Intelligent drainage metering early warning device and method based on multi-sensor fusion and LSTM prediction model

Abstract

The invention relates to the technical field of medical monitoring equipment and artificial intelligence crossing, in particular to an intelligent drainage metering early warning device and method based on multi-sensor fusion and an LSTM prediction model. The device comprises a drainage liquid collecting container and a four-dimensional sensor array, wherein the four-dimensional sensor array comprises a weighing sensor, an optical sensor, a capacitive sensor and a micro pressure sensor; according to the electronic drainage meter, various intelligent algorithms are fused, so that the electronic drainage meter realizes high-precision metering, advanced intelligent early warning, self-adaptive clinical adaptation, safe and efficient data transmission and rapid abnormity correction, and the safety and efficiency of postoperative drainage monitoring are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring equipment and artificial intelligence, and in particular to an intelligent drainage measurement and early warning device and method based on multi-sensor fusion and LSTM prediction model. Background Technology

[0002] Existing electronic drainage measurement devices suffer from the following technical bottlenecks: single sensor fusion algorithms have limited adaptability and are difficult to cover the diverse characteristics of fluids such as blood, bile, and cerebrospinal fluid; traditional prediction models do not pay enough attention to critical clinical periods, limiting their early warning capabilities; anomaly detection can only identify known patterns, resulting in a high rate of missed detection for rare complications; data transmission poses a risk of privacy breaches and is difficult to adapt to collaborative optimization of multi-center clinical data.

[0003] To address this, the present invention integrates cutting-edge algorithms such as federated learning, deep reinforcement learning, and variational autoencoders with four-dimensional sensing hardware to construct an intelligent system covering the entire "perception-analysis-decision-transmission" chain, thus overcoming the aforementioned technical limitations. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide an intelligent diversion metering and early warning device and method based on multi-sensor fusion and LSTM prediction model, so as to solve the technical problems existing in the prior art.

[0005] According to a first aspect of the present invention, an intelligent diversion metering and early warning device based on multi-sensor fusion and LSTM prediction model is provided, the device comprising:

[0006] The drainage fluid collection container has a superhydrophobic anti-wall coating on its inner wall; The four-dimensional sensor array includes a weighing sensor, an optical sensor, a capacitive sensor, and a miniature pressure sensor; wherein the weighing sensor has an accuracy of ±0.1g, and the miniature pressure sensor is used to monitor the fluid pressure inside the drainage tube. The microprocessor integrates an ARM Cortex-M7 core and an edge computing acceleration unit. The microprocessor is configured to run a multi-sensor federated fusion algorithm, an attention-enhanced LSTM prediction algorithm, a VAE-CNN joint anomaly detection algorithm, a wavelet-federated encrypted transmission algorithm, and a reinforcement learning adaptive threshold optimization algorithm. The intelligent alarm includes a directional speaker and a three-color indicator light, and is configured to output a three-level response signal according to the level of abnormality. The interface module supports Bluetooth 5.0, Wi-Fi 6, and USB 3.0 data transmission and is configured for encrypted data interaction with the hospital information system. The drainage fluid collection container is connected to the four-dimensional sensor array; the four-dimensional sensor array is connected to the microprocessor; the microprocessor is connected to the intelligent alarm and the interface module respectively; the intelligent alarm is connected to the interface module.

[0007] Furthermore, the four-dimensional sensor array is configured to synchronously acquire data every 0.2 seconds, the optical sensor is used to detect the refractive index of the drainage fluid to distinguish the liquid type, and the capacitive sensor is used to compensate for the interference of ambient temperature on the measurement.

[0008] Furthermore, the multi-sensor federated fusion algorithm in the microprocessor is configured to: receive real-time data from a four-dimensional sensor array, dynamically adjust the weights of each sensor based on a federated learning model and a recursive least squares algorithm, and output the fused drainage fluid volume and flow rate, with a measurement error ≤ ±0.3%.

[0009] Furthermore, the three-level response signals of the intelligent alarm include: a blue warning signal, a yellow warning signal, and a red alarm signal, corresponding to suspected abnormality, unconfirmed abnormality, and confirmed abnormality, respectively.

