-physiological depression" is generated, and the corresponding target signal segment is locked, that is, the 10-second current signal data from the start of the signal drop to the time of recovery and stabilization, which is transmitted to the depression compensation module.

[0028] The indentation compensation module uses an ARM Cortex-M7 processor. After receiving the identification results and the target signal segment, it performs the following operations: extracts the signal data (60 sets in total) within 60 seconds before the indentation in the target signal segment, calculates its arithmetic mean as the baseline data of the signal before the indentation, assuming that the average value of the 60 sets of data is 79.5nA, that is, the baseline data is 79.5nA; combines the overall blood glucose signal dynamic trend data (assuming that the overall trend is a slow decrease with a trend slope of -0.02nA / s), and uses a linear extrapolation algorithm to correct the target signal segment, and completes the signal correction for all moments within the target signal segment in sequence to obtain the compensated current signal; and transmits the compensated current signal to the blood glucose estimation module through the UART interface.

[0029] The specific steps for correction using the linear extrapolation algorithm are as follows: Select stable signal segments before and after the target signal segment as references. Assume the target segment is from time t3 to t7, with the preceding stable segment being t1-t2, where signal values ​​I1 = 10nA and I2 = 11nA, and the subsequent stable segment being t8-t9, where signal values ​​I8 = 15nA and I9 = 16nA. Calculate the linear trend slope: Based on the preceding stable segment, calculate the initial slope k1 = (I2-I1) / (t2-t1) = 1nA / time; based on the subsequent stable segment, calculate the recovery slope k2 = (I9-I8) / (t9-t8) = 1nA / time. Since both are consistent, the overall linear trend is determined to be a slope of 1nA / time. Based on the endpoint of the previous stationary segment (t2=2, I2=11nA) and the slope 1, the correction values ​​for each time step of the target segment are extrapolated sequentially: I3=11 + 1×(3-2)=12nA, I4=11 + 1×(4-2)=13nA, I5=11 + 1×(5-2)=14nA, I6=11 + 1×(6-2)=15nA I7 = 11 + 1 × (7-2) = 16nA. After correction, the compensated current signal is obtained, which makes the target segment and the preceding and following stable segments form a continuous linear trend, eliminating distortion.

[0030] The blood glucose estimation module pre-stores a blood glucose value calculation model, which is a multiple linear regression model obtained by fitting a large amount of clinical data: G = a×I + b×v + c, where G is the blood glucose concentration in mmol / L, I is the compensated current signal in nA, v is the rate of change of blood glucose in mmol / (L·s), and a, b, and c are model coefficients, with a in mmol / (L·nA), b in s, and c in mmol / L. Assume a=0.05, b=-2, and c=3. After receiving the compensated current signal (e.g., 79.4 nA), it is substituted into the model to calculate: G=0.05×79.4 + (-2)×(-0.05) + 3=7.07mmol / L, obtaining an accurate nighttime blood glucose monitoring result, which is then output in real time on the LCD display.

[0031] Example 2 The only difference from Example 1 is the addition of physiological feature data acquisition and fusion processing, as detailed below: In addition to acquiring blood glucose-related current signals, the signal acquisition module also integrates a heart rate sensor and a triaxial accelerometer to simultaneously acquire the subject's physiological characteristic data at night. The heart rate sensor has a sampling frequency of 1Hz and acquires the subject's nighttime heart rate data, ranging from 50 to 100 beats per minute. The triaxial accelerometer has a sampling frequency of 1Hz and acquires the subject's nighttime body movement data. The acceleration vector magnitude is calculated to reflect the body movement amplitude, ranging from 0 to 2g. After acquisition, the physiological characteristic data (heart rate + body movement amplitude) and blood glucose-related current signals are synchronously transmitted to the signal processing module.

[0032] After receiving physiological characteristic data, the signal processing module performs the following processing: It normalizes the heart rate data using the formula H = (H - Hmin) / (Hmax - Hmin), where H is the original heart rate. For example, if Hmin = 50 beats / minute, Hmax = 100 beats / minute, and the original heart rate H = 75 beats / minute, then the heart rate quantization value H = 0.5. It also performs threshold quantization on body movement data, setting the body movement amplitude threshold to 0.05g. When the body movement amplitude is ≥ 0.05g, the body movement quantization value M = 1 (indicating significant body movement); when the body movement amplitude is < 0.05g, M = 0 (indicating...). (If there is no obvious body movement), and the body movement amplitude is 0.8g at a certain moment, then M=1; The quantified physiological characteristic data (H=0.5, M=1) are correlated and integrated with the fusion characteristic data (30nA-10s-rapid decrease followed by slow recovery--0.05mmol / (L·s)) generated in Example 1 to generate joint characteristic data, in the format of decrease amplitude-duration-recovery trend-blood glucose change rate-heart rate quantification value-body movement quantification value, that is, 30nA-10s-rapid decrease followed by slow recovery--0.05mmol / (L·s)-0.5-1, which is transmitted to the signal identification module.

