A centralized monitoring-based infusion progress visualization early warning management system and method

Abstract

The application discloses a kind of infusion process visualization early warning management system and method based on centralized monitoring, system includes multiple image acquisition terminal, wireless data transmission module, centralized monitoring server and nurse station visualization terminal.Image acquisition terminal real-time acquisition infusion bottle or infusion bag visible light or infrared image, upload to centralized monitoring server by wireless network.Server uses optical character recognition to extract label information, calculates residual liquid volume using image segmentation and edge detection algorithm, combines flow rate to analyze dynamic prediction remaining time.Alarm decision module triggers multi-level warning according to remaining time, displays the infusion state of all patients in graphical interface at nurse station terminal, including residual volume, remaining time, next bottle of medicine and warning level.The application realizes the centralized monitoring and visualization early warning of infusion process, can remind nurses to prepare for drug change in advance, avoid liquid emptying risk, reduce the workload of nurse patrol.

Description

Technical Field

[0001] This invention relates to the fields of medical Internet of Things and intelligent nursing technology, and in particular to a visualization early warning management system and method for infusion process based on centralized monitoring. Background Technology

[0002] Intravenous infusion is one of the most basic and common methods of drug administration in clinical treatment. In large hospitals, nurses in a ward are often responsible for the intravenous infusion care of multiple patients simultaneously. Each patient may require multiple bottles of intravenous fluid, and the infusion sequence, types of medication, and infusion rate requirements vary from patient to patient. In the traditional nursing model, nurses need to frequently patrol between beds, visually observing the remaining volume in each bottle to determine when to replace the next bottle or remove the needle. This method has the following significant drawbacks: First, the workload is heavy. Nurses need to spend a lot of time monitoring IV fluids, which takes away time and energy that should be used to directly care for patients. Especially during night shifts or when staff are short-staffed, nurses are often overwhelmed, leading to a decline in the quality of care.

[0003] Second, the response is not timely. Because the precise remaining volume of each bottle of fluid cannot be known in real time, nurses can only make judgments during rounds. This can easily lead to situations where the fluid is empty and not detected in time, resulting in adverse events such as backflow of blood and air entering the blood vessels, which can endanger the patient's safety in severe cases.

[0004] Third, there is a lack of predictive ability. Even if a nurse knows that a particular IV drip is about to run out, they cannot accurately predict in advance which medication will need to be replaced, because the infusion order varies from patient to patient, and even for the same patient, the infusion order may change due to adjustments in doctor's orders. Nurses often need to run to the bedside to check the label before returning to prepare, which leads to inefficiency and increases the risk of administering the wrong medication in haste.

[0005] Fourth, existing infusion alarms have limited functionality. Currently, there are some individual infusion alarms on the market based on gravity or photoelectric sensors, which can only emit a local audible alarm when the infusion is finished. However, these devices cannot achieve centralized monitoring; nurses at the nursing station cannot simultaneously see the infusion status of all patients. They also cannot provide early warnings, only alarming when the infusion is exhausted, leaving nurses almost no preparation time. Furthermore, they cannot manage the infusion sequence or provide reminders for the next vial.

[0006] Fifth, there is a lack of data recording and traceability capabilities. The existing solution cannot automatically record key nursing data such as the start time, end time, flow rate changes, and nurse response time for each infusion, which is detrimental to nursing quality management and adverse event analysis.

[0007] Therefore, there is an urgent clinical need for an infusion process early warning management system and method that can achieve centralized monitoring, visual display, early warning, and intelligent management of the infusion sequence. Summary of the Invention

[0008] To address the technical problems of existing infusion monitoring technologies, such as reliance on manual processes, untimely response, lack of centralized visualization management, and data traceability, this invention provides a visualization-based early warning management system and method for infusion process monitoring. The technical solution is as follows: On the one hand, the present invention provides a centralized monitoring-based visual early warning management system for infusion process, which includes: multiple image acquisition terminals, a wireless data transmission module, a centralized monitoring server, and a visual terminal at the nurse station.

[0009] Multiple image acquisition terminals, each fixedly positioned on an infusion stand, are used to acquire visible light or infrared images of the infusion bottle or bag in real time during the infusion process. The images at least include the label information area of ​​the infusion bottle or bag and the area containing the remaining fluid.

[0010] The wireless data transmission module is connected to the image acquisition terminal and is used to wirelessly upload the acquired image data to the centralized monitoring server in real time.

[0011] The centralized monitoring server is the core processing unit of the system, and it integrates multiple functional modules: The image receiving module is used to receive image data from multiple image acquisition terminals.

[0012] The image preprocessing module is used to denoise, enhance, correct, and extract regions of interest from the received image to improve the accuracy and robustness of subsequent processing.

[0013] The label information recognition module uses an optical character recognition algorithm to extract key medical order information such as patient identity information, drug name, drug dosage, preparation time, and preset infusion sequence from the label information area.

[0014] The liquid level detection module uses image segmentation and edge detection algorithms to analyze the remaining liquid area and calculate the remaining liquid volume or remaining liquid level height in the current infusion bottle or infusion bag.

[0015] The flow rate analysis module calculates the current infusion flow rate in real time based on the changes in liquid level and the corresponding time intervals in multiple consecutive frames of images.

[0016] The remaining time prediction module dynamically predicts the estimated completion time of the current infusion bottle or bag based on the remaining liquid volume and the current flow rate.

[0017] The sequence management module automatically generates an infusion sequence queue for each patient based on the preset infusion order and label information recognition results, and tracks in real time the infusion bottle or bag currently being infused and the replacement status of subsequent infusion bottles or bags.

[0018] The early warning decision module automatically generates a replacement early warning signal for the infusion bottle or bag when it predicts that the remaining time of the current infusion bottle or bag is lower than a preset threshold, and determines the preparation prompt for the next infusion bottle or bag based on the sequence management module.

