Excrement character recognition and drug delivery closed-loop control method and system

Abstract

The invention discloses an excrement character recognition and drug delivery closed-loop control method and system, and the system comprises an excrement collection and preprocessing module which is used for carrying out the automatic collection, preliminary physical and chemical character detection and sample preprocessing of the excrement of a patient; the multi-modal data analysis module is used for carrying out deep fusion analysis on the original multi-modal data output by the excrement collection and preprocessing module and carrying out diarrhea type diagnosis and severity evaluation; the drug delivery closed-loop control module is used for accurately and dynamically infusing drugs according to the diagnosis result of the multi-modal data analysis module; and the man-machine interaction and early warning module is used for monitoring and intervening the state of the system by medical staff and receiving important alarms. According to the invention, objective and continuous evaluation of excrement properties is realized, the drug dosage and scheme can be dynamically adjusted according to the properties, and the accuracy, individualization level and safety of diarrhea treatment of critical patients are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, and in particular to a method and system for closed-loop control of excrement characteristics identification and drug delivery. Background Technology

[0002] Patients in the Intensive Care Unit (ICU) often face numerous clinical challenges due to the complexity of their conditions and their physiological vulnerability. Among these challenges, gastrointestinal dysfunction, particularly diarrhea, is a common problem that can lead to serious consequences. Infections, the use of broad-spectrum antibiotics, enteral nutrition intolerance, and multiple organ dysfunction can all induce or exacerbate diarrhea symptoms. If diarrhea is not addressed promptly and accurately, patients are highly susceptible to a series of life-threatening complications, such as electrolyte imbalances, malnutrition, secondary infection spread, and pressure ulcer formation, which severely impact prognosis and prolong hospital stays. Therefore, effective monitoring of the characteristics of ICU patients' excrement and adjusting treatment plans accordingly is a core aspect of critical care nursing and treatment.

[0003] In traditional clinical practice, healthcare professionals primarily rely on manual observation to monitor patients' excrement. Specifically, nurses assess a patient's diarrhea by visually judging the color, consistency, and amount of excrement, and by recording the frequency of bowel movements. While this method is simple and intuitive, its inherent limitations are obvious. First, manual observation is highly subjective; different healthcare professionals may interpret the same excrement characteristics differently, leading to a lack of standardization and comparability in the assessment results. Second, due to the workload and shift schedules of nursing staff, observations are often intermittent, making it difficult to achieve real-time, continuous monitoring of excrement characteristics. This results in inherent defects such as poor timeliness and incomplete data recording.

[0004] Therefore, how to achieve objective, multi-dimensional, and high-precision identification of the characteristics of patients' excrement, and on this basis, build a closed-loop control system that can adjust drug dosage and treatment in real time according to the dynamic changes in the characteristics of excrement, has become a key challenge and a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention provides a method and system for closed-loop control of excrement characteristics identification and drug delivery to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A closed-loop control system for excrement characteristic identification and drug delivery includes:

[0008] The excrement collection and pretreatment module is used for automated collection of patient excrement, preliminary physical and chemical property detection, and sample pretreatment.

[0009] The multimodal data analysis module is communicatively connected to the excrement collection and preprocessing module. It is used to perform deep fusion analysis on the raw multimodal data output by the excrement collection and preprocessing module in order to objectively and quantitatively identify the qualitative characteristics of the excrement and to diagnose the type of diarrhea and assess its severity.

[0010] A drug delivery closed-loop control module is communicatively connected to the multimodal data analysis module and is used to accurately and dynamically infuse drugs based on the diagnostic results of the multimodal data analysis module.

[0011] The human-computer interaction and early warning module is communicatively connected to the excrement collection and pretreatment module, the multimodal data analysis module, and the drug delivery closed-loop control module. It provides an intuitive user interface to enable medical staff to monitor and intervene in the system status, receive important alarms, and support limited information sharing with patients' families.

[0012] As a further improvement to this technical solution: the excrement collection and pretreatment module includes an intelligent toilet structure, a dilution and mixing unit, and a sample diversion unit;

[0013] The intelligent toilet structure consists of a receiving cavity and an integrated sensor array.

[0014] The integrated sensor array is precisely mounted on the bottom and sidewalls of the receiving cavity, including:

[0015] A high-precision piezoresistive force sensor array, consisting of at least four independent piezoresistive sensors, is used to measure the total mass of excrement and its center of gravity distribution within the toilet bowl in real time.

[0016] The multispectral image acquisition unit includes a high-resolution CMOS image sensor and a tunable multispectral light source, which is used to achieve automatic focusing and capture the reflection images of excrement in different spectral bands under different excrement amounts and forms.

[0017] A high-precision non-contact viscosity sensor uses the acoustic scattering principle based on microwaves or ultrasound to derive the dynamic viscosity coefficient of excrement.

[0018] An integrated electrochemical sensor assembly includes a glass electrode pH sensor, a platinum electrode conductivity sensor, and an ion-selective electrode (ISE) array for detecting the pH, conductivity, and specific ion concentrations of excrement; and a temperature sensor for real-time monitoring of excrement temperature to correct for temperature drift of the integrated electrochemical sensor assembly.