[0010] Furthermore, it also includes an intelligent negative pressure control module, which is linked with the microprocessor and the micro pressure sensor and is configured to dynamically adjust the negative pressure value according to the pressure inside the drainage tube, with an adjustment accuracy of ±2mmHg.

[0011] According to a second aspect of the present invention, a smart diversion metering and early warning method based on multi-sensor fusion and LSTM prediction model is applied to the smart diversion metering and early warning device based on multi-sensor fusion and LSTM prediction model described in any one of the above embodiments, the method comprising: The drainage fluid is collected using a drainage fluid collection container, and the weight, refractive index, dielectric constant, and pressure data inside the drainage tube of the drainage fluid are synchronously collected every preset time period using a four-dimensional sensor array. Based on the weight, refractive index, dielectric constant, and pressure data within the drainage tube of the drainage fluid, through... The microprocessor runs a multi-sensor federated fusion algorithm to dynamically adjust the weights of each sensor based on the type of drainage fluid, and calculates the real-time drainage volume and flow rate. Based on the real-time drainage volume and flow rate results, the drainage flow rate trend in the future period is predicted by the attention-enhanced LSTM model. Combined with the VAE-CNN joint algorithm, abnormal patterns are identified, triggering the corresponding level response of the intelligent alarm to obtain relevant drainage data. The drainage data is compressed by wavelet transform and homomorphically encrypted, and then transmitted to the hospital information system through the interface module.

[0012] Furthermore, based on the weight, refractive index, dielectric constant, and pressure data within the drainage tube of the drainage fluid, a multi-sensor federated fusion algorithm is run via a microprocessor to dynamically adjust the weights of each sensor based on the drainage fluid type, and to calculate the real-time drainage volume and flow rate, including: The weights of each sensor are initialized based on a pre-trained global weight model in the cloud, and then fine-tuned locally by combining real-time collected data on the refractive index and pressure of the drainage fluid, and the real-time drainage volume and flow rate are calculated. The weights satisfy w1 + w2 + w3 + w4 = 1; where w1 w4 represents the weights of the weighing, optical, capacitive, and pressure sensors, respectively.

[0013] Furthermore, the prediction of the traffic flow rate trend over a future period using the attention-enhanced LSTM model includes: The first attention weight was assigned to the drainage flow rate time series in the first postoperative time period using an attention-enhanced LSTM model, and the second attention weight was assigned to the drainage flow rate time series in the second postoperative time period. The prediction results are corrected based on the first attention weight, the second attention weight, and the pressure data from the micro pressure sensor.

[0014] Furthermore, the VAE-CNN joint algorithm identifies abnormal patterns and triggers corresponding levels of response from the intelligent alarm, including: The reconstruction error of the prediction result is calculated by variational autoencoder. When the reconstruction error is greater than the first threshold, it is marked as a suspected outlier. High-frequency features of the suspected outliers are extracted using a convolutional neural network, and then matched with a complication feature library to classify the outlier types. Based on the anomaly type classification, the corresponding level of response of the intelligent alarm is triggered.

[0015] Furthermore, the step of performing wavelet transform compression and homomorphic encryption on the drainage data, and transmitting it to the hospital information system through the interface module, includes: Wavelet transform was used to compress the drainage data to obtain the first compressed data; The first compressed data is encrypted using a homomorphic encryption algorithm, and the encrypted data is transmitted to the hospital information system through the interface module.