[0019] The nurse station's visual terminal communicates with a centralized monitoring server to display an overview of all patients' infusion status in a graphical interface. The infusion status includes: patient bed number / identifier, medication information for the current infusion bottle or bag, remaining volume, remaining time, medication information for the next infusion bottle or bag, warning status, and medication preparation status.

[0020] Furthermore, to improve image acquisition quality, the image acquisition terminal also includes a supplementary lighting unit and an ambient light sensor. The supplementary lighting unit automatically adjusts the brightness of the supplementary light according to the light intensity detected by the ambient light sensor, ensuring clear and stable images can be obtained under any lighting conditions. In addition, to optimize the shooting angle, the angle between the central axis of the lens of the image acquisition terminal and the vertical central axis of the infusion bottle is set to 0° to 15°, and the distance between the lens and the label area of ​​the infusion bottle is 15cm to 40cm.

[0021] Furthermore, to achieve high-precision liquid level detection, the liquid level detection module employs a deep learning-based semantic segmentation network to perform pixel-level segmentation of the liquid region in the infusion bottle image. The output of the semantic segmentation network is a binary mask of the liquid region, and the remaining liquid volume is calculated using a pre-established mapping function between liquid volume and the number of mask pixels. This mapping function is a nonlinear regression model (such as a multinomial regression or neural network model) obtained through calibration experiments on infusion bottles of different specifications and shapes (e.g., round, elliptical, and irregularly shaped bottles), which can accurately compensate for the nonlinear relationship between liquid level height and volume caused by the bottle shape.

[0022] Furthermore, the flow rate analysis module further includes an abnormal flow rate detection submodule, used to compare the real-time calculated infusion flow rate with a preset safe flow rate range (for example, for commonly used adult infusions, the safe range can be set to 60-120 drops / minute, or a specific flow rate value set according to the doctor's order). When the detected flow rate exceeds the safe range for a preset duration threshold (e.g., 30 seconds), an abnormal flow rate alarm signal is generated, which includes an alarm for excessively fast flow rate and an alarm for excessively slow flow rate.

[0023] Furthermore, the sequence management module supports intelligent sorting management of multiple bottles of continuous infusion, specifically including: The infusion sequence input submodule is used to input infusion sequence information, supporting three methods: manual input, importing by scanning barcodes / QR codes, and automatic acquisition from the hospital information system.

[0024] The dynamic reordering submodule is used to automatically update the queue and recalculate the warning time for all subsequent infusion bottles or bags when the actual infusion order is detected to be inconsistent with the preset order (for example, the next bottle of medication actually hung by the nurse is different from what the system expects), so as to prevent subsequent warnings from being wrong.

[0025] The interruption and recovery management submodule is used to automatically record the interruption time when it is detected that an infusion bottle has been paused or removed due to reasons such as the patient needing to check or adverse reactions, and to recalculate the remaining time and flow rate after the infusion is resumed.

[0026] Furthermore, the early warning decision module adopts a multi-level early warning strategy, including: Level 1 Warning (Preparation Level): When less than 10 minutes remain, the patient will be displayed in yellow on the nurse station's visual terminal, and a text prompt to prepare the next bottle of medication will be generated.

[0027] Level 2 warning (reminder level): When the remaining time is less than 3 minutes, it will be displayed in orange and a soft voice prompt will be issued on the nurse station terminal. At the same time, a medication change notification will be pushed to the corresponding nurse's mobile terminal (such as PDA or smartphone APP).

[0028] Level 3 Warning (Emergency Level): When the infusion is finished or air is detected entering the infusion tube through image analysis (for example, the liquid level detection module detects that the liquid level has disappeared for several seconds), it will be displayed with a flashing red indicator and will emit a continuous, loud audible alarm. At the same time, the system can automatically call the corresponding responsible nurse or the on-duty nurse station through the call system.

[0029] The thresholds for the multi-level early warning can be customized by the user based on the department (such as ICU, pediatrics, general ward) or the patient's condition (such as heart failure patients who need early warning).

[0030] Furthermore, the nurse station visualization terminal also includes a historical data traceability module, used to record and display complete data for each patient's infusion, including: infusion start time, time for each bottle change, actual flow rate curve, warning trigger records, nurse response time, and dressing change operation confirmation records. The historical data traceability module supports retrieval and statistical analysis by time range, patient, medication name, or nurse name, and automatically generates infusion nursing quality reports (such as average dressing change response time, frequency of abnormal flow rate occurrences, and nurse workload statistics).

[0031] On the other hand, the present invention provides a method for visual early warning management of infusion process based on centralized monitoring. This method uses the system described in any of the above claims and includes the following steps: S1: Image acquisition: Multiple image acquisition terminals installed on the infusion stand acquire image data of the corresponding infusion bottle or infusion bag in real time. The image data includes the label information area and the liquid remaining area.

[0032] S2: Image Upload: The collected image data is uploaded to the centralized monitoring server in real time via wireless network.

[0033] S3: Image preprocessing and region of interest extraction: The received image is denoised, grayscaled, contrast-enhanced and geometrically corrected in the centralized monitoring server, and the label information region sub-image and the liquid remaining region sub-image are extracted based on the preset region of interest template.

[0034] S4: Label Information Recognition and Verification: The optical character recognition algorithm is used to recognize the text in the label information area sub-image, extract the patient identification, drug name, drug dosage and infusion sequence information, and cross-verify it with the medical orders in the hospital information system. If the information is inconsistent, a verification alarm is issued.

[0035] S5: Liquid level detection and volume calculation: Edge detection algorithms (such as Canny edge detection) and region growing algorithms are used to process the liquid remaining region sub-image, identify the boundary line between liquid and air, and calculate the current remaining liquid volume V_rem by combining the pre-calibrated infusion bottle geometric parameters (such as the bottle cross-sectional area function).