[0019] As a further improvement to this technical solution: the high-precision piezoresistive force sensor array is composed of at least four independent silicon-based microelectromechanical systems (MEMS) piezoresistive sensors; the ion-selective electrode (ISE) array can simultaneously detect the concentrations of sodium ions, potassium ions and chloride ions.

[0020] As a further improvement to this technical solution: the dilution and mixing unit is located below the receiving cavity of the intelligent toilet structure, and includes two peristaltic pumps and a micro-mixing chamber with a volume of 20 mL. One peristaltic pump draws up the excrement sample and injects it into the micro-mixing chamber, while the other peristaltic pump injects sterile saline as a diluent at a specific ratio. The micro-mixing chamber is equipped with a magnetically driven micro-stirring paddle. The sample diversion unit is adjacent to the dilution and mixing unit and consists of a microfluidic channel and an array of electrically controlled micro-valves.

[0021] As a further improvement to this technical solution: the multimodal data analysis module includes an edge computing unit and a cloud-based intelligent processing platform;

[0022] The edge computing unit is responsible for receiving all raw data streams from the integrated sensor array and performing preprocessing operations such as high-frequency noise filtering, data standardization, image registration and correction.

[0023] The edge computing unit runs a set of deep learning models, including: an image feature extraction and texture analysis model based on a convolutional neural network (CNN), used to extract the color distribution, surface texture pattern, macroscopic morphology, and spatial features of particle distribution and size from multispectral images, and output a high-dimensional feature vector;

[0024] A sequence data analysis model based on a Long Short-Term Memory (LSTM) network is used to process time-series data from the high-precision piezoresistive force sensor array, the high-precision non-contact viscosity sensor, and the integrated electrochemical sensor group to extract the rheological properties, dynamic changes in chemical composition, and emptying patterns of excrement. A multimodal fusion model based on a Transformer architecture is also included, whose encoder consists of six stacked attention layers. This model deeply fuses the image feature vectors and the time-series feature vectors to generate a more discriminative joint feature representation. Finally, a fully connected network layer is used to classify and regress excrement characteristics to identify diarrhea types and predict diarrhea severity scores.

[0025] As a further improvement to this technical solution: the edge computing unit can complete all deep fusion analysis and output detailed diagnostic results and confidence levels within 5 seconds after data acquisition; the cloud intelligent processing platform and the edge computing unit perform real-time data synchronization and model updates through the encrypted WebSocket protocol; the cloud intelligent processing platform hosts a larger-scale clinical decision engine built on the deep reinforcement learning paradigm.

[0026] As a further improvement to this technical solution: the clinical decision engine includes:

[0027] Structured knowledge base, patient-personalized data interface, and decision agent based on deep reinforcement learning.

[0028] As a further improvement to this technical solution: the drug delivery closed-loop control module includes an intelligent infusion pump array and a feedback regulation controller.

[0029] As a further improvement to this technical solution: the human-computer interaction and early warning module includes a nurse station terminal, a mobile application, and an independent early warning system.

[0030] This invention proposes a closed-loop control method for excrement characteristics identification and drug delivery, comprising the following steps:

[0031] The excrement collection and pretreatment module automates the collection of patient excrement, performs preliminary physical and chemical property testing, and preprocesses the samples.

[0032] The multimodal data analysis module performs in-depth fusion analysis on the raw multimodal data output by the excrement collection and preprocessing module to objectively and quantitatively identify the qualitative characteristics of the excrement and to diagnose the type of diarrhea and assess its severity.

[0033] The drug delivery closed-loop control module delivers drugs accurately and dynamically based on the diagnostic results of the multimodal data analysis module;

[0034] The human-computer interaction and early warning module provides an intuitive user interface, enabling medical staff to monitor and intervene in the system status, receive important alarms, and support limited information sharing with patients' families.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] First, by introducing a multimodal sensor array (including high-precision pressure, multispectral imaging, non-contact viscosity, integrated electrochemistry, etc.) and combining it with dilution mixing and sample splitting pretreatment, an objective, refined, and multidimensional quantitative assessment of excrement properties is achieved, overcoming the subjectivity and intermittency of traditional manual observation and the shortcomings of existing automated equipment in qualitative property analysis.

[0037] Second, by using deep learning models based on CNN, LSTM and Transformer architectures to perform deep fusion analysis of multimodal data, it is possible to automatically and accurately identify the specific types of diarrhea and their potential pathophysiological mechanisms, and quantify their severity, which greatly improves the depth and accuracy of diagnosis and makes up for the lack of diagnostic depth in existing monitoring information.

[0038] Third, a clinical decision engine based on reinforcement learning was constructed. By combining individualized patient data, drug PK / PD models and a structured knowledge base, it can generate personalized and real-time optimized drug delivery plans, rather than relying on fixed empirical medical orders, thereby realizing intelligent and precise treatment plans.