[0016] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: 1. Breakthrough in accuracy: The federated fusion algorithm reduces the measurement error to ≤±0.3%, which is 40% lower than that of traditional electronic devices, making it suitable for neonatal micro-drainage (<5ml / h) scenarios; 2. Early warning capability: The attention-enhanced LSTM, combined with stress signals, can provide early warning of abnormalities 15-20 minutes in advance, which is 33% longer than existing technologies, thus buying time for intervention of complications; 3. Comprehensive anomaly detection: The VAE-CNN joint algorithm achieves an unknown anomaly detection rate of ≥95%, solving the pain point of traditional algorithms "missing rare complications"; 4. Clinical adaptability: The reinforcement learning threshold model reduces the false alarm rate by another 25% (from the original 40%), making it suitable for multiple departments, special populations (children / elderly), and complex postoperative stages; 5. Data security and efficiency: Wavelet-federated encryption reduces data volume by 50% and transmission latency to <1s, while meeting medical privacy protection requirements; 6. Academic value: Constructing an intelligent paradigm for medical devices based on "multimodal sensing-federated learning-reinforced decision-making" to support the construction of postoperative rehabilitation assessment models under the ERAS concept.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0019] Figure 1 This is a schematic diagram illustrating the composition of an intelligent diversion metering and early warning device based on multi-sensor fusion and LSTM prediction model, according to an exemplary embodiment. Figure 2 This is a schematic diagram illustrating the process of an intelligent diversion metering and early warning method based on multi-sensor fusion and LSTM prediction model, according to an exemplary embodiment. Detailed Implementation

[0020] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0021] Example 1 Please see Figure 1 , Figure 1 This is a schematic diagram illustrating the composition of an intelligent diversion metering and early warning device based on multi-sensor fusion and LSTM prediction model according to an exemplary embodiment. The device includes: The drainage fluid collection container 10 has a superhydrophobic anti-fouling coating on its inner wall, suitable for 12 types of fluids such as blood, bile, and cerebrospinal fluid, with a residual amount of <0.1ml; The four-dimensional sensor array 20 includes a weighing sensor, an optical sensor, a capacitive sensor, and a miniature pressure sensor; wherein, the weighing sensor has an accuracy of ±0.1g, and the miniature pressure sensor is used to monitor the fluid pressure in the drainage tube, achieving synchronous sampling at the 0.2s level; Weighing sensors, optical sensors, capacitive sensors, and miniature pressure sensors are integrated into a four-dimensional sensor array 20. The signal outputs of each sensor are aggregated within the array to form a collaborative acquisition sensor network.

[0022] Specifically, in the four-dimensional sensor array 20, the weighing sensor is located directly below and attached to the bottom of the drainage fluid collection container 10, the optical sensors (transmitter and receiver) are symmetrically installed on both sides of the transparent sidewall in the middle section of the container, the capacitive sensor is attached to the outer wall of the container in the form of an electrode sheet, and the miniature pressure sensor is connected in series in the section of the drainage tube near the container inlet; the signal outputs of the four sensors are combined and connected to the microprocessor 30 through a unified data bus to form a four-dimensional synchronous acquisition network of "weight-optical-capacitive-pressure" to transmit real-time data to the microprocessor 30.

[0023] The microprocessor 30 integrates an ARM Cortex-M7 core and an edge computing acceleration unit. The microprocessor 30 is configured to run a multi-sensor federated fusion algorithm, an attention-enhanced LSTM prediction algorithm, a VAE-CNN joint anomaly detection algorithm, a wavelet-federated encrypted transmission algorithm, and a reinforcement learning adaptive threshold optimization algorithm. The intelligent alarm 40 includes a directional speaker and a tri-color indicator light, and is configured to output a three-level response signal according to the level of abnormality. Specifically, the intelligent alarm 40 is a three-level response device based on the level of abnormality (blue warning: suspected abnormality; yellow warning: pending confirmation; red alarm: confirmed abnormality), including a directional speaker (to reduce patient interference) and a three-color indicator light; The interface module 50 supports Bluetooth 5.0, Wi-Fi 6 and USB 3.0 data transmission, and is configured to perform encrypted data interaction with the hospital information system; The drainage fluid collection container 10 is connected to the four-dimensional sensor array 20; the four-dimensional sensor array 20 is connected to the microprocessor 30; the microprocessor 30 is connected to the intelligent alarm 40 and the interface module 50 respectively.

[0024] Furthermore, it also includes an intelligent negative pressure control module, which is linked with the microprocessor 30 and the micro pressure sensor and is configured to dynamically adjust the negative pressure value according to the pressure inside the drainage tube, with an adjustment accuracy of ±2mmHg.

