Hepatitis patient personalized nursing scheme generation system

Abstract

The invention discloses a personalized nursing scheme generation system for hepatitis patients. The personalized nursing scheme generation system comprises a signal acquisition and processing unit, a dynamic decision-making unit and a feedback optimization unit, the signal acquisition and processing unit adopts a flexible wearable hepatic region monitoring patch and a bio-impedance spectroscopy sensor array, acquires data of electrophysiological signals and metabolic characteristics of a patient in real time, and eliminates medical environmental noise from the data; the dynamic decision-making unit is used for cooperatively transmitting data by integrating a fractal antenna array and a medical special LoRaWAN protocol, establishing a three-layer cascade decision-making model, integrating patient medical history, gene detection data and clinical guidelines, and generating a personalized nursing scheme; the feedback optimization unit carries out dynamic optimization processing on the personalized nursing scheme; the multi-modal data of electrophysiological signals, metabolic characteristics, medical history and gene detection are fused to generate a highly personalized nursing scheme, so that patient requirements can be accurately matched, and the accuracy and pertinence of the nursing scheme are improved.

Description

Technical Field

[0001] This invention relates to a system for generating nursing care plans for hepatitis patients, and more particularly to a system for generating personalized nursing care plans for hepatitis patients. Background Technology

[0002] Hepatitis, as a common liver disease, has a complex course and significant individual differences. Patient care needs to consider liver function protection, drug metabolism management, and quality of life intervention. Current hepatitis care protocols are mainly based on clinical guidelines and the experience of healthcare professionals, focusing on monitoring liver function indicators (such as ALT and AST) and basic vital signs, and implementing interventions through standardized dietary recommendations, medication guidance, and regular follow-up examinations. However, these protocols are significantly inadequate in adapting to individual patient genetic differences (such as drug metabolism-related CYP450 genotypes), real-time fluctuations in physiological signals (such as changes in liver electrophysiological characteristics), and the risk of sudden onset of illness (such as a sudden increase in bilirubin during an acute attack).

[0003] To address these issues, existing technologies attempt to optimize nursing care plans through the following methods: introducing wearable devices to collect basic vital signs such as heart rate and respiratory rate, and combining this with static medical history data to generate preliminary intervention recommendations; employing traditional signal processing algorithms to denoise electrophysiological signals and extracting time-domain features such as mean and standard deviation to aid in judgment; and integrating clinical standards such as the "Guidelines for the Prevention and Treatment of Chronic Hepatitis B" based on rule engines to form an "indicator-plan" correspondence. These methods have improved the standardization of nursing care to some extent, but they still have not broken through the limitations of the "standardization framework."

[0004] The core shortcomings of the existing solution are: firstly, the data collection dimension is limited, covering only basic vital signs and lacking real-time dynamic monitoring of key physiological parameters such as liver metabolic characteristics and bilirubin metabolic spectrum, thus failing to capture subtle changes in liver function; secondly, the noise processing technology is outdated, and traditional filtering algorithms have limited ability to suppress noise such as motion artifacts and electromagnetic interference, resulting in a high feature extraction error rate and affecting the reliability of the input data of the decision model. Summary of the Invention

[0005] This invention overcomes the shortcomings of existing technologies and provides a system for generating personalized care plans for hepatitis patients.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is: a personalized care plan generation system for hepatitis patients, comprising: a signal acquisition and processing unit, a dynamic decision-making unit, and a feedback optimization unit;

[0007] The signal acquisition and processing unit uses a flexible wearable liver area monitoring patch and a bioimpedance spectrum sensor array to acquire real-time data on the patient's electrophysiological signals and metabolic characteristics, and uses an improved wavelet packet transform algorithm to eliminate medical environment noise.

[0008] The dynamic decision-making unit integrates a fractal antenna array and a medical-grade LoRaWAN protocol for collaborative data transmission, and establishes a three-layer cascaded decision-making model to integrate patient history, gene testing data, and clinical guidelines to generate personalized care plans.

[0009] The feedback optimization unit constructs a vital sign monitoring module based on a millimeter-wave radar array and an evaluation model for quantifying nursing effects, and dynamically optimizes the personalized nursing plan.