[0033] The signal identification module performs a two-dimensional judgment based on joint feature data: the time domain dimension judgment is the same as in Example 1, and it is determined to be a short-term depression. Physiological features are used to corroborate this judgment. The quantified physiological feature data is analyzed. Since the body movement quantification value M=1 (there is obvious body movement) and the heart rate quantification value H=0.5 (no significant heart rate abnormality), it is consistent with the physiological features of non-physiological depression caused by changes in sleep posture. Therefore, it is confirmed to be a non-physiological depression. The identification result of non-physiological depression and the corresponding target signal segment are generated and transmitted to the depression compensation module.

[0034] Example 3 The only difference from Example 2 is the addition of a time synchronization judgment mechanism for multi-source data in the signal processing module, as follows: When collecting blood glucose-related current signals and physiological characteristic data, the collection timestamp is recorded synchronously, and the blood glucose-related current signals, physiological characteristic data, and their corresponding timestamps T1 and T2 are synchronously transmitted to the signal processing module. After receiving the data, the signal processing module extracts the differences between timestamps T1 and T2, for example, Δt = 50ms; the preset synchronization threshold is 100ms. Since Δt = 50ms ≤ 100ms, it is determined to be synchronous data, and the signal processing module transmits the integrated joint feature data to the signal identification module. If Δt = 150ms > 100ms, it is determined to be asynchronous data, and the signal processing module only transmits the fused feature data extracted based on the blood glucose-related current signal to the signal identification module, avoiding misjudgment caused by asynchronous data.

[0035] Example 4 The only difference from Example 3 is the optimized asynchronous data processing strategy and the addition of a time calibration mechanism, as follows: After acquiring the acquisition timestamp T1 of the blood glucose-related current signal and the acquisition timestamp T2 of the physiological characteristic data, the time delay ΔT = T1 - T2 is calculated. If ΔT = 0, it is determined to be synchronous data without delay, and the joint feature data is directly integrated and transmitted to the signal identification module. If ΔT > 0, it is determined to be asynchronous data with a time delay, and the time calibration process is initiated. The historical data cache library (caching historical data of the last 1 minute) is called, and the two types of signals are calibrated based on the changing trend of the historical data: The corrected current signal at the intermediate time T_mid = (T1 + T2) / 2 is generated using the linear extrapolation method. Assuming that in the historical data, the current signal 5 seconds before T1 is 82nA, the current signal at T1 is 80nA, and the historical changing trend is a linear decrease with a slope k = -0.4nA / s, then the corrected current signal I_cal at T_mid is 79.96nA. The corrected physiological characteristic data at T_mid is generated using the moving average method. Assuming the historical data shows a heart rate of 70 beats / minute 3 seconds before T2, 71 beats / minute 2 seconds before T2, 72 beats / minute 1 second before T2, and 73 beats / minute at T2, and setting the sliding window size to 4, the corrected heart rate H_cal = 71.5 beats / minute is used to calibrate the body movement data in the same way. The calibrated current signal and physiological characteristic data are then integrated into joint feature data and transmitted to the signal discrimination module.