[0036] S6: Flow velocity calculation and smoothing: Based on the time interval Δt between two consecutive frames and the corresponding remaining volume change ΔV, the current instantaneous flow velocity is calculated as v_inst = ΔV / Δt, and a stable flow velocity V_cur is obtained by using a moving average filter (e.g., taking the average of the most recent 5-10 instantaneous flow velocities).

[0037] S7: Remaining Time Prediction and Correction: Based on the current remaining volume V_rem and the steady flow rate V_cur, calculate the expected remaining time T_rem = V_rem / V_cur, and use a Kalman filter to smooth and dynamically correct the prediction results to suppress noise and abrupt changes.

[0038] S8: Infusion Queue Generation and Tracking: Based on the infusion order extracted from the tag information, automatically generate or update the patient's infusion queue Q = {B1, B2, ..., Bn}, and mark the position of the currently infused bottle Bk in the queue.

[0039] S9: Multi-level early warning and medication preparation prompts: The estimated remaining time T_rem is compared with the preset multi-level early warning thresholds T1, T2, and T3. When T_rem ≤ T3, a level 3 early warning is generated; when T_rem ≤ T2 and greater than T3, a level 2 early warning is generated; and when T_rem ≤ T1 and greater than T2, a level 1 early warning is generated. At the same time, medication preparation prompts are generated based on the medication information of the next bottle B(k+1) in the queue.

[0040] S10: Visual Display: The infusion status of all patients is displayed in real time on the nursing station visualization terminal in the form of dashboard, progress bar or Gantt chart, including patient identification, current medication, remaining volume, remaining time, next bottle of medication, warning level and nurse response status.

[0041] Furthermore, in step S5, the liquid level detection step, for the case of using a flexible infusion bag, the accuracy of traditional edge detection algorithms is limited due to the easily changeable shape of the bag. Therefore, the edge detection algorithm is replaced by a key point detection network based on deep learning (such as a stacked hourglass network or HRNet). This network directly regresses and outputs the coordinates of two key points on the boundary of the liquid surface inside the infusion bag, and then calculates the liquid level height according to a pre-established deformation compensation model. The deformation compensation model considers the nonlinear relationship between the bottom tensile deformation caused by gravity after the infusion bag is suspended and the actual volume of the liquid surface. This relationship can be obtained by fitting experimental data.

[0042] Furthermore, the method also includes a closed-loop confirmation step after the infusion is completed: S11: Automatic determination of infusion completion: When the system detects that the remaining volume of the current infusion bottle or infusion bag is less than the preset residual liquid threshold or the liquid level is lower than the neck of the infusion bottle, it automatically determines that the infusion of the bottle is complete.

[0043] S12: Send dressing change notification: The system sends a dressing change notification to the nursing station terminal and the mobile terminal of the corresponding responsible nurse.

[0044] S13: Nurse confirms dressing change: After completing the dressing change, the nurse confirms the dressing change by scanning the barcode of the next bottle of medication or by clicking the confirmation button on the nurse station terminal interface.

[0045] S14: System Status Update: After receiving the dressing change confirmation signal, the system automatically moves the infusion queue pointer to the next bottle, resets the remaining time and flow rate calculation, and records the response time and operator information for this dressing change operation.

[0046] S15: Unconfirmed Apprehension After Timeout: If no confirmation signal for medication change is received within the preset waiting time (e.g., 5 minutes) after the medication change notification is sent, the system will automatically upgrade the current warning to a Level 3 emergency alarm and call the shift leader or a nearby nurse for assistance.

[0047] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: 1. This invention, through real-time and automatic liquid level monitoring and flow rate analysis, can immediately issue multi-level warnings when the infusion is completed or the flow rate is abnormal, effectively preventing adverse events such as blood backflow and air embolism, and significantly reducing medical risks.

[0048] 2. This invention enables centralized monitoring and visualization of infusion status, eliminating the need for nurses to frequently travel between wards. Nurses can easily monitor the infusion status of the entire ward from the nurses' station, freeing up valuable time for bedside care.

[0049] 3. The system automatically generates and tracks the infusion queue, can verify sequence errors and provide medication preparation prompts, thus technically eliminating infusion sequence errors caused by human negligence.

[0050] 4. This invention dynamically predicts the remaining time based on algorithms such as Kalman filtering, enabling nurses to make forward-looking and planned work arrangements and avoid being overwhelmed. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a schematic diagram of the infusion stand and information acquisition module structure of the centralized monitoring and visualization early warning management system for infusion process provided in an embodiment of the present invention; Figure 2 This is a block diagram of the principle of the centralized monitoring and visualization early warning management system for infusion process provided in this embodiment of the invention; Figure 3 This is a block diagram of the centralized monitoring server provided in an embodiment of the present invention; Figure 4 This is a diagram of a semantic segmentation network architecture based on deep learning provided in an embodiment of the present invention; Figure 5 This is a flowchart of the working method of the infusion process visualization early warning management system based on centralized monitoring provided in an embodiment of the present invention. Detailed Implementation

[0053] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0054] Example 1: System Overall Architecture

[0055] Please refer to Figure 1This diagram illustrates the overall structure of a centralized monitoring-based visualized early warning management system for infusion processes, as provided in an embodiment of the present invention. The system aims to address issues such as monitoring blind spots, response delays, and chaotic processes in traditional infusion management. The system mainly consists of three core components: a front-end data acquisition layer, a cloud / server processing layer, and a user interaction layer.

[0056] The front-end acquisition layer comprises multiple image acquisition terminals 100 deployed within the ward. Each image acquisition terminal 100 is fixedly mounted on an IV stand 200 using a dedicated clamp or integrated bracket, maintaining a relatively fixed position and angle with its corresponding IV bottle or IV bag 300. At the core of each image acquisition terminal 100 is a high-resolution, low-light-performance camera capable of simultaneously acquiring visible light and near-infrared images. To adapt to different ambient light conditions, the terminal also integrates an ambient light sensor 101 and a variable-brightness LED supplementary lighting unit 102. Furthermore, to optimize the shooting angle, the angle between the lens center axis of the image acquisition terminal 100 and the vertical center axis of the IV bottle is set to 0° to 15°, and the distance between the lens and the label area of ​​the IV bottle is 15cm to 40cm. Figure 2 As shown, this installation specification enables the terminal to reliably and clearly capture key image areas containing patient label information and the liquid interface.