[0010] In a preferred embodiment of the present invention, the system further includes a time diversity controller, which alternately activates the bioimpedance spectroscopy sensor array and the millimeter-wave radar array according to a preset timing sequence; simultaneously, a motion compensation unit is established to correct the signal amplitude of the electrophysiological signal in real time through an accelerometer counter, wherein the compensation coefficient in the motion compensation unit is: In the formula, v represents the speed at which the electrophysiological signal moves.

[0011] It should be noted that, to address the electromagnetic coupling interference issue when the bioimpedance spectroscopy sensor array and the millimeter-wave radar array operate simultaneously, a time-division activation mechanism is employed to avoid signal crosstalk between sensors, ensuring independent and accurate acquisition of electrophysiological signals and vital sign data. Through time-division activation hardware design and cross-modal collaboration of dynamic compensation algorithms, the signal distortion problem of wearable devices in complex physiological environments is solved. This approach is more adaptable than traditional solutions that only handle fixed noise, and is particularly suitable for hepatitis patients requiring long-term dynamic monitoring.

[0012] In a preferred embodiment of the present invention, the specific steps of the signal acquisition and processing unit in eliminating medical environmental noise in the data include:

[0013] S100. Construct a noise baseline correction module based on a nonlinear threshold function, and introduce the data into the noise baseline correction module to perform baseline correction;

[0014] S200. The complex Morlet wavelet basis function is selected as the time-frequency joint analysis module to dynamically calculate the above-corrected data and obtain the optimal number of decomposition layers.

[0015] S300. Construct a noise-signal classification model, if If the value is , then it is determined to be medical environmental noise; where, Representing empirical coefficients, preferably ; Represents the wavelet coefficients of the k-th node in the j-th layer; This represents the maximum amplitude of all wavelet coefficients in the j-th layer; N represents the total number of dynamic decomposition nodes.

[0016] S400. The signal is reconstructed using an improved wavelet packet transform algorithm, and an amplitude compensation factor is added. The value of the reconstructed signal is obtained through the formula: The calculation yields the result; where, This represents the value of the reconstructed signal; This represents the value of the signal before reconstruction. This represents the energy difference before and after signal reconstruction; Represents a reference energy value, preferably .

[0017] In a preferred embodiment of the present invention, the three-layer cascaded decision model includes a signal feature layer, a data fusion layer, and a decision generation layer; the signal feature layer employs an improved TCN network, adding a perforated convolutional layer to the traditional temporal convolution; the data fusion layer is based on a Transformer module with a multi-head attention mechanism, setting gene data weight coefficients. The decision generation layer combines the Q-learning algorithm with a reinforcement learning optimizer to set the reward function. .

[0018] In a preferred embodiment of the present invention, the feedback optimization unit adjusts the operating frequency of the millimeter-wave radar array to 75-85 GHz, with a resolution of [missing information]. Based on the FMEA risk prediction algorithm, a risk threshold is preset; the assessment model is a weighted assessment system of liver function indicators and quality of life scores.

[0019] In a preferred embodiment of the present invention, the feedback optimization unit further includes a decision tracing module for displaying the influence weights of each data source on the nursing plan. The calculation formula for the influence weights of each data source by the decision tracing module is as follows:

[0020] ;

[0021] In the formula, This represents the influence weight of the i-th data source. This represents the influence weight of the i-th data source; This represents the sum of weight coefficients for all data sources across all dimensions; n represents the total number of data sources; and m represents the coefficient dimension of the total weight coefficients.

[0022] In a preferred embodiment of the present invention, the feedback optimization unit is further provided with a manual correction interface, which triggers the physician review process when the confidence level of the generated personalized solution is lower than a preset threshold.

[0023] In a preferred embodiment of the present invention, the dynamic decision-making unit integrates a federated learning framework to support distributed model training across multiple medical institutions; the bioimpedance spectrum sensor array is equipped with an adaptive calibration module to compensate for changes in skin contact impedance through a differential amplifier circuit.