[0039] Fourth, the human-computer interaction and early warning module provides intuitive real-time information display, convenient manual intervention interface and multi-level early warning mechanism, which enhances medical staff's control over the patient's condition and treatment process, shortens response time and further protects the life safety of critically ill patients.

[0040] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0041] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0042] Figure 1 This is a schematic flowchart of a closed-loop control method for excrement property identification and drug delivery proposed in this invention.

[0043] Figure 2 This is a schematic diagram of a closed-loop control system for excrement property identification and drug delivery proposed in this invention. Detailed Implementation

[0044] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically in the following paragraphs by way of example with reference to the accompanying drawings. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.

[0045] This invention proposes a closed-loop control system for excrement characteristics identification and drug delivery, comprising four core modules: an excrement collection and pretreatment module, a multimodal data analysis module, a drug delivery closed-loop control module, and a human-computer interaction and early warning module. Deployed in the intensive care unit, the system enables real-time, non-invasive, or minimally invasive continuous monitoring and in-depth analysis of patients' excrement, and dynamically adjusts drug treatment plans accordingly.

[0046] Specifically, the excrement collection and pretreatment module is deployed at the patient's bedside. Its main functions are automated collection of patient excrement, preliminary physical and chemical property testing, and sample pretreatment. The core component of this module is a smart bedpan structure, which cleverly integrates a sensor array, a dilution and mixing unit, and a sample diversion unit. The receiving cavity of the smart bedpan is integrally molded from high-strength medical-grade polypropylene material using a precision injection molding process. Its inner surface is coated with a hydrophilic polytetrafluoroethylene nanocomposite anti-stick coating with a thickness of approximately 50 micrometers. This coating has extremely low surface tension (approximately 18 mN / m), effectively preventing excrement adhesion and residue, facilitating subsequent automated cleaning and disinfection, and ensuring biocompatibility.

[0047] An integrated sensor array is precisely installed in pre-drilled grooves on the bottom and side walls of the receiving cavity, directly or indirectly contacting the excrement and capturing various physicochemical parameters of the excrement in real time. The array includes:

[0048] First, a high-precision piezoresistive force sensor array, consisting of four independent silicon-based microelectromechanical systems (MEMS) piezoresistive sensors, is symmetrically distributed in a rectangular array at the bottom of the receiving cavity to ensure accurate measurement of the total mass and center of gravity distribution of excrement. Each sensor has a measurement accuracy of 0.01 kg and a repeatability of 0.005 kg within a range of 0 kg to 20 kg. The analog voltage signal output by each sensor is converted into a digital signal by a TI ADS1256 24-bit high-precision analog-to-digital converter (ADC) at a sampling rate of 100 times per second, and then transmitted to the edge computing unit via an I2C bus. This array not only monitors the weight of excrement but also calculates the change in the center of gravity of the excrement within the toilet bowl in real time by analyzing the differences in the signals from the four sensors, providing auxiliary data for subsequent morphological analysis.

[0049] Secondly, a multispectral image acquisition unit, consisting of a CMOS image sensor and an autofocus liquid lens, is integrated into a rotatable robotic arm above the receiving cavity. This arm automatically adjusts the shooting angle and focus after excrement enters, ensuring image clarity. The CMOS sensor boasts a frame rate of up to 60fps, sufficient to capture the dynamic changes during excrement discharge. The multispectral light source comprises eight independently controlled narrowband LED modules with center wavelengths of 470nm (blue), 530nm (green), 630nm (red), 750nm (near-infrared-1), 850nm (near-infrared-2), 940nm (near-infrared-3), 1050nm (near-infrared-4), and 1200nm (near-infrared-5). The full width at half maximum (FWHM) of each LED is strictly controlled within ±10nm. Through a high-speed timing controller, these LEDs can be sequentially triggered for illumination, with an exposure time of 20 milliseconds for each band, thereby capturing the reflectance images of excrement at different spectral levels in a short time. All image data is transmitted in real time to the edge computing unit via a Gigabit Ethernet interface supporting the GigE Vision protocol, in a lossless compression format.

[0050] Third, a high-precision non-contact viscosity sensor. This sensor transmits a 24GHz microwave signal and receives its reflection and attenuation signals in excrement. Combined with Doppler effect analysis, it derives the dynamic viscosity coefficient of the excrement. It has a wide measurement range, covering 1 mPa·s to 5000 mPa·s, with a measurement accuracy of ±5%, ensuring accurate characterization of stool samples ranging from thin, watery to thick, pasty. The sensor uses a non-contact installation, avoiding the risk of cross-contamination, and communicates with the edge computing unit via an RS-485 interface using the Modbus protocol, with a data update frequency of 10Hz.