[0025] Furthermore, the four-dimensional sensor array 20 is configured to synchronously acquire data every 0.2s, the optical sensor is used to detect the refractive index of the drainage fluid to distinguish the liquid type, and the capacitive sensor is used to compensate for the interference of ambient temperature on the measurement.

[0026] Furthermore, the multi-sensor federated fusion algorithm in the microprocessor 30 is configured as follows: receiving real-time data from the four-dimensional sensor array 20, dynamically adjusting the weights of each sensor based on the federated learning model and the recursive least squares algorithm, and outputting the fused drainage fluid volume and flow rate, with a measurement error ≤ ±0.3%.

[0027] Furthermore, the three-level response signals of the intelligent alarm 40 include: a blue warning signal, a yellow warning signal, and a red alarm signal, which correspond to suspected abnormality, unconfirmed abnormality, and confirmed abnormality, respectively.

[0028] Furthermore, it also includes an intelligent negative pressure control module, which is linked with the microprocessor 30 and the micro pressure sensor and is configured to dynamically adjust the negative pressure value according to the pressure inside the drainage tube, with an adjustment accuracy of ±2mmHg.

[0029] For specific implementation, please refer to the following examples: In the neonatal intensive care unit (NICU), a premature infant weighing 2.5 kg underwent closed thoracic drainage for congenital pneumothorax. Precise monitoring of pleural effusion drainage volume (normal range 0-30 ml / h) was necessary to prevent excessively rapid drainage from causing mediastinal shift or insufficient drainage from worsening the pneumothorax. Therefore, the "intelligent drainage measurement and early warning device based on multi-sensor fusion and LSTM prediction model" described in this invention was used for real-time monitoring.

[0030] During the device initialization phase, a 50ml soft collection container adapted for neonatal micro-drainage is prepared. Its inner wall has a superhydrophobic anti-fouling coating (contact angle 155°) to ensure that the residual amount of pleural effusion (containing a small amount of blood) is <0.1ml. The container inlet is connected to the infant's chest drainage port via a sterile drainage tube. The four-dimensional sensor array 20 is deployed as follows: a weighing sensor (accuracy ±0.1g) is installed at the bottom of the container to detect weight changes; the optical sensor transmitter and receiver are symmetrically positioned in the transparent window in the middle of the container to detect refractive index; the capacitive sensor electrode is attached to the outer wall of the container to compensate for room temperature fluctuations (24-26℃); and a miniature pressure sensor is connected in series in the drainage tube near the container to monitor the pressure inside the tube (normal range -5 to -10cmH2O). After medical staff input patient information "neonatal, thoracic surgery, 0 hours post-closed thoracic drainage" via USB interface, the microprocessor 30 automatically completes initialization: the reinforcement learning adaptive threshold optimization algorithm sets the abnormal upper limit to 30ml / h (40% lower than that for adults), and relaxes the threshold to 25ml / h for the high-risk period 6 hours post-surgery; the federated learning global weight model is loaded, and the initial weight allocation for "pleural effusion with a small amount of blood" is 0.55 for weighing, 0.25 for optics, 0.15 for capacitance, and 0.05 for pressure; at the same time, self-calibration is initiated, and the signal from the empty container is collected to establish a measurement baseline and eliminate sensor zero drift.

[0031] During real-time monitoring for 1-3 hours post-surgery, the sensor array synchronously collected data every 0.2 seconds: in the first hour, the weighing sensor detected a weight increase from 0g to 12g, the optical sensor detected a refractive index of 1.338 (containing a small number of red blood cells), the capacitance sensor output a dielectric constant of 2.5 (compensated for room temperature effects +0.02ml), and the pressure sensor monitored the pressure inside the tube, which remained stable at -6cmH2O. The microprocessor 30 ran a multi-sensor federated fusion algorithm. Because the optical sensor detected "blood components," the weighing weight was increased to 0.58 and the optical weight was increased to 0.27, resulting in a calculated real-time flow rate of 12ml / h (measurement error ±0.2ml, meeting the ±0.3% accuracy requirement). Meanwhile, the attention-enhanced LSTM model predicts the flow rate for the next 20 minutes based on the flow rate sequence of the past 30 minutes [8ml / h, 10ml / h, 12ml / h]. Due to the high-risk period after surgery, the time attention weight is set to 0.7. Combined with the pressure stability signal (no risk of tube blockage), the predicted flow rate is 14ml / h (<threshold 25ml / h). The display screen refreshes in real time with "current flow rate 12ml / h, cumulative volume 12ml, pressure -6cmH2O" and plots a gentle trend curve.