[0024] In a preferred embodiment of the present invention, the feedback optimization unit establishes a dual-modal early warning mechanism, including abnormal detection of the sympathetic / parasympathetic balance index based on HRV and early warning of the dynamic growth threshold of ALT.

[0025] In a preferred embodiment of the present invention, the signal acquisition and processing unit further includes a photonic crystal fiber sensing module for non-invasively acquiring the spectral characteristics of bilirubin metabolism in blood; the dynamic decision-making unit adopts a transfer learning algorithm to achieve cross-disease nursing strategy recommendation through a pre-trained hepatobiliary disease knowledge graph.

[0026] This invention addresses the shortcomings of the prior art and has the following beneficial effects:

[0027] (1) By integrating multimodal data such as electrophysiological signals, metabolic characteristics, medical history, and gene testing, highly personalized nursing plans can be generated, thereby accurately matching patient needs and improving the accuracy and pertinence of nursing plans.

[0028] (2) By monitoring the patient’s vital signs and changes in condition in real time, the nursing plan can be dynamically adjusted, thereby enabling timely response to emergencies and improving nursing effectiveness.

[0029] (3) By making full use of patients’ multi-dimensional data resources, we can provide a scientific basis for the development of nursing plans and reduce interference from human factors. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a system flowchart of a preferred embodiment of the present invention. Detailed Implementation

[0032] 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.

[0033] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein. Therefore, the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0034] like Figure 1 As shown, the personalized care plan generation system for hepatitis patients includes: a signal acquisition and processing unit, a dynamic decision-making unit, and a feedback optimization unit.

[0035] The flexible wearable liver monitoring patch uses a medical-grade flexible PCB substrate and integrates an 8-channel differential electrode bioimpedance spectroscopy sensor array and a photonic crystal fiber optic sensing module. The former acquires the liver tissue impedance value through a 100~10kHz sweep frequency signal to reflect metabolic characteristics, while the latter acquires the bilirubin metabolism spectrum through an 850nm near-infrared light source transmitted through the skin, with a sampling rate of 20Hz.

[0036] The millimeter-wave radar array module is a 77GHz frequency-modulated continuous wave radar with a resolution of 0.3mm. It is integrated into the back of the wearable device and monitors vital signs such as respiratory rate and heart rate in real time. It is activated alternately with the bioimpedance sensor in a 100ms cycle through a time diversity controller, thereby reducing electromagnetic interference.

[0037] The motion compensation unit has a built-in triaxial accelerometer with a range of ±8g and a sampling rate of 100Hz. It monitors the patch displacement velocity v in real time and calculates the compensation coefficient β=1 / (1+0.05v²) online through a hardware multiplier to correct the amplitude of the electrophysiological signal in real time.

[0038] This solution integrates an 8-channel bioimpedance spectroscopy sensor array and a photonic crystal fiber optic sensing module into a flexible wearable liver region monitoring patch, enabling continuous acquisition of liver tissue metabolic characteristics and bilirubin metabolic spectra, overcoming the periodic limitations of traditional blood sampling tests. Combined with a 77GHz millimeter-wave radar and triaxial accelerometer integrated into the back, it simultaneously acquires respiratory rate, heart rate variability, and motion displacement velocity, constructing a real-time monitoring system encompassing electrophysiology, metabolism, and motion status, providing comprehensive data support for personalized care.

[0039] The signal processing engine is developed based on Python / PyTorch and deployed on edge computing nodes. It uses a nonlinear threshold function for piecewise polynomial fitting to correct DC drift in the raw electrophysiological signal, with a window length of 5 seconds.

[0040] The center frequency of the complex Morlet wavelet basis is 50Hz, and the bandwidth parameter is... =3, perform 5-level wavelet packet decomposition, and dynamically calculate the optimal decomposition level using the minimum entropy criterion.

[0041] According to the formula Noise is identified, where N is the number of nodes in the current layer. After setting the noise coefficients to zero, the signal is recovered using an improved wavelet packet reconstruction algorithm, and then... Perform amplitude compensation.