[0051] Fourth, an integrated electrochemical sensor array, consisting of multiple microelectrodes, is integrated onto a replaceable microfluidic chip. This includes a Swiss Hamilton VisiFerm glass electrode pH sensor with a measurement range of pH 1.0 to pH 14.0, an accuracy of ±0.02 pH units at 25°C, and a response time of less than 10 seconds. A custom-designed platinum electrode conductivity sensor measures from 0.1 mS / cm to 200 mS / cm with an accuracy of ±1%, and its electrode constant is calibrated at multiple points. Additionally, it includes an array of ion-selective electrodes (ISEs) made of polymer membrane electrodes, capable of simultaneously detecting the concentrations of sodium (Na+), potassium (K+), and chloride (Cl-) ions. These ISE electrodes all have a detection limit of 0.1 mmol / L and a measurement range of 0.5 mmol / L to 200 mmol / L. Within 5 seconds of the excrement discharge being completed, the entire electrochemical sensor assembly undergoes automated cleaning and two-point calibration by precisely injecting 5 mL of standard buffer solution (such as pH 7.0 phosphate buffer and sodium chloride solution of known concentration) via a micro-injection pump, ensuring the long-term stability and accuracy of the measurement data.

[0052] Fifth, a temperature sensor, employing a high-performance NTC thermistor package (e.g., VishayNTCLE100E3), has a measurement range covering 20°C to 50°C with an accuracy of ±0.1°C. This sensor is located adjacent to the electrochemical sensor array and is used for real-time monitoring of excrement temperature. It also provides temperature compensation correction to the electrochemical sensor measurements to eliminate the potential impact of temperature variations on pH, conductivity, and ion concentration measurements.

[0053] The dilution and mixing unit is located below the receiving chamber of the smart toilet bowl and is connected to the bottom of the chamber via a self-sealing electric valve. Once the excrement collection is complete, the valve opens, allowing a quantitative sample of excrement to flow into the dilution and mixing unit. The core components of this unit are two high-precision peristaltic pumps and a micro-mixing chamber. The first peristaltic pump (e.g., the Masterflex L / S series from Cole-Parmer) precisely controls the rotation angle of its stepper motor to draw 2.0 mL to 10.0 mL of excrement sample from the bottom of the receiving chamber with a minimum precision of 0.5 mL, and injects it into the 20 mL micro-mixing chamber at a constant flow rate of 1.0 mL / s. Simultaneously, the second peristaltic pump precisely injects sterile saline as a diluent at a specific ratio, such as 1:5 (i.e., 1 part excrement sample to 5 parts diluent). The interior of the micro-mixing chamber is made of biocompatible materials and is equipped with a magnetically driven micro-stirring paddle that can rotate at up to 1500 rpm. This ensures that the excrement sample and the diluent are fully and uniformly mixed within a 15-second stirring time, forming a homogenized diluent that provides a reliable analytical sample for subsequent fine-grained testing.

[0054] The sample splitting unit, adjacent to the dilution and mixing unit, consists of a series of precision-machined microfluidic channels and an array of electrically controlled microvalves. The homogenized sample, after dilution and mixing, is precisely split through the microvalve array, according to instructions from the edge computing unit, into the measurement cell of the integrated electrochemical sensor array. It can also be further split into other reserved micro-biochip detection units for automated detection of specific biomarkers such as fecal calprotectin and occult blood. The microvalves utilize piezoelectric-driven MEMS technology, boasting a fast switching response speed of up to 20 milliseconds, ensuring precise and efficient flow path switching and minimizing sample loss and cross-contamination.

[0055] Furthermore, the multimodal data analysis module, as the core intelligent processing unit of this invention, primarily performs in-depth fusion analysis on the raw multimodal data output by the excrement collection and preprocessing module. This allows for the objective and quantitative identification of the qualitative characteristics of the excrement, as well as the diagnosis of diarrhea type and severity assessment. This module works collaboratively with an edge computing unit and a cloud-based intelligent processing platform.

[0056] The edge computing unit employs a high-performance embedded computing platform. This unit is responsible for receiving raw data streams from all sensors in the smart toilet, including image data, spectral data, force sensor data, viscosity data, electrochemical data, and temperature data. Before the data is stored, the edge computing unit performs a series of preprocessing operations, including high-frequency noise filtering (such as using Kalman filtering or wavelet denoising), data normalization (Min-Max or Z-score normalization), and image registration and correction to eliminate the impact of uneven ambient lighting and minor sensor position deviations on data analysis.

[0057] The edge computing unit embeds and runs a set of pre-trained deep learning models, specifically including:

[0058] First, an image feature extraction and texture analysis model based on a convolutional neural network (CNN). This model employs an improved variant of the U-Net architecture, with its encoder consisting of up to 18 layers of residual blocks (ResNet18 Blocks), effectively capturing multi-scale features in the image. The input layer receives a normalized RGB image (resolution adjusted to 512x512 pixels) and a calibrated near-infrared image (e.g., 850nm and 1050nm bands). Through multi-layer convolution, pooling, and activation function operations, it extracts the color distribution features, surface texture patterns, macroscopic morphology (e.g., presence of undigested food residue, foam, mucus), and the distribution and size information of particulate matter (e.g., fiber, blood clots) from the excrement. The encoder outputs a high-dimensional feature vector (e.g., 1024-dimensional), accurately representing the visual attributes of the excrement.