[0036] Example 5 The only difference from Example 4 is the establishment of a dynamic time synchronization mechanism to improve the accuracy of time delay determination. Specifically, the following steps are taken: a baseline timestamp T1' is recorded synchronously when collecting blood glucose-related current signals, and a baseline timestamp T2' is recorded synchronously when collecting physiological characteristic data; the initial time delay ΔT' = T1' - T2' is calculated; the preset time interval is 10 seconds, and a set of ΔT' values ​​of synchronized data is extracted every 10 seconds to construct a delay monitoring sequence. For example, the ΔT' values ​​of 10 consecutively extracted sets of data are 20ms, 22ms, 19ms, 21ms, 23ms, 20ms, 18ms, 22ms, 19ms, 21ms, 23ms, 20ms, 18ms, 22ms, 23ms, 20ms, 22ms, 23ms, 20ms, 22ms, 23ms, 20ms, 22ms, 23ms, 20ms, 22ms, 23ms, 20ms, 22ms, 23ms, 20ms, 22ms, 23ms, 24ms, 25 ... ms, 22ms, 21ms, 19ms; The mean delay μ and standard deviation σ are calculated using a sliding window algorithm (window size is 10 groups). If the calculated μ=20.5ms and σ=1.5ms, then the normal time delay range is [μ-2σ,μ+2σ]=[20.5-3,20.5+3]=[17.5ms,23.5ms]; If the extracted ΔT'=25ms at a certain moment exceeds the normal range, it is determined to be an abnormal time delay, triggering a synchronization calibration command to control the signal acquisition module to resynchronize the reference time source and correct the timestamp recording deviation.

[0037] Example 6 The only difference from Example 5 is that the dual-dimensional judgment rule is fused and optimized, and quantitative fusion is achieved by feature weight allocation, as follows: The calculation formula for the preset fusion judgment value S is: S=α×A+β×B+γ×C+δ×D, where α, β, γ, and δ are weight coefficients, which are determined based on historical data statistics and specific performance objectives (such as error minimization and stability maximization), satisfying α+β+γ+δ=1; A is a time-domain feature parameter: the ratio of the signal drop amplitude to a preset amplitude threshold, where the preset amplitude threshold is 30nA. If the signal drop amplitude at a certain moment is 30nA, then A=30 / 30=1; B is a duration parameter: the ratio of the drop duration to 60 minutes. If the drop duration is 10 minutes, then... B = 10 / 60 ≈ 0.167. If the duration exceeds 60 minutes, then B = 1 (typically, the default sampling frequency of mainstream continuous glucose monitoring (CGM) systems is to record blood glucose values ​​once every 5 minutes, i.e., 12 data points per hour and 288 data points per day); C is the recovery trend parameter: the ratio of the recovery rate within 30 minutes after the dip to the dip rate. If the dip rate is 30 nA and the recovery rate within 30 minutes is 25 nA, then C = 25 / 30 ≈ 0.833; D is the physiological characteristic parameter: the ratio of the rate of change of blood glucose to the upper limit of the physiological hypoglycemia rate threshold (0.042 mmol / (L·s)). If the rate of change of blood glucose is 0.05 mmol / (L·s), then D = 0.05 / 0.042 ≈ 1.19; Assuming a fusion judgment value of 0.8, α=0.3, β=0.2, γ=0.2, and δ=0.3, a depression of S≤0.8 is considered non-physiological, while a depression of S>0.8 is considered true hypoglycemia. The fusion judgment value S is calculated to be 0.857. Since S>0.8, it is initially determined to be true hypoglycemia. This is further validated by physiological characteristic data. If the physiological characteristic data shows a slow increase in heart rate and small body movement amplitude, consistent with the physiological response of true hypoglycemia, then it is ultimately determined to be true hypoglycemia, and the original target signal segment is retained.

[0038] Example 7 The only difference from Example 6 is the refinement of the sinking time point determination and baseline data extraction process, specifically as follows: High-frequency sampling analysis is performed on the blood glucose-related current signal, with the sampling frequency set to 10Hz, generating a continuous waveform sequence of time and signal intensity. The sinking trigger threshold parameters are set as follows: M=3, preset instantaneous change threshold=2nA, first preset percentage=15%, N=5, second preset percentage=80%, K=10, and third percentage=5%. The waveform sequence is iterated. When the signal intensity of three consecutive sampling points decreases by more than the preset instantaneous change threshold 2nA compared to the previous sampling point, and the signal intensity of the third sampling point decreases by 14.5% compared to the average value of the stable segment before the sinking, which is close to 15%, the condition is met, and this is marked as the sinking start time point t0. Continuing to traverse the waveform sequence, when the cumulative rise of 5 consecutive sampling points exceeds the second preset percentage, and the signal fluctuation amplitude of the subsequent 10 sampling points is less than 5%, 1800ms is marked as the end time point t1 of the dip, with the dip time point range being [t0, t1]. Signal data (300 sampling points) within 30 seconds before the start time point t0 of the dip is extracted. A weighted average algorithm is used to calculate the baseline data, with weight coefficients distributed according to a linear increasing law. The closer the sampling point is to t0, the higher its weight. The weight coefficient of the sampling point in the nearest second is 3, and the weight coefficient of the sampling point in the farthest second is 1. The weight coefficients of the intermediate sampling points increase linearly. Assuming the baseline data obtained through weighted average calculation is 79.2 nA.