[0042] 32 features, including heart rate variability (HRV), impedance phase angle reflecting cell membrane integrity, and bilirubin absorption peak intensity, were extracted from the processed signal.

[0043] The signal feature layer is an improved TCN network, which contains three layers of perforated convolutions with dilation coefficients of 1, 2, and 4. It takes a 32-dimensional feature sequence as input and outputs a 50-dimensional depth feature vector.

[0044] The data fusion layer, based on eight head-attention Transformer modules, integrates structured data from patients' electronic medical records, HLA genotypes, CYP450 metabolic site mutation information, UpToDate standard nursing procedures, and gene data weighting coefficients. .

[0045] The decision generation layer is based on the Q-learning reinforcement learning engine, and the state space is defined as follows: Liver function classification, medication history, and real-time vital signs. The action space is Dietary recommendations, medication adjustments, and exercise interventions. reward function The optimal care plan is generated through the ε-greedy strategy.

[0046] The federated learning module supports collaborative training of data from multiple hospitals. It adopts a horizontal federated learning framework, where each institution trains model parameters locally and uploads gradients through a secure aggregation protocol. The global model is updated every 24 hours to protect patient privacy.

[0047] The dynamic decision-making unit integrates a federated learning framework, supporting distributed training across multiple medical institutions. Homomorphic encryption protects patient privacy and addresses the issue of insufficient data from a single institution. By combining a pre-trained hepatobiliary disease knowledge graph with transfer learning algorithms, it enables cross-disease nursing strategy recommendations for hepatitis B, hepatitis C, and other diseases, significantly expanding clinical application scenarios.

[0048] The effectiveness evaluation module establishes a weighted evaluation system, with liver function indicators having a weight of 0.6 and quality of life scores having a weight of 0.4. A comprehensive score is calculated daily, and exceeding the threshold triggers scheme optimization.

[0049] The risk warning module also features a dual-modal warning mechanism: HRV analysis triggers an autonomic nervous system imbalance warning, and when ALT growth exceeds 50 U / L / 24h, a liver injury warning is triggered, which notifies medical staff via SMS / pop-up window.

[0050] The decision tracing interface visualizes the influence weights of each data source, categorized by... calculate.

[0051] To further explain, the data flow interaction process is as follows:

[0052] The wearable device is configured to alternately activate the bioimpedance sensor and millimeter-wave radar at 100ms intervals to simultaneously collect electrophysiological signals, impedance spectra, spectral data and acceleration data.

[0053] The accelerometer is noise-removed by Kalman filtering, the instantaneous velocity v is calculated, and the electrophysiological signal is corrected point by point.

[0054] A second-order Butterworth low-pass filter with a cutoff frequency of 100Hz is used in the hardware circuit to filter out high-frequency noise and generate a raw data frame containing timestamps, sensor IDs, and raw signal values.

[0055] The raw data is uploaded to the edge node, and after baseline correction, wavelet packet decomposition and noise reduction, and signal reconstruction, a clean electrophysiological signal with a signal-to-noise ratio improved by ≥15dB is output.

[0056] The mean, standard deviation, peak-to-peak value, and time-domain characteristics of the electrophysiological signal, as well as the power-to-impedance spectrum from 5 to 30 Hz, are extracted. Principal component analysis is used to extract the top three principal components of the bilirubin metabolism spectrum, generating a standardized feature vector, which is then uploaded to the cloud via LoRaWAN.

[0057] The decision generation and solution output include: generating a patient state vector by combining real-time features, gender, age, disease course, complications, rs12979860 genotype from the electronic medical record, and dietary restrictions and medication dosage adjustment rules from the "Guidelines for the Prevention and Treatment of Chronic Hepatitis B" in the guidelines.

[0058] Then, temporal features are extracted through the signal feature layer, and the attention weights for each data modality are calculated using the data fusion layer and the Transformer algorithm, such as the weights for acute-phase gene data. =0.4, stable period =0.3. Finally, the decision generation layer outputs the optimal solution containing three core intervention measures based on the Q-learning optimizer.

[0059] The optimal approach for the three core interventions includes: each with parameters for frequency and intensity of implementation, such as "low-fat diet (daily fat intake <50g) + 3 light aerobic exercises per week + entecavir medication reminders".