[0032] Four hours after the operation, the patient's body position changed, which temporarily increased the patency of the drainage tube. The sensor array collected abnormal data: within 10 minutes, the weighing sensor detected that the weight increased from 48g to 78g (corresponding to a volume of 12ml → 78ml, flow rate of 180ml / h), the optical sensor detected that the refractive index was still 1.338, and the pressure sensor monitored that the pressure inside the tube decreased from -6cmH2O to -8cmH2O. The microprocessor 30 immediately initiates the anomaly identification process: the variational autoencoder (VAE) reconstructs the flow rate curve (12ml / h→180ml / h), and the reconstruction error reaches 4.2σ (>3σ threshold), marking it as "suspected anomaly" and triggering a blue warning from the intelligent alarm 40; the convolutional neural network (CNN) extracts the feature of "sudden increase in flow rate slope of 168ml / h² + brief drop in pressure", which matches the feature database of "rapid drainage" complications with a 98% match rate. After confirming the anomaly type, the intelligent alarm 40 switches to red alarm mode - the red light stays on and the directional speaker emits a low-frequency alarm, while simultaneously pushing alarm information to the nurse station through the interface module 50. At the same time, the microprocessor 30 outputs the "target pressure -6cmH2O" command to the intelligent negative pressure control module, and the module reduces the pressure in the tube to -6cmH2O within 30 seconds, and the flow rate gradually decreases to 25ml / h (<threshold 30ml / h).

[0033] During the data processing and transmission phase, the microprocessor 30 performs wavelet transform compression (compression ratio 20:1, 5MB→250KB) on the 4-hour monitoring data (including 14,400 sets of raw sensor data, 240 sets of flow velocity calculation values, and 1 abnormal alarm record), and then encrypts it using AES-256 to generate an encrypted data packet. After being transmitted to the NICU central monitoring system via Wi-Fi 6, the system generates a "Postoperative Drainage Report 4 Hours" (including a cumulative volume of 78ml and an abnormal timestamp of 14:30) in encrypted mode, and feeds back a reinforcement learning reward signal (Q value +0.9) to the device. Based on this, the microprocessor 30 increases the warning sensitivity of the "neonatal drainage" scenario by 2%.

[0034] In this scenario, the device's measurement error is ±0.2ml (<±0.3%), the abnormal response time is 1.2 seconds, and the intelligent negative pressure adjustment takes effect within 30 seconds, fully demonstrating high-precision measurement, advanced early warning, and clinical adaptability, effectively ensuring the safety of newborns after surgery.

[0035] Please see Figure 2 , Figure 2 This is a schematic flowchart illustrating an intelligent diversion metering and early warning method based on multi-sensor fusion and LSTM prediction model according to an exemplary embodiment. The method includes: S1. Collect drainage fluid using a drainage fluid collection container, and synchronously collect data on the weight, refractive index, dielectric constant, and pressure inside the drainage tube of the drainage fluid every preset time period using a four-dimensional sensor array; S2. Based on the weight, refractive index, dielectric constant and pressure data inside the drainage tube of the drainage fluid, a multi-sensor federated fusion algorithm is run by a microprocessor to dynamically adjust the weight of each sensor based on the type of drainage fluid and calculate the real-time drainage volume and flow rate. S3. Based on the real-time drainage volume and flow rate results, the drainage flow rate trend in the future period is predicted by the attention-enhanced LSTM model, and the abnormal patterns are identified by the VAE-CNN joint algorithm to trigger the corresponding level response of the smart alarm and obtain relevant drainage data. S4. Perform wavelet transform compression and homomorphic encryption on the drainage data, and transmit it to the hospital information system through the interface module.