[0059] Second, a sequence data analysis model based on Long Short-Term Memory (LSTM) networks. This model is specifically designed to process time-series data from force sensors, viscosity sensors, and electrochemical sensor arrays. Input sequences include pressure change curves during excrement emptying (sampling frequency 100 Hz), viscosity dynamic change sequences (sampling frequency 10 Hz), and time-series data of pH, conductivity, and ion concentrations (Na+, K+, Cl-) (sampling frequency 1 Hz). Through its unique input, forget, and output gate mechanisms, the LSTM model can effectively capture long-term dependencies and transient patterns in these time-series data, thereby extracting the rheological properties of excrement (such as flowability and shear-thinning behavior), dynamic changes in chemical composition (such as rapid pH decreases or sudden increases in ion concentrations), and emptying patterns (such as intermittent or continuous discharge).

[0060] Third, a multimodal fusion model based on the Transformer architecture. This model aims to deeply fuse the image feature vectors output by the CNN model and the temporal feature vectors output by the LSTM model to generate a more discriminative joint feature representation. The encoder part of the Transformer model consists of six stacked attention layers, each containing a multi-head attention sublayer (Head=8) and a feedforward network sublayer. The multi-head attention mechanism allows the model to focus on information from different modalities in different representation subspaces, while the self-attention mechanism is used to capture the correlation between elements within the same modality. The cross-attention mechanism is the core of this model, enabling image features to "query" temporal features and vice versa, thereby learning and fusing complex intrinsic relationships between different modalities. For example, foam observed in an image and a rapid decrease in pH value in electrochemical data may jointly indicate a specific fermentation process.

[0061] The final output layer of the multimodal fusion model is a fully connected network used for classification and regression prediction of excrement characteristics. The classification task identifies specific types of diarrhea (e.g., osmotic diarrhea, secretory diarrhea, infectious diarrhea, inflammatory diarrhea) and outputs the probability distribution for each type, for example: "Secretory diarrhea: 0.85, Osmotic diarrhea: 0.10, Infectious diarrhea: 0.03". The regression task predicts the severity score of diarrhea, for example, a continuous numerical score based on the improved BristolStool Scale, ranging from 0 to 7 (where 0 represents severe constipation and 7 represents severe watery stool), and can predict the probability of the presence of specific biomarkers (such as fecal calprotectin concentration and hematuria). The edge computing unit completes all the above analyses within 5 seconds of data acquisition and outputs detailed diagnostic results and confidence levels.

[0062] The cloud-based intelligent processing platform and edge computing units synchronize data and update models in real time via encrypted WebSocket protocol, ensuring secure data transmission and continuous algorithm optimization. The platform hosts a larger-scale clinical decision engine built on a deep reinforcement learning (RL) paradigm, used to generate personalized drug delivery plans based on the patient's real-time condition.

[0063] The clinical decision engine is a highly integrated intelligent system that includes:

[0064] First, a structured knowledge base stores rigorously reviewed, up-to-date clinical guidelines (such as domestic and international diarrhea treatment guidelines), pharmacokinetic (PK) / pharmacodynamic (PD) models (covering commonly used antidiarrheal drugs, antibiotics, probiotics, and fluid replacement in the ICU), drug interaction databases (e.g., interactive data based on Lexicomp or Medscape), lists of patient contraindications, and standardized recommended treatment regimens for different types of diarrhea. This knowledge base utilizes Web Ontology Language (OWL) for semantic modeling, achieving a high degree of structure and reasonability, and is regularly updated and maintained by a dedicated team based on the latest research findings and guidelines.

[0065] Second, a patient-specific data interface employs a data transmission protocol compliant with the HL7 FHIR (Fast Healthcare Interoperability Resources) standard. This interface enables real-time and secure acquisition of individualized patient information from multiple data sources, including the hospital's Electronic Health Record (EHR) system, vital signs monitors, and Laboratory Information System (LIS). This information includes, but is not limited to, the patient's age, weight, height, medical history (e.g., diabetes, renal insufficiency, abnormal liver function), allergy history, comorbidities, current vital signs (e.g., heart rate, blood pressure, body temperature, blood oxygen saturation, CVP, MAP), and the latest laboratory test results (e.g., complete blood count, electrolytes, liver and kidney function, blood gas analysis, C-reactive protein, procalcitonin, stool culture and drug sensitivity results, urine output, etc.). This data is integrated to form a comprehensive physiological state vector for the patient.

[0066] Third, a decision agent based on deep reinforcement learning (Deep Q-Network, DQN). This agent takes the patient's current state as input, and its state space consists of high-dimensional vectors representing the results of excrement trait identification (diarrhea type, severity score, biomarker prediction), the patient's individualized physiological data, current medication status, and drug responses from the previous decision-making cycle. Its output is the drug delivery action (action space), including drug type (e.g., montmorillonite powder, loperamide, aspergillus oryzae, Clostridium butyricum live bacteria powder, broad-spectrum antibiotics such as imipenem, electrolyte supplements, parenteral nutrition solutions, etc.), precise dosage (e.g., 3 to 12 grams of montmorillonite powder per dose, 0.5 to 4 mg of loperamide per dose, antibiotic dosage precisely calculated based on body weight), dosing frequency (e.g., every 4 hours, every 6 hours, every 8 hours, continuous infusion), and route of administration (oral, nasogastric tube, intravenous injection). The reward function of the decision-making agent is carefully designed to maximize the speed and stability of the recovery of excretory characteristics to normal ranges while ensuring patient safety (e.g., avoiding side effects such as constipation, intestinal obstruction, electrolyte imbalance, drug allergies, and liver and kidney damage), and simultaneously considering reducing patient discomfort and lowering total drug consumption. The agent is trained on a large scale with a simulated environment based on patient physiology and drug pharmacokinetic / pharmacodynamic (PK / PD) models, and fine-tuned using historical real-world clinical data to learn optimal treatment strategies for various complex clinical scenarios.