[0060] The confidence level is verified through a manual correction interface. If the confidence level is below 80%, a physician review process is triggered. Finally, a PDF nursing plan is generated and synchronized to the patient's APP and the medical workstation.

[0061] Effectiveness evaluation and dynamic optimization: Respiratory rate and heart rate variability are output every 1 minute by millimeter-wave radar, combined with liver function test results, to calculate a weighted evaluation score. A three-level risk threshold (low / medium / high risk corresponding to different response mechanisms) is preset using the FMEA algorithm. For example, a medium-risk trigger will result in a minor adjustment of the protocol, while a high-risk trigger will result in an emergency warning, simultaneously notifying the attending physician.

[0062] The feedback optimization unit iterates the solution every 24 hours, adjusts the decision model parameters based on the latest evaluation results, and updates the edge node processing algorithm through OTA, forming a closed loop of "collection-decision-feedback".

[0063] Example 1

[0064] Daily care for patients with chronic hepatitis B: Activate the bioimpedance sensor every morning from 6:30 to 7:00 to detect the metabolic status of the liver area and generate breakfast suggestions based on the heart rate variability during the previous night's sleep.

[0065] The timing of nucleoside (acid) analogue administration is adjusted based on genetic data (CYP450 metabolites), and medication reminders are pushed through the APP.

[0066] At midday, monitor your resting heart rate using millimeter-wave radar. If it is >80 bpm and ALT is normal, it is recommended to take a 20-minute walk and use an accelerometer to monitor exercise intensity in real time to ensure your heart rate is ≤100 bpm.

[0067] Example 2

[0068] Acute hepatitis exacerbation warning: A sudden increase in bilirubin absorbance at night, exceeding the baseline value by 20%, coupled with an ALT increase >30 U / L / 6h, triggers a dual-modal warning. The decision generation layer automatically switches to the acute phase nursing protocol and displays "bilirubin test data contribution 72%, ALT increase contribution 28%" through the decision tracing interface. The protocol takes effect after confirmation by the attending physician, and the nursing record system is updated synchronously.

[0069] Furthermore, the bioimpedance sensor has a built-in differential amplifier circuit that is automatically calibrated by a standard impedance module before each use to compensate for changes in skin contact impedance, with an error of ≤5%.

[0070] The wearable device uses a flexible lithium battery with a capacity of 500mAh and features a low-power design, meaning a sleep current of <10μA. It has a battery life of 72 hours on a single charge and supports wireless charging.

[0071] LoRaWAN transmission uses CRC-16 checksum and triggers a retransmission mechanism when the packet loss rate is >5%; the cloud receiving module is equipped with a data buffer queue with a capacity of 1000 frames and a processing latency of ≤100ms.

[0072] Federated learning sets up a data quality screening process for participants, with ≥5 participating institutions in each round of training. The model is updated when the global accuracy is ≥90%, and the optimization effect is verified through A / B testing. The error rate of the control group is reduced by 15%.

[0073] Based on the preferred embodiments of the present invention described above, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A system for generating personalized care plans for hepatitis patients, characterized in that, include: Signal acquisition and processing unit, dynamic decision-making unit, and feedback optimization unit; The signal acquisition and processing unit uses a flexible wearable liver area monitoring patch and a bioimpedance spectrum sensor array to acquire real-time data on the patient's electrophysiological signals and metabolic characteristics, and uses an improved wavelet packet transform algorithm to eliminate medical environment noise. The dynamic decision-making unit integrates a fractal antenna array and a medical-grade LoRaWAN protocol for collaborative data transmission, and establishes a three-layer cascaded decision-making model to integrate patient history, gene testing data, and clinical guidelines to generate personalized care plans. The feedback optimization unit constructs a vital sign monitoring module based on a millimeter-wave radar array and an evaluation model for quantifying nursing effects, and dynamically optimizes the personalized nursing plan.