[0039] Example 8 This embodiment discloses a depression compensation method for CGM sensors, as shown in the attached figure. Figure 2 As shown, the specific steps are as follows: Blood glucose-related current signals of the subject at night are collected using an implanted CGM sensor at a sampling frequency of 1Hz. The collected current signals are transmitted to the microcontroller in real time via Bluetooth. After receiving the blood glucose-related current signals, temporal and physiological features are extracted. The signal decrease amplitude, duration, and recovery trend are extracted using a 5-second sliding window. Blood glucose concentration is obtained through a calibration curve of "current signal - blood glucose concentration," and the rate of blood glucose change is calculated using differential calculation. Fusion feature data is generated in the format of "decline amplitude - duration - recovery trend - rate of blood glucose change" and transmitted to the FPGA chip. The fusion feature data is analyzed based on a two-dimensional judgment rule. From the temporal dimension, it is determined whether the signal is a short-term dip, and from the physiological dimension, it is verified whether it is true hypoglycemia. The identification result and the corresponding target signal segment are obtained, for example, determining that "the 10-second signal from 23:05:00 to 23:05:10 is a non-physiological dip," and transmitted to the ARM processor. After receiving the identification results and the target signal segment, if it is a non-physiological depression, the signal data of the 60 seconds before t0 is extracted, the baseline data is calculated, and combined with the overall dynamic trend of blood glucose signal, a linear extrapolation algorithm is used to correct the target signal segment to obtain the compensated current signal. If it is a true hypoglycemia, the original target signal segment is retained as the compensated current signal. The compensated current signal is then transmitted to the blood glucose estimation module. Upon receiving the compensated current signal, it is processed based on the preset blood glucose value calculation model G=a×I+b×v+c (a=0.05, b=-2, c=3). For example, if the compensated current signal is 79.4nA and the blood glucose change rate is -0.05mmol / (L·s), the model is substituted to obtain G=7.07mmol / L, generating an accurate nighttime blood glucose monitoring result and outputting it through the LCD display.

[0040] Example 9 The only difference from Example 8 is the addition of steps for collecting, processing, and fusing physiological characteristic data, as detailed below: While collecting blood glucose-related current signals, the MAX30102 heart rate sensor and MPU6050 accelerometer simultaneously collected the subject's nighttime heart rate and body movement data at a sampling frequency of 1Hz. The blood glucose-related current signals and physiological characteristic data (heart rate + body movement) were transmitted to the signal processing module in real time. Upon receiving the data, the signal processing module extracted temporal and physiological correlation features from the blood glucose-related current signals to generate fused feature data. It then quantized the physiological characteristic data, normalizing the heart rate data and thresholding the body movement data. The quantized physiological characteristic data was then correlated and integrated with the fused feature data to generate joint feature data containing temporal features, physiological correlation features, and quantized physiological features, which was transmitted to the signal identification module. The signal identification module analyzed the joint feature data based on a two-dimensional judgment rule: first, it determined whether the depression was short-term using the temporal dimension rule, and then corroborated this by the quantified physiological characteristic data. The identification result and the corresponding target signal segment were generated and transmitted to the depression compensation module. Subsequent signal compensation and blood glucose estimation followed the same steps as in Example 8, ultimately yielding accurate nighttime blood glucose monitoring results, which were then output.

[0041] The above are merely embodiments of the present invention. The invention is not limited to the fields covered by these embodiments. Commonly known structures and characteristics in the solutions are not described in detail here. Those skilled in the art are aware of all common technical knowledge in the field prior to the application date or priority date, are able to access all existing technologies in that field, and have the ability to apply conventional experimental methods prior to that date. Those skilled in the art can, under the guidance of this application, improve and implement this solution in combination with their own capabilities. Some typical known structures or methods should not be obstacles for those skilled in the art to implement this application. It should be noted that those skilled in the art can make several modifications and improvements without departing from the structure of the present invention. These should also be considered within the scope of protection of the present invention, and will not affect the effectiveness of the implementation of the present invention or the practicality of the patent. The scope of protection claimed in this application should be determined by the content of its claims, and the specific embodiments described in the specification can be used to interpret the content of the claims.