[0067] Furthermore, the drug delivery closed-loop control module is the system execution layer, used to deliver drugs precisely and dynamically according to the plan generated by the clinical decision engine. This module includes an intelligent infusion pump array and a feedback regulation controller.

[0068] The intelligent infusion pump array comprises at least two independent infusion pump units, typically configured as a combination of a peristaltic pump (for enteral nutrition solutions and oral medication suspensions) and an injection pump (for intravenous drug administration). Each unit can independently control its infusion channel. Each pump unit is driven by a high-precision stepper motor, with a flow rate control range of 0.1 mL / h to 100 mL / h, flow rate accuracy up to ±2%, and volume accuracy up to ±1%, ensuring accurate drug infusion. To ensure patient safety, the pump units integrate air alarm (ultrasonic sensor), obstruction alarm (pressure sensor), and free-flow protection functions. Each pump unit is equipped with a drug identification unit at the front end, which includes a high-performance 2D barcode scanner and an RFID reader. By scanning the GS1 DataMatrix code on the medicine bottle or infusion bag or reading the RFID tag, the system automatically identifies the drug type, batch number, production date, expiration date, and precise concentration. This information is compared in real time with the medication regimen generated by the clinical decision engine. If any mismatch is found, the system will immediately issue an alarm and pause the infusion to prevent medication errors. The pump unit's drug solution piping is made entirely of medical-grade silicone or DEHP-free polyvinyl chloride (PVC) material, which has excellent biocompatibility and corrosion resistance to common drug solutions.

[0069] The feedback control controller is an embedded microcontroller, such as the STM32F4 series chip based on the ARM Cortex-M4 processor, equipped with a floating-point unit (FPU) and rich communication interfaces. This controller receives drug delivery instructions (including drug name, dosage, frequency, and route) from the clinical decision engine and converts them into precise control signals for the intelligent infusion pump array. The controller continuously monitors the actual infusion status of the intelligent infusion pumps (including real-time flow rate, infused volume, and pump pressure) and feeds this execution data back to the clinical decision engine in real time, forming a micro-closed loop of execution status. Simultaneously, the controller maintains communication with the excrement collection and preprocessing module via a high-real-time data link, continuously acquiring the latest excrement characteristics data, and obtaining key physiological parameters such as the patient's vital signs (heart rate, blood pressure) and urine output through the bedside monitor interface.

[0070] The feedback control system incorporates a multivariate model predictive control (MPC) algorithm. This MPC algorithm uses fecal morphology parameters (such as diarrhea severity score, excretion frequency, and fecal electrolyte concentration) as the core control objective and the drug infusion rate (including drug type, dosage, and flow rate) of the intelligent infusion pump as the control variable. Within each preset control cycle (e.g., every 30 minutes), the MPC algorithm combines current fecal morphology data, patient physiological status data, and a drug action model calibrated based on individualized PK / PD parameters to predict the evolution trend of fecal morphology over a future period (e.g., the next 4 hours). Based on this prediction, the MPC algorithm solves a constrained optimization problem to calculate the optimal drug infusion strategy while satisfying medication safety constraints (e.g., maximum drug dose limits, minimum dosing intervals, known drug interactions, and patient liver and kidney function limitations). The objective function of this optimization process is designed to minimize the deviation between fecal morphology (e.g., Bristol score) and the target value, while minimizing the total drug usage to reduce side effects and costs, and smoothing drug dosage adjustments as much as possible to avoid drastic fluctuations. Then, the MPC algorithm sends the calculated first-step optimal infusion command (e.g., Loperamide 1.5 mg, continuous infusion of enteral nutrition solution 30 mL / h) to the intelligent infusion pump array for execution. In the next control cycle, the system will recalculate and optimize based on the latest monitoring data to achieve continuous closed-loop feedback regulation.

[0071] Furthermore, the human-computer interaction and early warning module aims to provide an intuitive user interface, enabling medical staff to monitor and intervene in the system status, receive important alerts, and support limited information sharing with patients' families.

[0072] The human-computer interaction and early warning module includes a high-performance nurse station terminal, a mobile application, and an independent early warning system.