[0057] The wireless data transmission module 103 can be a Wi-Fi module, ZigBee module, or NB-IoT module integrated inside the image acquisition terminal 100. It is responsible for transmitting the acquired raw image data or simply compressed image data to the back-end centralized monitoring server 500 in real time and reliably through the hospital's internal wireless network 400.

[0058] The cloud / server processing layer is the centralized monitoring server 500. This server can be a physical server, a virtual machine, or a cloud server instance, possessing powerful computing and storage capabilities. Figure 3 As shown, the centralized monitoring server 500 integrates multiple functional modules internally through software definition: Image receiving module 501: As a network server, it is responsible for listening to and receiving TCP / UDP data streams from all online image acquisition terminals 100, and performing unpacking and preliminary data integrity verification.

[0059] Image preprocessing module 502 performs a series of standardization operations on each received image frame. First, it removes salt-and-pepper noise or Gaussian noise introduced during image acquisition and transmission using median filtering or Gaussian filtering. Second, it enhances image contrast using histogram equalization or adaptive gamma correction algorithms, making label text and liquid surface boundaries clearer. Then, it performs geometric correction for possible lens distortion based on pre-calibrated camera parameters. Finally, based on a preset region of interest (ROI) template that matches the infusion bottle template, it quickly crops out sub-images of the "label information area" and "remaining liquid area" for subsequent processing by dedicated modules.

[0060] Label Information Recognition Module 503: This module takes a "label information region" sub-image as input. First, it applies a text detection algorithm based on Maximum Stable Extreme Region (MSER) ​​or Convolutional Neural Network (CNN) to locate the specific region containing text. Then, it uses an open-source or commercial Optical Character Recognition (OCR) engine (such as Tesseract or PaddleOCR) to convert the text in the image into an editable string. To improve recognition accuracy, the system has a built-in medical dictionary (including common drug names, department names, dosage units, etc.). The extracted key information includes: patient name, hospital number / bed number, drug name, dosage, preparation time, expiration date, and the infusion order clearly specified in the medical order (such as "1 / 3", "2 / 3", "3 / 3" or "BID", "QD", etc.). After recognition, the module interfaces with the hospital's HIS system via the HL7 / FHIR interface to cross-verify the recognized information with the electronic medical order. If a discrepancy is found in the patient, drug, or dosage, a high-level alarm such as "medical order verification error" is immediately generated.

[0061] Liquid level detection module 504: This module is fundamental for achieving accurate monitoring. In a preferred embodiment, this module employs a deep learning-based semantic segmentation network, the principle of which is as follows... Figure 4As shown. The network structure can adopt classic architectures such as U-Net and DeepLabV3+. During the training phase, the researchers collected tens of thousands of images of infusion bottles under different lighting conditions, angles, and sizes (50ml, 100ml, 250ml, 500ml), and labeled the liquid regions in each image at the pixel level (foreground is liquid, background is non-liquid). After training, the network can perform end-to-end inference on the input liquid remaining region sub-image and output a binary mask with the same resolution as the original image, where white pixels represent liquid regions and black pixels represent air regions. Next, the system counts the total number of white pixels S_mask in the mask. Since the geometry of an infusion bottle of the same size is fixed, there is a definite, but potentially non-linear, functional relationship between the liquid volume V and the number of mask pixels S_mask: V = f(S_mask). To obtain an accurate mapping function f, during the system calibration phase, experiments are conducted beforehand for each type of infusion bottle: different volumes of liquid (e.g., 0ml, 50ml, 100ml, ..., full bottle) are filled, and images are captured, calculating the corresponding mask pixel count. Then, multinomial regression (e.g., cubic multinomial) or an RBF neural network is used to fit these experimental data points, resulting in a high-precision nonlinear regression model. In practical applications, the liquid level detection module 504 only needs to substitute the segmented S_mask into this model to quickly and accurately calculate the current remaining liquid volume V_rem.

[0062] Flow rate analysis module 505: This module works closely with the liquid level detection module 504. It is responsible for caching the timestamps t_i and the calculated remaining volume V_rem_i for N consecutive frames (e.g., N=10) of images. For the two most recent frames (i-1 and i), the instantaneous flow rate v_i = (V_rem_{i-1} - V_rem_i) / (t_i - t_{i-1}). Since the instantaneous flow rate may fluctuate due to factors such as liquid level fluctuations and image noise, the module further employs a moving average filter: V_cur = (v_{i-M+1} + v_{i-M+2} + ... + v_i) / M, where M is the size of the sliding window (e.g., M=5). This stable flow rate V_cur represents the current relatively reliable infusion rate. In addition, an abnormal flow rate detection submodule is integrated within the module. This submodule compares V_cur with a configurable safe flow rate range [V_min, V_max]. V_min and V_max can be set by the doctor in the prescription based on factors such as the nature of the drug (e.g., chemotherapy drugs need to be infused slowly, and antibiotics need to be infused within a certain time), the patient's age and weight, etc. If V_cur > V_max, and the duration exceeds the threshold T_abnormal (e.g., 30 seconds), an "overly fast flow alarm" is triggered; if V_cur < V_min, and the duration exceeds the threshold, an "overly slow flow alarm" or a "potential blockage / pause" alarm is triggered.