Claims

1. A recess compensation system for a CGM sensor, characterized by, The method comprises the following steps: a signal acquisition module is used to acquire blood glucose related current signals of a subject at night, and the blood glucose related current signals are transmitted to a signal processing module in real time; the signal processing module is used to receive the blood glucose related current signals, extract time domain features and physiological correlation features of the blood glucose related current signals, obtain fusion feature data of blood glucose change rate, and transmit the fusion feature data to a signal discrimination module; the signal discrimination module is used to receive the fusion feature data, analyze the fusion feature data based on a preset double-dimension judgment rule, determine whether it is short-term decline through a time domain dimension rule, and verify whether it is true hypoglycemia through a physiological dimension rule, to obtain a discrimination result and a corresponding target signal segment, and transmit the discrimination result and the corresponding target signal segment to a depression compensation module; the depression compensation module is used to receive the discrimination result and the target signal segment, extract signal baseline data before the depression in the target signal segment if the discrimination result is non-physiological depression, combine overall blood glucose signal dynamic trend data, correct the target signal segment by using a linear extrapolation algorithm, and obtain a compensated current signal; if it is true hypoglycemia, the original target signal segment is retained as the compensated current signal, and the compensated current signal is transmitted to a blood glucose estimation module; the blood glucose estimation module is used to receive the compensated current signal, process the compensated current signal based on a preset blood glucose value calculation model, and obtain an accurate night blood glucose monitoring result and output the result.

2. The recess compensation system for a CGM sensor of claim 1, wherein, The signal acquisition module is also used to acquire physiological feature data of the subject at night, and the physiological feature data are transmitted to the signal processing module in real time; after receiving the physiological feature data, the signal processing module is used to perform feature quantization processing on the physiological feature data, correlate and integrate the quantized physiological feature data and the fusion feature data, generate joint feature data containing time domain features, physiological correlation features and quantized physiological features, and transmit the joint feature data to the signal discrimination module; the signal discrimination module is used to analyze the joint feature data based on the preset double-dimension judgment rule: on the basis of the determination of short-term decline through the time domain dimension rule, whether it is non-physiological depression is verified according to the quantized physiological feature data.

3. The recess compensation system for a CGM sensor of claim 2, wherein, The signal processing module is also used to receive time stamps of the acquisition of the blood glucose related current signals and the physiological feature data, and calculate a time difference Δt between the two: if the time difference Δt is less than or equal to a preset synchronization threshold, it is determined that the blood glucose related current signals and the physiological feature data are synchronous data collected at the same time, and the signal processing module transmits the integrated joint feature data to the signal discrimination module; if the time difference Δt is greater than the preset synchronization threshold, it is determined that the two are non-synchronous data, and the signal processing module only transmits the fusion feature data extracted based on the blood glucose related current signals to the signal discrimination module.

4. The recess compensation system for a CGM sensor of claim 3, wherein, The signal processing module is also used to obtain an acquisition time stamp T1 of the blood glucose related current signals and an acquisition time stamp T2 of the physiological feature data transmitted by the signal acquisition module, and calculate a time delay ΔT = |T1-T2| between the two; if ΔT = 0, it is determined that the data are synchronous data without delay, the joint feature data are generated and transmitted to the signal discrimination module; If ΔT>0, it is determined that there is time-delayed asynchronous data, the signal processing module calls the historical data cache library, and time calibration is performed on the two types of signals respectively based on the change trend of the historical data: For blood glucose related current signals, a linear extrapolation method is used to generate a corrected current signal at (T1+T2) / 2; For physiological characteristic data, a moving average method is used to generate corrected physiological characteristic data at (T1+T2) / 2; After correction, time-matched synchronous data is obtained, which is integrated into joint feature data and transmitted to the signal discrimination module.

5. The recess compensation system for a CGM sensor of claim 4, wherein, The signal processing module establishes a dynamic time synchronization mechanism through a reference time source, the signal acquisition module records a reference time stamp T1' when collecting blood glucose related current signals, and a reference time stamp T2' when collecting physiological characteristic data; After receiving T1' and T2', the signal processing module calculates an initial time delay ΔT'=|T1'-T2'|, and simultaneously starts dynamic monitoring: every preset time interval extracts a group of ΔT' values of synchronous data, constructs a delay monitoring sequence, and calculates the delay mean μ and standard deviation σ through a sliding window algorithm; When ΔT' is within the interval [μ-2σ, μ+2σ], it is determined that the time delay is within the normal range; when ΔT' exceeds the interval, it is determined that the time delay is abnormal, triggering a synchronization calibration instruction.