[0036] Specifically, based on the weight, refractive index, dielectric constant, and pressure data within the drainage tube of the drainage fluid, a multi-sensor federated fusion algorithm is run via a microprocessor to dynamically adjust the weights of each sensor based on the drainage fluid type, and to calculate the real-time drainage volume and flow rate, including: The weights of each sensor are initialized based on a pre-trained global weight model in the cloud, and then fine-tuned locally by combining real-time collected data on the refractive index and pressure of the drainage fluid, and the real-time drainage volume and flow rate are calculated. The weights satisfy w1 + w2 + w3 + w4 = 1; where w1 w4 represents the weights of the weighing, optical, capacitive, and pressure sensors, respectively.

[0037] Specifically, the method of predicting the trend of traffic flow rate over a future period using an attention-enhanced LSTM model includes: The first attention weight was assigned to the drainage flow rate time series in the first postoperative time period using an attention-enhanced LSTM model, and the second attention weight was assigned to the drainage flow rate time series in the second postoperative time period. The prediction results are corrected based on the first attention weight, the second attention weight, and the pressure data from the micro pressure sensor.

[0038] Specifically, the VAE-CNN joint algorithm identifies abnormal patterns and triggers the corresponding level of response of the intelligent alarm, including: The reconstruction error of the prediction result is calculated by variational autoencoder. When the reconstruction error is greater than the first threshold, it is marked as a suspected outlier. High-frequency features of the suspected outliers are extracted using a convolutional neural network, and then matched with a complication feature library to classify the outlier types. Based on the anomaly type classification, the corresponding level of response of the intelligent alarm is triggered.

[0039] Specifically, the process of performing wavelet transform compression and homomorphic encryption on the drainage data, and then transmitting it to the hospital information system through the interface module, includes: Wavelet transform was used to compress the drainage data to obtain the first compressed data; The first compressed data is encrypted using a homomorphic encryption algorithm, and the encrypted data is transmitted to the hospital information system through the interface module.

[0040] In a practical application scenario involving pelvic drainage monitoring after gynecological laparoscopic surgery, a 35-year-old female patient underwent laparoscopic cystectomy for ovarian cyst removal. Postoperatively, monitoring of pelvic drainage fluid (primarily wound exudate) was required. A normal flow rate of ≤80 ml / h was mandated; a rate consistently exceeding this indicated potential postoperative bleeding, while a rate below 5 ml / h suggested possible tube blockage. The method of this invention operates fully automatically in this scenario. First, a 150 ml soft drainage fluid collection container is connected to the patient's pelvic drainage port. A four-dimensional sensor array synchronously collects data every 0.2 seconds according to a preset program. A weighing sensor detects changes in the total weight of the container and the liquid; an optical sensor detects the refractive index of the drainage fluid to distinguish between exudate and blood; a capacitive sensor detects the dielectric constant of the liquid to compensate for the influence of room temperature (25°C); and a miniature pressure sensor monitors the pressure within the drainage tube.

[0041] The microprocessor invokes a pre-trained global weight model in the cloud to initialize weights for the "pelvic effusion" scenario, assigning weights w1=0.5 to the weighing sensor, w2=0.3 to the optical sensor, w3=0.15 to the capacitive sensor, and w4=0.05 to the pressure sensor, satisfying w1+w2+w3+w4=1. Two hours post-surgery, the optical sensor detected a refractive index of 1.335 for the drainage fluid (typical characteristics of effusion), and the pressure sensor showed a stable pressure of -9cmH2O. The microprocessor fine-tunes the weights based on real-time data, maintaining the initial values ​​due to the stable fluid properties. Combining this with the weight changes detected by the weighing sensor over the two hours, the real-time flow rate is calculated to be 60ml / h, with an error of ±0.2ml.