[0063] The remaining time prediction module 506 calculates the estimated remaining time for the current bottle based on the basic formula T_rem = V_rem / V_cur. However, to overcome the fluctuations in prediction results caused by short-term flow rate variations, this module introduces a Kalman filter. The Kalman filter is an optimal state estimation algorithm that establishes the system's state equation (e.g., assuming a linear relationship between remaining volume and time, with some process noise) and observation equation (the actual measured V_rem). By combining historical information and current observations, it provides an optimal estimate with minimum mean square error. Through smoothing and correction by the Kalman filter, the final output T_rem curve is smoother and more stable, avoiding significant jumps in remaining time caused by instantaneous anomalies in liquid level detection.

[0064] Sequence Management Module 507: This module is key to solving the problem of continuous infusion of multiple vials. It first obtains the patient's complete infusion plan through the infusion sequence input submodule 5071. The input method is very flexible: nurses can manually enter the information through the nurse station terminal interface; a more convenient method is for nurses to scan the barcode / QR code on each vial with a PDA before preparing medication or hanging the vial, and the system automatically identifies and establishes a queue according to the scanning order or the sequence information contained in the barcode; the ideal method is to interface with the HIS system to automatically download long-term or temporary infusion orders issued by doctors and generate a structured infusion queue Q ={B1, B2, ..., Bn}, where each element Bk contains information such as drug name, dosage, and infusion requirements. The Dynamic Reordering submodule 5072 handles common real-world scenarios of sequence reversal. For example, suppose the system expects the next vial to be B2, but image recognition reveals that the nurse has actually hung B3. At this point, the module will not generate a serious error alarm. Instead, it will intelligently and automatically adjust the queue order to {B1, B3, B2, B4...} and recalculate the expected start time and warning time for all subsequent bottles, ensuring that the system logic is consistent with clinical practice. The interruption and recovery management submodule 5073 is used to handle infusion interruptions. For example, if the system detects that the flow rate is 0 and the fluid level has not changed for more than 1 minute, it automatically determines it as an "infusion interruption" and records the interruption start time T_stop. When the flow rate is subsequently detected to be positive again, the system determines it as a "recovery" and dynamically adjusts the expected end time of subsequent bottles based on the interruption duration.

[0065] Early Warning Decision Module 508: This module triggers different levels of early warnings and alerts based on the outputs of the Remaining Time Prediction Module 506 and the Sequence Management Module 507. The system defines three main thresholds: T1 (e.g., 10 minutes), T2 (e.g., 3 minutes), and T3 (0 minutes or air detection). The initial state is normal. When T_rem ≤ T1, it enters the first-level early warning state, and the interface displays a yellow warning. When T_rem ≤ T2, it enters the second-level early warning state, and the interface displays an orange warning and pushes a message. When T_rem ≤ T3 or air is detected, it enters the third-level early warning state, and the interface flashes red and emits a strong audible and visual alarm. Each early warning level is associated with a specific action (e.g., preparing medicine, notification, or calling), and these thresholds and actions can be flexibly configured through the backend management interface.

[0066] The user interaction layer primarily refers to the nurse station visualization terminal 600. This terminal can be a high-performance industrial control computer or a regular PC, connected to a large-size high-definition monitor. It maintains real-time communication with the centralized monitoring server 500 via WebSocket or HTTP protocol. The terminal displays an overview of the infusion status of all beds in a graphical interface. Each patient corresponds to a card or icon, which intuitively displays: bed number (e.g., "Bed 12"), patient name, current medication name and dosage, remaining volume (in both numerical and analog level bars), estimated remaining time (countdown format), next medication name, current warning level (indicated by different colors and icons), and nurse response status (e.g., "Notified" or "Confirming"). The interface also supports global sorting (sorted by remaining time or warning level) to help nurses prioritize the most urgent tasks. In addition, the nurse station visualization terminal 600 also includes a historical data traceability module, connected to a time-series database (e.g., InfluxDB) or relational database, recording detailed logs of each infusion event. Nurses can use this module to query any patient's historical infusion records, view flow rate change curves, calculate average dressing change response time, and even generate daily / weekly / monthly nursing quality reports to provide data support for management decisions.

[0067] Example 2: Detailed Visual Early Warning Management Method for Infusion Process

[0068] Please refer to Figure 5 This document illustrates the main flowchart of a visualized early warning management method for infusion process based on centralized monitoring, provided by an embodiment of the present invention. The method, based on the system described in Embodiment 1, includes the following steps: Step S1: Image acquisition.

[0069] After the nurse attaches the first IV bag to the patient and starts the system, the corresponding image acquisition terminal 100 begins to work. The terminal continuously captures images of the IV bottle or IV bag at a fixed frame rate (e.g., 1 frame / second during normal operation, automatically increasing to 5 frames / second when entering the alarm state).

[0070] Step S2: Image upload.

[0071] The terminal's built-in wireless module 103 encapsulates the collected JPEG or PNG compressed images into data packets and uploads them in real time to the designated IP address and port of the centralized monitoring server 500 via a Wi-Fi network.

[0072] Step S3: Image preprocessing and region of interest extraction.

[0073] After receiving the data packet, the image receiving module 501 on the server side restores it to an image matrix. Subsequently, the image preprocessing module 502 performs denoising, grayscale conversion, contrast enhancement, and geometric correction on the image. Finally, using a shape-matching-based template matching algorithm, two sub-images, the "label information area" and the "remaining liquid area," are quickly located and cropped.

[0074] Step S4: Tag information identification and verification.

[0075] After receiving the "label information area" sub-image, the label information recognition module 503 performs OCR recognition. Assume the recognition result contains the patient's name. The system displays the patient's name, bed number "08", medication "levofloxacin sodium chloride injection", dosage "100ml". The system immediately compares this information with the current valid medical order for patient 08 retrieved from the HIS system. If the comparison matches, the process proceeds normally; if the comparison reveals that the order for that period was "normal saline", the system determines it as a "medication mismatch", immediately issues a red "medication order verification error" alarm on the nurse station terminal, and suspends monitoring of that bottle, awaiting nurse intervention.

[0076] Step S5: Liquid level detection and volume calculation.