6. The recess compensation system for a CGM sensor of claim 5, wherein, The dual-dimension judgment rule realizes the merging and fusion of the time domain dimension and the physiological dimension through feature weight distribution: Set a fusion judgment value S, the calculation formula is S = α×A+β×B +γ×C +δ×D, wherein A is a time domain feature parameter, B is a duration parameter, C is a recovery trend parameter, and D is a physiological feature parameter, and α, β, γ, and δ are weight coefficients; when S≤a preset fusion judgment value, it is determined that it is not a physiological decline; when S>the preset fusion judgment value, it is determined that it is a true hypoglycemia.

7. The recess compensation system for a CGM sensor of claim 6, wherein, When the recess compensation module extracts the signal baseline data before the decline in the target signal segment, the decline time point is determined in advance: The blood glucose related current signals in the fusion feature data transmitted by the signal processing module are analyzed by high frequency sampling to generate a continuous waveform sequence of time signal intensity; Set a decline trigger threshold: when the signal intensity of the continuous M sampling points decreases by more than a preset instantaneous change threshold compared with the previous sampling point, and the signal intensity of the Mth sampling point decreases by more than a first preset percentage compared with the average value of the stable segment before the decline, mark the time corresponding to the Mth sampling point as the decline starting time point t0; Track the waveform sequence to the recovery amplitude of the continuous N sampling points, which accumulates more than a second preset percentage of the total decline amplitude, and the signal fluctuation amplitude of the subsequent K sampling points is less than a third percentage, mark the time corresponding to the Nth sampling point as the decline ending time point t1, and the decline time point range is [t0, t1]; Extract the signal data in a preset period before the decline starting time point t0, and calculate the signal baseline data before the decline by using a weighted average algorithm, the closer the sampling point to t0, the higher the weight, and the weight coefficients are distributed according to a linear increasing rule.

8. A recess compensation method for a CGM sensor, characterized by, The steps include: Step 1: collecting blood glucose related current signals of the subject at night through a signal acquisition module, and transmitting the blood glucose related current signals to a signal processing module in real time; Step 2: the signal processing module receives the blood glucose related current signals, extracts time domain features and physiological correlation features, obtains fusion feature data containing blood glucose change rate, and transmits the fusion feature data to a signal discrimination module; Step 3: the signal discrimination module receives the fusion feature data, analyzes the fusion feature data based on a preset two-dimensional judgment rule: first determines whether it is a short-term decline through a time domain dimension rule, and then verifies whether it is a true hypoglycemia through a physiological dimension rule, obtains a discrimination result and a corresponding target signal segment, and transmits the discrimination result and the corresponding target signal segment to a depression compensation module; Step 4: the depression compensation module receives the discrimination result and the target signal segment, extracts the signal baseline data before the depression in the target signal segment if the discrimination result is a non-physiological depression, combines the overall blood glucose signal dynamic trend data, uses a linear extrapolation algorithm to correct the target signal segment, and obtains a compensated current signal; if it is a true hypoglycemia, the original target signal segment is retained as the compensated current signal, and the compensated current signal is transmitted to a blood glucose estimation module; Step 5: the blood glucose estimation module receives the compensated current signal, processes the compensated current signal based on a preset blood glucose value calculation model, obtains an accurate night blood glucose monitoring result, and outputs the accurate night blood glucose monitoring result.

9. The depression compensation method for a CGM sensor according to claim 8, characterized in that, the signal acquisition module in step 1 also collects physiological feature data of the subject at night, and transmits the physiological feature data to the signal processing module in real time; the signal processing module receives the physiological feature data, performs feature quantization processing on the physiological feature data, integrates the quantized physiological feature data with the fusion feature data, generates joint feature data containing time domain features, physiological correlation features and quantized physiological features, and transmits the joint feature data to the signal discrimination module; the signal discrimination module analyzes the joint feature data based on a preset two-dimensional judgment rule: on the basis of the determination of short-term decline through the time domain dimension rule, whether it is a non-physiological depression is verified according to the quantized physiological feature data.

Images