[0073] The nurse station terminal is equipped with a 24-inch high-resolution (1920x1200) capacitive touchscreen display. Its interface, presented through a carefully designed graphical user interface (GUI), provides real-time comprehensive patient information, including: raw data on stool characteristics (multispectral images, spectral curves, viscosity values, pH values, and trend graphs of various ion concentrations), diarrhea type diagnosis results and their confidence levels, dynamic change curves of severity scores, the current medication delivery protocol recommended by the clinical decision engine, the amount of medication delivered by the intelligent infusion pump, the pumping status, and the patient's real-time vital signs data. The terminal provides a comprehensive role-based access control system, requiring healthcare personnel to log in via biometrics (such as fingerprints or facial recognition) or multi-factor authentication (such as employee ID + password + dynamic password). The terminal interface includes an intuitive manual intervention interface, allowing authorized senior healthcare personnel to pause, modify, or completely take over the medication delivery protocol in special circumstances. The system rigorously records all manual operations and their reasons, operator IDs, and timestamps, ensuring traceability and clear accountability. In addition, the terminal also has integrated remote consultation function, which allows doctors to communicate in real time with remote experts through encrypted video conferencing that complies with HIPAA standards, and jointly discuss and optimize treatment plans.

[0074] The mobile application supports mainstream iOS and Android operating systems and synchronizes encrypted data with the system via a secure cloud platform. For patients' families, the application provides limited, anonymized, and non-sensitive data views, such as excretion frequency trend graphs and treatment progress overviews (without specific diagnostic information or detailed medication details), aiming to protect patient privacy while providing information of concern to families. For authorized physicians and senior nurses, the application provides the ability to remotely view detailed patient monitoring data, diagnostic results, and system-recommended treatment plans, and supports remote approval or modification of treatment plans. All remote operations require two-factor authentication (e.g., SMS verification code or token) to ensure the security and compliance of operations.

[0075] The independent early warning system includes an independently powered audible and visual alarm and a multi-channel message notification service. The audible and visual alarm uses a high-decibel buzzer and a high-brightness LED flashlight. When specific emergency conditions are met, such as when the smart bedpan detects a large amount of fresh blood in the excrement (e.g., image analysis identifies heme spectral characteristics, or the occult blood detection unit alarms), or when the electrolyte sensor detects abnormally critical electrolyte concentrations (e.g., Na+ below 100 mmol / L or above 180 mmol / L), or when the drug delivery system experiences mechanical failure (e.g., infusion tubing detachment, pump malfunction), or when the patient's vital signs exceed safe ranges (e.g., heart rate consistently below 40 bpm or above 150 bpm), or when the system determines that the patient is at serious risk of dehydration (e.g., urine output consistently below 0.5 mL / kg / h for 2 hours, and serum sodium concentration exceeding 155 mmol / L), the system will immediately trigger a local audible and visual alarm and issue a visual prompt through the nurse station terminal. Meanwhile, the notification service sends detailed alerts in real time to pre-defined medical team members (such as on-duty doctors and head nurses) through multiple secure channels (including SMS messages, push notifications from the hospital's internal instant messaging system, and notifications from smart wearable devices). The alert information includes the triggering conditions, patient ID, time of occurrence, and the system's suggested initial handling measures. The alert system has a sophisticated alarm classification function, divided into critical, urgent, and general alerts. Different alert levels correspond to different response times and notification scopes, ensuring effective resource allocation and timely intervention.

[0076] This invention also provides a closed-loop control method for excrement characteristics identification and drug delivery, which is implemented through the above-mentioned system and includes the following steps:

[0077] The excrement collection and pretreatment module automates the collection of patient excrement, performs preliminary physical and chemical property testing, and preprocesses the samples.

[0078] The multimodal data analysis module performs in-depth fusion analysis on the raw multimodal data output by the excrement collection and preprocessing module to objectively and quantitatively identify the qualitative characteristics of the excrement and to diagnose the type and severity of diarrhea.

[0079] The drug delivery closed-loop control module delivers drugs accurately and dynamically based on the diagnostic results of the multimodal data analysis module;

[0080] The human-computer interaction and early warning module provides an intuitive user interface, enabling medical staff to monitor and intervene in the system status, receive important alarms, and support limited information sharing with patients' families.

[0081] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Those skilled in the art can readily implement the present invention based on the accompanying drawings and the above description. However, any modifications, alterations, or variations made by those skilled in the art without departing from the scope of the present invention, utilizing the disclosed technical content, are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, or variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.

Claims

1. A system for excrement property identification and drug delivery closed loop control, comprising: Comprise: a feces collection and pretreatment module for automated collection, preliminary physical and chemical property detection and sample pretreatment of patient feces; a multi-modal data analysis module in communication with the feces collection and pretreatment module for deep fusion analysis of raw multi-modal data output by the feces collection and pretreatment module to objectively and quantitatively identify qualitative property characteristics of feces and make diarrhea type diagnosis and severity assessment; a drug delivery closed-loop control module in communication with the multi-modal data analysis module for precise and dynamic infusion of drugs according to diagnostic results of the multi-modal data analysis module; a human-computer interaction and early warning module in communication with the feces collection and pretreatment module, the multi-modal data analysis module and the drug delivery closed-loop control module for providing an intuitive user interface to realize monitoring, intervention and receiving important alarms of system status by medical staff and support limited information sharing with patient family members.