[0077] The liquid level detection module 504 processes the "remaining liquid area" sub-image. In this embodiment, for standard glass or plastic infusion bottles, the Canny edge detection algorithm is used to extract all edges in the image. Then, Hough transform is used to identify vertical and horizontal straight lines. Combined with the prior shape of the bottle, the crescent-shaped liquid surface boundary is located. Finally, a region growing algorithm is used to fill downwards from the liquid surface boundary to obtain the precise contour of the liquid area. The pixel area within this contour is calculated and substituted into a pre-calibrated "area-volume" mapping function for this specification of infusion bottle (e.g., for a 250ml round bottle, the function is: V = 0.086). S - 2.3 (unit ml, where S is the pixel area), to obtain the current remaining volume V_rem.

[0078] Step S6: Flow rate calculation and smoothing.

[0079] The flow rate analysis module 505 records the system time t_cur and volume V_rem_cur during the current calculation, and reads the previously recorded time t_prev and volume V_rem_prev. The instantaneous flow rate v_inst is calculated as (V_rem_prev - V_rem_cur) / (t_cur - t_prev), in ml / s. The system maintains a circular queue of length 10 to store the most recent v_inst. The stable flow rate V_cur is equal to the arithmetic mean of all non-zero values ​​in the queue. Simultaneously, the abnormal flow rate detection submodule checks whether V_cur is within the doctor's preset safe range; if it exceeds this range, a corresponding alarm is triggered.

[0080] Step S7: Remaining time prediction and correction.

[0081] The remaining time prediction module 506 first calculates the basic prediction value T_rem_raw = V_rem_cur / V_cur. Then, this value is used as the input to the Kalman filter. The Kalman filter internally maintains a state vector (containing the estimated remaining volume and velocity), and through prediction and update steps, outputs an optimal, smooth remaining time estimate T_rem.

[0082] Step S8: Infusion queue generation and tracking.

[0083] The sequence management module 507 is responsible for this step. During the initial infusion, the module has already generated the patient's complete infusion queue via scanning or HIS import. For example, the queue for patient in bed 08 is: [1. Levofloxacin 100ml, 2. Metronidazole 100ml, 3. Vitamin C 50ml]. The module has an internal pointer that initially points to the first element. The system continuously analyzes the label information to confirm that the currently infused medication matches the medication pointed to by the pointer. Once one bottle of infusion is completed, the pointer automatically moves to the next element.

[0084] Step S9: Multi-level early warning and medication preparation reminders.

[0085] The early warning decision module 508 obtains T_rem from step S7 and queue information from step S8. Assume the system's configured early warning thresholds are: T1 = 10 minutes, T2 = 3 minutes, T3 = 0 minutes. Currently, the patient in bed 08 has 100ml of levofloxacin remaining at a flow rate of 5ml / min, so T_rem = 20 minutes, which is greater than T1, and no early warning is issued. When T_rem drops to 8 minutes, T_rem ≤ T1, triggering a Level 1 early warning. The display card for bed 08 on the nurse station terminal turns yellow, and the text message "Prepare the next bottle of medication: Metronidazole" appears. When T_rem drops to 2 minutes, a Level 2 early warning is triggered; the card turns orange, and a notification is pushed to the responsible nurse's PDA. When T_rem drops to 0, the system confirms the fluid level has disappeared by analyzing the image, triggering a Level 3 early warning; the card flashes red, and a continuous alarm sound plays at the nurse station.

[0086] Step S10: Visualization.

[0087] The nurse station's visual terminal 600 receives real-time status data (bed number, remaining time, warning level, etc.) for each patient from the server and dynamically updates and displays it in the form of dashboards, progress bars, Gantt charts, etc. Nurses can have a clear understanding of the infusion status of the entire ward without leaving their seats.

[0088] Example 3: Optimized Liquid Level Detection Method for Flexible Infusion Bags

[0089] For flexible infusion bags (especially PVC or non-PVC soft bags), their shape deforms as the liquid volume decreases (bottom shrinks, sidewalls concave), and traditional methods based on edge detection and fixed geometric parameters have significant errors. Therefore, the present invention provides another preferred embodiment.

[0090] In this embodiment, the liquid level detection method in step S5 is replaced by a keypoint detection method based on deep learning. Specifically: Network selection: A lightweight keypoint detection network is used, such as a simplified version of the Stacked Hourglass Network, or a lightweight pose estimation network specifically designed for mobile devices (such as a simplified version of BlazePose). The network's output layer is modified to regress the coordinates of two keypoints: the left liquid surface point P_left(x1, y1) and the right liquid surface point P_right(x2, y2).

[0091] Training data preparation: Collect a large number of images of infusion bags of different sizes (e.g., 500ml, 1000ml), brands, and liquid levels (from full to empty) in a suspended state. For each image, manually label the coordinates of the intersection points between the liquid surface and the left and right boundaries of the bag. Simultaneously, record the actual liquid volume inside the bag at that moment. This is a typical training set for a regression problem.

[0092] Deformation Compensation Model: The coordinates of two keypoints output by the keypoint detection network can be used to calculate the pixel height h_pixel of the midpoint of the liquid surface. However, due to bag deformation, the relationship between h_pixel and the actual volume is not a simple linear one. Therefore, we establish a deformation compensation model. One effective method is to train another small neural network (such as a 3-layer MLP) whose input is the liquid surface height h_pixel and the bag width w (which can be calculated from the difference in the x-coordinates of P_left and P_right), and whose output is the predicted remaining volume V_pred. Alternatively, a simpler method is online calibration: before the infusion begins, the system acquires images of a full and empty bottle (or from the previous few frames), calculates the correspondence between the liquid surface height and volume at this time, forming a personalized linear or polynomial mapping.

[0093] Example 4: Closed-loop confirmation process after infusion completion

[0094] To create a closed-loop management system and ensure that every dressing change is confirmed, thus avoiding safety issues caused by system misjudgments or nurse oversights, the method of this invention also includes a closed-loop confirmation step: S11: Automatically determines when the infusion is complete.