[0042] The first 6 hours post-surgery are considered a high-risk period. The attention-enhanced LSTM model assigned a first attention weight of 0.7 to the flow rate sequence over the past 30 minutes, and combined with pressure sensor data, predicted a flow rate of 61 ml / h for the next 15 minutes, without triggering an alarm. Eight hours post-surgery, a slight abnormality appeared: the weighing sensor detected a weight increase of only 3 ml within one hour, and the pressure sensor showed a pressure rise to -4 cmH2O. The variational autoencoder calculated a reconstruction error of 3.2σ (>3σ first threshold), marking it as "suspected abnormality." The convolutional neural network extracted features and matched them with a 91% match rate in the "partial blockage of the drainage tube" feature library. After confirming the abnormality type, the intelligent alarm triggered a yellow warning.

[0043] The microprocessor performs wavelet transform compression on the 8-hour monitoring data, achieving a compression ratio of 20:1. The compressed data is then encrypted using a homomorphic encryption algorithm, and the encrypted data packet is transmitted to the nurse station terminal via Bluetooth. The system generates a "Postoperative Drainage Report 8 Hours" in encrypted mode. This method involves real-time data acquisition every 0.2 seconds, with a flow rate calculation error of <±0.3%, an anomaly detection response time of <2 seconds, and a 95% reduction in data transmission volume. It meets the accuracy requirements of clinical monitoring while reducing operational complexity through standardized procedures, making it suitable for routine postoperative monitoring scenarios in primary hospitals.

[0044] Specifically, this invention significantly improves the accuracy of drainage measurement through a multi-sensor federated fusion algorithm, effectively eliminating interference factors such as fluid properties and environment; it combines an attention-enhanced LSTM model with a VAE-CNN joint algorithm to achieve early warning of abnormal trends, with high recognition rates for both known and unknown abnormalities, significantly reducing false alarm rates; it features reinforcement learning adaptive threshold adjustment, which can dynamically adjust the warning threshold according to department, patient characteristics, and postoperative stage, adapting to multiple clinical scenarios; the fully automated operation reduces the workload of medical staff, and efficient compression and encryption technologies ensure secure data transmission and seamless integration into hospital systems; the intelligent negative pressure control module can quickly correct abnormal drainage states, ultimately achieving high-precision measurement, advanced intelligent early warning, strong clinical adaptability, and in-depth data utilization, significantly improving the safety and efficiency of postoperative drainage monitoring.

[0045] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0046] It should be noted that in the description of this invention, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this invention, unless otherwise stated, "a plurality of" means at least two.

[0047] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0048] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0049] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0050] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0051] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0052] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0053] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An intelligent diversion metering and early warning device based on multi-sensor fusion and LSTM prediction model, characterized in that, The device includes: The drainage fluid collection container has a superhydrophobic anti-wall coating on its inner wall; The four-dimensional sensor array includes a weighing sensor, an optical sensor, a capacitive sensor, and a miniature pressure sensor; wherein the weighing sensor has an accuracy of ±0.1g, and the miniature pressure sensor is used to monitor the fluid pressure inside the drainage tube. The microprocessor integrates an ARM Cortex-M7 core and an edge computing acceleration unit. The microprocessor is configured to run a multi-sensor federated fusion algorithm, an attention-enhanced LSTM prediction algorithm, a VAE-CNN joint anomaly detection algorithm, a wavelet-federated encrypted transmission algorithm, and a reinforcement learning adaptive threshold optimization algorithm. The intelligent alarm includes a directional speaker and a three-color indicator light, and is configured to output a three-level response signal according to the level of abnormality. The interface module supports Bluetooth 5.0, Wi-Fi 6, and USB 3.0 data transmission and is configured for encrypted data interaction with the hospital information system. The drainage fluid collection container is connected to the four-dimensional sensor array; the four-dimensional sensor array is connected to the microprocessor; the microprocessor is connected to the intelligent alarm and the interface module respectively; the intelligent alarm is connected to the interface module.

2. The apparatus according to claim 1, characterized in that, The four-dimensional sensor array is configured to synchronously acquire data every 0.2 seconds. The optical sensor is used to detect the refractive index of the drainage fluid to distinguish the liquid type, and the capacitive sensor is used to compensate for the interference of ambient temperature on the measurement.