2. The fecal property identification and medication delivery closed loop control system according to claim 1, wherein, The feces collection and pretreatment module comprises an intelligent toilet structure, a dilution mixing unit and a sample diversion unit; The intelligent toilet structure is composed of a receiving cavity and an integrated sensor array; The integrated sensor array is accurately installed at the bottom and sidewall of the receiving cavity and comprises: a high-precision piezoresistive force sensor array composed of at least four independent piezoresistive sensors for real-time measurement of total mass of feces and its distribution center of gravity in the toilet; a multi-spectral image acquisition unit comprising a high-resolution CMOS image sensor and a tunable multi-spectral light source for automatic focusing and capturing reflection images of feces under different spectral bands under different amounts and forms of feces; a high-precision non-contact viscosity sensor adopting acoustic scattering principle based on microwaves or ultrasonic waves for deriving dynamic viscosity coefficient of feces; an integrated electrochemical sensor group comprising a glass electrode pH sensor, a platinum gold electrode conductivity sensor and an ion selective electrode array for detecting pH value, conductivity and specific ion concentration of feces; and a temperature sensor for real-time monitoring of feces temperature to correct temperature drift of the integrated electrochemical sensor group.

3. The fecal property identification and medication delivery closed loop control system according to claim 2, wherein, The high-precision piezoresistive force sensor array is composed of at least four independent silicon-based micro-electro-mechanical system piezoresistive sensors; and the ion selective electrode array can simultaneously detect sodium ion, potassium ion and chloride ion concentration.

4. The fecal property identification and medication delivery closed loop control system according to claim 2, wherein, The dilution mixing unit is arranged below the receiving cavity of the intelligent toilet structure and comprises two peristaltic pumps and a 20 mL micro mixing chamber, one of the peristaltic pumps sucks feces sample and injects into the micro mixing chamber, the other peristaltic pump injects sterile saline as diluent at a specific ratio, the micro mixing chamber is internally provided with a micro stirring paddle driven by magnetic force; and the sample diversion unit is adjacent to the dilution mixing unit and is composed of a microfluidic channel and an electrically controlled micro valve array.

5. The fecal property identification and medication delivery closed loop control system of claim 1, wherein, The multi-modal data analysis module comprises an edge computing unit and a cloud intelligent processing platform; The edge computing unit is responsible for receiving all raw data streams from the integrated sensor array and performing preprocessing operations such as high-frequency noise filtering, data standardization, image registration, and correction. A set of deep learning models runs within the edge computing unit, including: An image feature extraction and texture analysis model based on convolutional neural networks, which extracts color distribution, surface texture patterns, macroscopic morphology, and particle distribution and size spatial features from multispectral images and outputs high-dimensional feature vectors.

6. The fecal property identification and medication delivery closed loop control system according to claim 5, wherein, A sequence data analysis model based on long short-term memory networks, which processes time-series data from the high-precision piezoresistive force sensor array, high-precision non-contact viscosity sensors, and integrated electrochemical sensors to extract rheological properties, dynamic chemical composition changes, and emptying patterns of the excrement.

7. The fecal property identification and medication delivery closed loop control system according to claim 6, wherein, A multi-modal fusion model based on the Transformer architecture, with an encoder consisting of six stacked attention layers, which deeply fuses image feature vectors and time-series feature vectors to generate more discriminative joint feature representations and classifies and regresses excrement properties through a final fully connected network layer to identify diarrhea types and predict diarrhea severity scores. The edge computing unit can complete all deep fusion analysis and output detailed diagnostic results and confidence within 5 seconds after data collection is complete.

8. The fecal property identification and medication delivery closed loop control system of claim 1, wherein, The cloud intelligent processing platform and the edge computing unit perform real-time data synchronization and model updates through an encrypted WebSocket protocol.

9. The fecal property identification and medication delivery closed loop control system of claim 1, wherein, The cloud intelligent processing platform hosts a larger-scale clinical decision engine based on deep reinforcement learning.

10. A method of excrement property identification and drug delivery closed-loop control, characterized in that, The clinical decision engine includes: a structured knowledge base, a patient individualized data interface, and a deep reinforcement learning-based decision agent. The drug delivery closed-loop control module includes an intelligent infusion pump array and a feedback regulation controller. The human-machine interaction and warning module includes a nurse station terminal, a mobile application, and a standalone warning system. The system includes the following steps: The excrement collection and preprocessing module automatically collects patient excrement, performs preliminary physical and chemical property detection, and pre-processes samples. The multi-modal data analysis module deeply fuses raw multi-modal data output by the excrement collection and preprocessing module to objectively and quantitatively identify qualitative and quantitative features of excrement and diagnose diarrhea types and severity. The drug delivery closed-loop control module accurately and dynamically infuses drugs based on the diagnostic results of the multi-modal data analysis module. The human-machine interaction and warning module provides an intuitive user interface for healthcare professionals to monitor system status, intervene, and receive important alerts, and supports limited information sharing with patient families.

Images