[0095] The liquid level detection module 504 continuously monitors the remaining volume. When V_rem is less than the preset residual liquid threshold (e.g., 5ml, configurable), or for hard bottles, when the liquid level is detected to be below the bottleneck position, the system determines that the infusion of the current bottle is complete.

[0096] S12: Send a notification to change medication.

[0097] The early warning decision module 508 generates a "medication change request" event. The system sends a notification to two targets through the built-in message push service (such as MQTT): first, the nurse station visualization terminal 600 displays a "medication change request" icon in the corresponding patient area; second, it pushes the notification to the mobile terminal (PDA or mobile APP) of the nurse in charge of the patient, which vibrates and makes a prompt sound, and displays "Please change the next bottle of medication for bed 08: metronidazole".

[0098] S13: Nurse confirms dressing change.

[0099] After receiving the notification, the nurse goes to the patient's bedside and performs the dressing change. After completing the dressing change, the nurse needs to confirm. The system supports two confirmation methods: Method 1 (Automatic Confirmation): The nurse uses a PDA or barcode scanner to scan the barcode on the newly installed IV bag. After scanning, the system automatically compares the barcode information with the information in queue B(k+1) in step S8. If they match, the confirmation is successful.

[0100] Method 2 (Manual Confirmation): The nurse directly locates the patient in bed 08 on the 600 terminal interface at the nursing station and clicks the "Dressing Change Confirmation" button. A pop-up window will request the nurse's ID and password or fingerprint confirmation to ensure the traceability of the operation.

[0101] S14: System status update.

[0102] After receiving the confirmation signal, the server performs a series of internal state update operations: Move the queue pointer in the sequence management module 507 from Bk to B(k+1).

[0103] Reset the Kalman filter state of the remaining time prediction module 506 so that it recalculates based on the new bottle.

[0104] Record key information about this dressing change in the database: dressing change time, nurse ID, and response time (the time difference between sending the notification in S12 and confirming it in S13).

[0105] S15: Timeout without confirmation of upgrade alarm.

[0106] Simultaneously with sending the notification in S12, the system starts a timer. If no confirmation signal is received within the preset waiting time (e.g., 5 minutes for general wards, 2 minutes for ICU), the system determines that an anomaly may have occurred (nurse has left, PDA has run out of power, or the system has been delayed by other matters). At this point, the system automatically escalates the alert from Level 2 (or Level 1) to Level 3 emergency alarm. In addition to the interface and sound upgrades, the system will also automatically call the nurses' station supervisor via the hospital's internal call system or broadcast an emergency assistance request to the mobile terminals of other nurses in the same group, ensuring everything goes smoothly.

[0107] Furthermore, embodiments of the present invention also provide a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the centralized monitoring-based visualized early warning management method for infusion process described in any of the above embodiments. The computer-readable storage medium can be any medium capable of storing program code, such as a USB flash drive, external hard drive, ROM, RAM, magnetic disk, or optical disk.

[0108] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0109] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A centralized monitoring-based infusion progress visualization early warning management system, characterized in that, include: Multiple image acquisition terminals are provided, each of which is set at a fixed position on an infusion stand. They are used to acquire visible light or infrared images of the infusion bottle or bag in real time during the infusion process. The images include at least the label information area of ​​the infusion bottle or bag and the remaining liquid area. The wireless data transmission module is connected to the image acquisition terminal and is used to wirelessly upload the acquired image data to the centralized monitoring server in real time. Centralized monitoring server, including: The image receiving module is used to receive image data from multiple image acquisition terminals; The image preprocessing module is used to denoise, enhance, correct, and extract regions of interest from the received image; The label information recognition module uses an optical character recognition algorithm to extract patient identity information, drug name, drug dosage, preparation time, and preset infusion sequence from the label information area; The liquid level detection module uses image segmentation and edge detection algorithms to analyze the remaining liquid area and calculate the remaining liquid volume or remaining liquid level height in the current infusion bottle or infusion bag. The flow rate analysis module calculates the current infusion flow rate in real time based on the change in liquid level and the corresponding time interval in multiple consecutive frames of images. The remaining time prediction module dynamically predicts the estimated completion time of the current infusion bottle or bag based on the remaining liquid volume and the current flow rate. The sequence management module automatically generates an infusion sequence queue for each patient based on the preset infusion order and label information recognition results, and tracks in real time the infusion bottle or bag currently being infused and the replacement status of subsequent infusion bottles or bags. The early warning decision module automatically generates a replacement early warning signal for the infusion bottle or bag when it predicts that the remaining time of the current infusion bottle or bag is lower than a preset threshold, and determines the preparation prompt for the next infusion bottle or bag according to the sequence management module. The nurse station visualization terminal communicates with the centralized monitoring server and is used to display an overview of the infusion status of all patients in a graphical interface. The infusion status includes: patient bed number / identifier, medication information of the current infusion bottle or bag, remaining volume, remaining time, medication information of the next infusion bottle or bag, warning status, and medication preparation status.

2. The system of claim 1, wherein, The image acquisition terminal also includes: The supplementary lighting unit and the ambient light sensor are provided. The supplementary lighting unit automatically adjusts the brightness of the supplementary lighting according to the light intensity detected by the ambient light sensor.

3. The system of claim 1, wherein, The liquid level detection module uses a deep learning-based semantic segmentation network to perform pixel-level segmentation of the liquid region in the infusion bottle image. The output of the semantic segmentation network is a binary mask of the liquid region, and the remaining liquid volume is calculated through a pre-established mapping function between the liquid volume and the number of mask pixels. The mapping function is a nonlinear regression model obtained through calibration experiments on infusion bottles of different specifications and shapes.