2. The personalized care plan generation system for hepatitis patients according to claim 1, characterized in that: The system also includes a time diversity controller, which alternately activates the bioimpedance spectroscopy sensor array and the millimeter-wave radar array according to a preset timing sequence; simultaneously, a motion compensation unit is established to correct the signal amplitude of the electrophysiological signal in real time through an accelerometer counter, wherein the compensation coefficient in the motion compensation unit is: In the formula, v represents the speed at which the electrophysiological signal moves.

3. The personalized care plan generation system for hepatitis patients according to claim 1, characterized in that: The specific steps for the signal acquisition and processing unit to eliminate medical environmental noise in the data include: S100. Construct a noise baseline correction module based on a nonlinear threshold function, and introduce the data into the noise baseline correction module to perform baseline correction; S200. The complex Morlet wavelet basis function is selected as the time-frequency joint analysis module to dynamically calculate the above-corrected data and obtain the optimal number of decomposition layers. S300. Construct a noise-signal classification model, if If the value is , then it is determined to be medical environmental noise; where, Represents the empirical coefficient; Represents the wavelet coefficients of the k-th node in the j-th layer; This represents the maximum amplitude of all wavelet coefficients in the j-th layer; N represents the total number of dynamic decomposition nodes. S400. The signal is reconstructed using an improved wavelet packet transform algorithm, and an amplitude compensation factor is added. The value of the reconstructed signal is obtained through the formula: The calculation yields the result; where, This represents the value of the reconstructed signal; This represents the value of the signal before reconstruction. This represents the energy difference before and after signal reconstruction; This represents the baseline energy value.

4. The personalized care plan generation system for hepatitis patients according to claim 1, characterized in that: The three-layer cascaded decision model includes a signal feature layer, a data fusion layer, and a decision generation layer. The signal feature layer uses an improved TCN network, adding a perforated convolutional layer to the traditional temporal convolution. The data fusion layer is based on a Transformer module with a multi-head attention mechanism, setting gene data weight coefficients. The decision generation layer combines the Q-learning algorithm with a reinforcement learning optimizer to set the reward function. .

5. The personalized care plan generation system for hepatitis patients according to claim 1, characterized in that: The feedback optimization unit adjusts the operating frequency of the millimeter-wave radar array to 75-85 GHz, with a resolution of [missing information]. ; Based on the FMEA risk prediction algorithm, a preset risk threshold is set; the assessment model is a weighted assessment system of liver function indicators and quality of life scores.

6. The personalized care plan generation system for hepatitis patients according to claim 5, characterized in that: The feedback optimization unit also includes a decision tracing module, used to display the influence weight of each data source on the nursing plan. The calculation formula for the influence weight of each data source by the decision tracing module is as follows: ； In the formula, This represents the influence weight of the i-th data source. This represents the influence weight of the i-th data source; This represents the sum of weight coefficients for all data sources across all dimensions; n represents the total number of data sources; and m represents the coefficient dimension of the total weight coefficients.

7. The personalized care plan generation system for hepatitis patients according to claim 6, characterized in that: The feedback optimization unit is also equipped with a manual correction interface. When the confidence level of the generated personalized plan is lower than a preset threshold, the physician review process is triggered.

8. The personalized care plan generation system for hepatitis patients according to claim 1, characterized in that: The dynamic decision-making unit integrates a federated learning framework to support distributed model training across multiple medical institutions; the bioimpedance spectrum sensor array is equipped with an adaptive calibration module that compensates for changes in skin contact impedance through a differential amplifier circuit.

9. The personalized care plan generation system for hepatitis patients according to claim 1, characterized in that: The feedback optimization unit establishes a dual-modal early warning mechanism, including abnormal detection of the sympathetic / parasympathetic nerve balance index based on HRV and early warning of the dynamic growth threshold of ALT.

10. The personalized care plan generation system for hepatitis patients according to claim 1, characterized in that: The signal acquisition and processing unit also includes a photonic crystal fiber sensing module, used to non-invasively acquire the spectral characteristics of bilirubin metabolism in blood. The dynamic decision-making unit employs a transfer learning algorithm to recommend cross-disease nursing strategies through a pre-trained knowledge graph of hepatobiliary diseases.

Images