Intelligent ward-oriented medical instrument internet of things sensing and intelligent scheduling system

Abstract

The application discloses a smart ward-oriented medical instrument internet of things sensing and intelligent scheduling system, which comprises five modules of multi-modal sensing, internet of things gateway, cloud edge processing, intelligent scheduling and visual interaction; the multi-modal sensing module collects valid sensing data of the medical instrument, and after preprocessing by the internet of things gateway, realizes nearby real-time processing and full-amount data modeling by the cloud edge processing module; the intelligent scheduling module outputs optimal instructions based on the model and clinical priority; and the visual interaction module realizes state display, instruction sending and receiving and manual intervention. The application realizes intelligent management of the whole life cycle of the medical instrument, improves scheduling accuracy and efficiency, guarantees clinical diagnosis and treatment requirements, and is suitable for the smart ward scene.

Description

Technical Field

[0001] This invention belongs to the field of medical device and medical Internet of Things technology, specifically a medical device Internet of Things sensing and intelligent scheduling system for smart wards. Background Technology

[0002] With the advancement of medical informatization and the construction of smart wards, medical devices, as core resources for clinical diagnosis and treatment, directly impact the smoothness of the treatment process and patient safety through their management efficiency and accuracy. Currently, smart wards have an increasingly urgent need for real-time location tracking, status monitoring, dynamic scheduling, and full lifecycle management of medical devices, necessitating the intelligent upgrade of device management through technologies such as the Internet of Things and big data.

[0003] In existing technologies, medical device management mostly relies on manual registration and statistics or single-dimensional sensing solutions. Some systems use IoT technology to collect data on device location or usage status, but these are mostly limited to single-point data collection and lack multi-dimensional data fusion and verification. Data processing mostly relies on centralized cloud computing, which results in transmission delays. Scheduling decisions are mostly based on experience-based allocation and do not combine the device's full life cycle status with clinical needs for intelligent coordination.

[0004] The aforementioned existing technologies have significant shortcomings: First, they have a single perception dimension, resulting in insufficient data accuracy and completeness, and are prone to abnormal data interference; second, data processing and scheduling response are lagging, failing to meet the needs of high-frequency data real-time processing and emergency diagnosis and treatment scenarios; and third, they lack full-process closed-loop management, making it difficult to accurately track equipment disinfection, maintenance, and other aspects, resulting in low scheduling efficiency and resource conflicts, and making it difficult to adapt to the refined and intelligent management needs of smart wards. Summary of the Invention

[0005] To address the shortcomings of the existing technologies, the present invention aims to provide a medical device IoT sensing and intelligent scheduling system for smart wards. This system solves the technical problems of existing technologies, such as the single sensing dimension of medical devices, insufficient data accuracy and completeness, abnormal data easily interfering with decision-making, data processing relying on centralized cloud computing leading to response delays and inability to meet the real-time processing needs of high-frequency data, lack of closed-loop management throughout the entire life cycle, inaccurate tracking of disinfection and maintenance processes, and inefficient resource allocation and potential conflicts of use due to scheduling not being combined with clinical needs priority.

[0006] To achieve the above objectives, embodiments of the present invention disclose a medical device Internet of Things (IoT) sensing and intelligent scheduling system for smart wards, the system comprising: a multimodal sensing module, an IoT gateway module, a cloud-edge processing module, an intelligent scheduling module, and a visual interaction module; The multimodal sensing module is deployed on the medical device body, synchronously collects the sensing data of the device and uploads it to the Internet of Things gateway module, and obtains effective sensing data by cross-comparison and verification of sensor data and elimination of abnormal data. The IoT gateway module and the multimodal sensing module establish a connection via wireless communication, aggregate the effective sensing data, complete protocol conversion and data preprocessing, obtain new sensing data, and interconnect with the cloud-edge processing module. The cloud-edge processing module includes edge computing nodes and a cloud server; the edge computing nodes process frequently occurring new sensing data after aggregation; the cloud server is used to store the new sensing data and to build a full life cycle model of medical devices based on the new sensing data using machine learning algorithms. The intelligent scheduling module is deployed on the cloud server. Based on the medical device life cycle model and clinical demand priority, it realizes dynamic tracking of devices, prediction and coordination of usage conflicts, closed-loop management of disinfection process and dispatch of maintenance reminders, and generates and outputs the optimal scheduling instructions. The visualization interaction module includes a nurse station terminal, a mobile medical terminal, and a device control interface, which is used to present the full-dimensional status of medical devices, send and receive scheduling instructions, and provide feedback on execution results.

[0007] Furthermore, the multimodal sensing module includes: a positioning sensor, a status monitoring sensor, a disinfection detection sensor, and a maintenance timing unit; the positioning sensor uses UWB ultra-wideband positioning technology to collect the real-time position of the instrument; the status monitoring sensor identifies the instrument's use or idle status through current and pressure feedback, and the data collected by each sensor are cross-checked to eliminate abnormal data.

[0008] Furthermore, the multimodal sensing module uploads sensing data using a combination of periodic and triggered uploads. Under normal conditions, data is uploaded at preset time intervals, while triggered uploads are used when a sudden change in the device's state is detected.