4. The system of claim 1, wherein, The flow rate analysis module further includes: The abnormal flow rate detection submodule is used to compare the real-time calculated infusion flow rate with the preset safe flow rate range. When the flow rate exceeds the safe range for a period of time exceeding the preset duration threshold, an abnormal flow rate alarm signal is generated. The abnormal flow rate alarm signal includes an alarm for excessively fast flow rate and an alarm for excessively slow flow rate.

5. The system of claim 1, wherein, The sequence management module supports intelligent sorting management of multiple bottles of continuous infusion, specifically including: The infusion sequence input submodule is used to input infusion sequence information, and supports three methods: manual input, import by scanning barcode or QR code, and automatic acquisition from the hospital information system. The dynamic reordering submodule is used to automatically update the queue and recalculate the warning time for all subsequent infusion bottles or bags when the actual infusion order is detected to be inconsistent with the preset order. The interruption and recovery management submodule is used to automatically record the interruption time when it is detected that a bottle of infusion has been paused or removed, and to recalculate the remaining time and flow rate after the infusion is resumed.

6. The system of claim 1, wherein, The early warning decision module adopts a multi-level early warning strategy, including: Level 1 warning: When less than 10 minutes remain, the patient will be displayed in yellow on the visual terminal at the nurse station, and a text prompt to prepare the next bottle of medication will be generated. Level 2 warning: When the remaining time is less than 3 minutes, it will be displayed in orange and an audio prompt will be issued on the nurse station terminal. At the same time, a medication change notification will be pushed to the corresponding nurse's mobile terminal. Level 3 warning: When the infusion is finished or air is detected entering the infusion tube, a red flashing indicator will be displayed and a continuous audible alarm will be emitted, while the corresponding responsible nurse will be automatically called.

7. The system of claim 1, wherein, The nurse station visualization terminal also includes: The historical data traceability module is used to record and display complete data for each patient's infusion, including: infusion start time, time of each bottle change, actual flow rate curve, early warning trigger record, nurse response time, and dressing change operation confirmation record. The historical data traceability module supports retrieval and statistical analysis by time range, patient, drug name, or nurse name, and automatically generates infusion nursing quality reports.

8. A method for infusion process visualization and early warning management based on centralized monitoring, using the system of any one of claims 1 to 7, characterized in that, Includes the following steps: S1: Multiple image acquisition terminals installed on the infusion stand acquire image data of the corresponding infusion bottle or infusion bag in real time. The image data includes the label information area and the liquid remaining area. S2: The collected image data is uploaded to the centralized monitoring server in real time via wireless network; S3: The received image is denoised, grayscale converted, contrast enhanced and geometrically corrected in the centralized monitoring server, and the label information region sub-image and the liquid remaining region sub-image are extracted based on the preset region of interest template. S4: Use optical character recognition algorithm to perform text recognition on the label information area sub-image, extract patient identification, drug name, drug dosage and infusion sequence information, and cross-verify with the medical orders in the hospital information system. If the information is inconsistent, a verification alarm will be issued. S5: The edge detection algorithm and region growing algorithm are used to process the liquid remaining region sub-image, identify the boundary line between liquid and air, and calculate the current remaining liquid volume V_rem in combination with the pre-calibrated infusion bottle geometric parameters; S6: Based on the time interval Δt between two consecutive frames and the corresponding change in remaining volume ΔV, calculate the current instantaneous flow velocity as v = ΔV / Δt, and obtain the stable flow velocity V_cur through moving average filtering; S7: Based on the current remaining volume V_rem and the steady flow rate V_cur, calculate the expected remaining time T_rem = V_rem / V_cur, and use a Kalman filter to smooth and dynamically correct the prediction results; S8: Based on the infusion sequence extracted from the tag information, automatically generate or update the patient's infusion queue Q = {B1,B2, …, Bn}, and mark the position of the currently infused bottle Bk in the queue; S9: Compare the estimated remaining time T_rem with the preset multi-level warning thresholds T1, T2, and T3. When T_rem ≤ T3, generate a level 3 warning; when T_rem ≤ T2 and greater than T3, generate a level 2 warning; when T_rem ≤ T1 and greater than T2, generate a level 1 warning. At the same time, generate a drug preparation prompt based on the drug information of the next bottle B(k+1) in the queue. S10: The infusion status of all patients is displayed in real time on the nursing station's visual terminal in the form of a dashboard, progress bar, or Gantt chart, including patient identification, current medication, remaining volume, remaining time, next bottle of medication, warning level, and nurse response status.

9. The method of claim 8, wherein, In step S5, the liquid level detection step, for the case of using a flexible infusion bag, the edge detection algorithm is replaced by a key point detection network based on deep learning, which directly regresses and outputs the coordinates of two key points of the liquid surface boundary inside the infusion bag, and then calculates the liquid level height according to the pre-established deformation compensation model. The deformation compensation model considers the nonlinear relationship between the bottom tensile deformation caused by gravity after the infusion bag is suspended and the actual volume of the liquid surface.

10. The method of claim 8, wherein, The method also includes a closed-loop confirmation step after the infusion is completed: S11: When the system detects that the remaining volume of the current infusion bottle or infusion bag is less than the preset residual liquid threshold or the liquid level is lower than the neck of the infusion bottle, it automatically determines that the infusion of the bottle is completed. S12: The system sends a notification for dressing change to the nursing station terminal and the mobile terminal of the corresponding responsible nurse; S13: After completing the dressing change, the nurse confirms the dressing change by scanning the barcode of the next bottle of medication or by clicking the confirmation button on the nurse station terminal interface. S14: After receiving the dressing change confirmation signal, the system automatically moves the infusion queue pointer to the next bottle, resets the remaining time and flow rate calculation, and records the response time and operator information for this dressing change operation. S15: If no confirmation signal for medication change is received within the preset waiting time after sending the medication change notification, the system will automatically escalate to a Level 3 emergency alarm and call the shift leader or a nearby nurse for assistance.

Images