3. The apparatus according to claim 1, characterized in that, The multi-sensor federated fusion algorithm in the microprocessor is configured to: receive real-time data from a four-dimensional sensor array, dynamically adjust the weights of each sensor based on a federated learning model and a recursive least squares algorithm, and output the fused drainage fluid volume and flow rate with a measurement error ≤ ±0.3%.

4. The apparatus according to claim 1, characterized in that, The intelligent alarm device has three levels of response signals: a blue warning signal, a yellow warning signal, and a red alarm signal, which correspond to suspected abnormality, abnormality pending confirmation, and confirmed abnormality, respectively.

5. The apparatus according to claim 1, characterized in that, It also includes an intelligent negative pressure control module, which is linked with the microprocessor and the micro pressure sensor and is configured to dynamically adjust the negative pressure value according to the pressure inside the drainage tube, with an adjustment accuracy of ±2mmHg.

6. A smart diversion metering and early warning method based on multi-sensor fusion and LSTM prediction model, applied to the smart diversion metering and early warning device based on multi-sensor fusion and LSTM prediction model as described in any one of claims 1-5, characterized in that, The method includes: The drainage fluid is collected using a drainage fluid collection container, and the weight, refractive index, dielectric constant, and pressure data inside the drainage tube of the drainage fluid are synchronously collected every preset time period using a four-dimensional sensor array. Based on the weight, refractive index, dielectric constant, and pressure data within the drainage tube of the drainage fluid, through... The microprocessor runs a multi-sensor federated fusion algorithm to dynamically adjust the weights of each sensor based on the type of drainage fluid, and calculates the real-time drainage volume and flow rate. Based on the real-time drainage volume and flow rate results, the drainage flow rate trend in the future period is predicted by the attention-enhanced LSTM model. Combined with the VAE-CNN joint algorithm, abnormal patterns are identified, triggering the corresponding level response of the smart alarm to obtain relevant drainage data. The drainage data is compressed using wavelet transform and homomorphically encrypted, and then transmitted to the hospital information system through the interface module.

7. The method according to claim 6, characterized in that, Based on the weight, refractive index, dielectric constant, and pressure data within the drainage tube of the drainage fluid, a multi-sensor federated fusion algorithm is run via a microprocessor to dynamically adjust the weights of each sensor based on the drainage fluid type, calculating the real-time drainage volume and flow rate, including: The weights of each sensor are initialized based on a pre-trained global weight model in the cloud, and then fine-tuned locally by combining real-time collected data on the refractive index and pressure of the drainage fluid, and the real-time drainage volume and flow rate are calculated. The weights satisfy w1 + w2 + w3 + w4 = 1; where w1 w4 represents the weights of the weighing, optical, capacitive, and pressure sensors, respectively.

8. The method according to claim 6, characterized in that, The method of predicting the flow rate trend in the future using an attention-enhanced LSTM model includes: The first attention weight was assigned to the drainage flow rate time series in the first postoperative time period using an attention-enhanced LSTM model, and the second attention weight was assigned to the drainage flow rate time series in the second postoperative time period. The prediction results are corrected based on the first attention weight, the second attention weight, and the pressure data from the micro pressure sensor.

9. The method according to claim 6, characterized in that, The VAE-CNN joint algorithm identifies abnormal patterns and triggers corresponding levels of response from the intelligent alarm, including: The reconstruction error of the prediction result is calculated by variational autoencoder. When the reconstruction error is greater than the first threshold, it is marked as a suspected outlier. High-frequency features of the suspected outliers are extracted using a convolutional neural network, and then matched with a complication feature library to classify the outlier types. Based on the anomaly type classification, the corresponding level of response of the intelligent alarm is triggered.

10. The method according to claim 6, characterized in that, The process of performing wavelet transform compression and homomorphic encryption on the drainage data, and then transmitting it to the hospital information system through the interface module, includes: Wavelet transform was used to compress the drainage data to obtain the first compressed data; The first compressed data is encrypted using a homomorphic encryption algorithm, and the encrypted data is transmitted to the hospital information system through the interface module.

Images