[0009] Furthermore, the wireless communication supports WiFi, Bluetooth and LoRa wireless communication protocol adaptation. During the protocol conversion process, the effective sensed data is uniformly converted into the MQTT protocol format. Data preprocessing includes noise filtering, data format standardization and redundant data removal.

[0010] Furthermore, the edge computing node performs localized processing on the frequently occurring new sensing data after aggregation, specifically including: real-time data noise reduction, abnormal operating condition threshold judgment, and instant feedback signal generation; the real-time data noise reduction uses an adaptive filtering algorithm to remove environmental interference signals; the abnormal operating condition threshold judgment and instant feedback signal generation are based on setting threshold ranges according to the standard operating parameters of different types of medical devices, and when the threshold range is exceeded, an abnormal warning signal is generated and instantly fed back to the IoT gateway module and the visualization interaction module; the machine learning algorithm includes neural network algorithm and decision tree algorithm.

[0011] Furthermore, the neural network algorithm adopts a fusion model of convolutional neural network and long short-term memory network, and is trained based on new sensing data to extract data features and predict the risk of device failure, remaining service life and maintenance requirement cycle; the decision tree algorithm adopts the C4.5 decision tree algorithm to classify and make decisions on the process node data of device procurement and warehousing, clinical use, cleaning and disinfection, maintenance and repair and scrapping and recycling, and classify the health level of the device.

[0012] Furthermore, the priority of clinical needs in the intelligent scheduling module is calculated based on a weighted average of three indicators: the urgency of the patient's condition, the timeline of the treatment plan, and the department's equipment quota; the conflict prediction and coordination uses a greedy algorithm to allocate equipment resources. Furthermore, the intelligent scheduling module supports tiered reminders for maintenance reminders, pushing reminder information to the equipment administrator and head nurse respectively based on the remaining time of the maintenance cycle. The reminder methods include terminal pop-ups, SMS, and voice prompts.

[0013] Furthermore, the nurse station terminal adopts a large-screen display mode, showing the distribution of ward equipment, status statistics, and scheduling task progress in sections; the mobile medical terminal supports handheld operation and pushes scheduling instructions and abnormal reminders in real time; the equipment control interface is linked with the medical device electronic control unit to realize remote start and stop control of some equipment.

[0014] Furthermore, the visual interaction module supports manual intervention in the scheduling process by medical staff. The manual intervention permissions are set according to job level. Nurses are responsible for adjusting the temporary use allocation of equipment, and head nurses are responsible for modifying the priority weight of clinical needs and scheduling rule parameters. The full-dimensional status of the medical device includes: device location, real-time use, cleaning and disinfection, maintenance cycle information, abnormal operating conditions, and health level.

[0015] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention integrates multiple types of sensors such as positioning, status monitoring and disinfection detection through a multimodal sensing module, and combines a sensor data cross-comparison and verification mechanism to synchronously collect full-dimensional data such as device location and usage status and remove abnormal data. This solves the problems of single sensing dimension and insufficient data accuracy in the prior art, and realizes synchronous collection and accurate verification of multi-source data of medical devices, providing reliable data support for subsequent management and scheduling.

[0016] (2) The present invention adopts a cloud-edge collaborative processing architecture. Edge computing nodes process high-frequency new sensing data nearby to achieve real-time early warning. The cloud uses a convolutional neural network and decision tree fusion algorithm to build a full life cycle model, which can reduce data transmission delay and accurately predict the risk of device failure and maintenance needs. It breaks through the limitations of existing technologies that rely on centralized cloud computing, resulting in delayed response and lack of full-process prediction, and realizes real-time and intelligent management of devices.

[0017] (3) This invention proposes a collaborative mode of intelligent algorithm scheduling and hierarchical permission intervention. It calculates priority by weighting multiple indicators of clinical needs and coordinates resource conflicts by greedy algorithm. At the same time, it sets manual intervention permissions according to job positions. This not only ensures the scientific and efficient nature of scheduling decisions, but also takes into account the flexibility of clinical scenarios. It solves the problems of existing technologies such as lack of priority basis for scheduling, frequent resource conflicts and lack of rules for manual intervention, and adapts to the refined management needs of smart wards. Attached Figure Description

[0018] Figure 1 This is a structural diagram of the system of the present invention; Figure 2 This is a diagram of the algorithm architecture for the medical device lifecycle model of the present invention. Figure 3 This is a schematic diagram of the terminal interface of the visual interactive module of the present invention. Detailed Implementation

[0019] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 , Figure 2 and Figure 3 This application provides a detailed description of the technical solutions provided in each embodiment.

[0021] like Figure 1As shown in the embodiment of this application, a medical device IoT sensing and intelligent scheduling system for smart wards is provided, including: a multimodal sensing module, an IoT gateway module, a cloud-edge processing module, an intelligent scheduling module, and a visualization interaction module; The perception layer consists of multimodal perception modules deployed on each medical device, responsible for collecting raw data from all dimensions of the device and performing preliminary verification. The gateway layer consists of IoT gateway modules, undertaking data aggregation, protocol conversion, and preprocessing functions, and serving as a communication bridge between the perception layer and the processing layer. The processing layer is the cloud-edge processing module, containing edge computing nodes and cloud servers. The scheduling layer is the intelligent scheduling module, which generates optimal scheduling instructions based on the models and data output from the processing layer and combined with clinical needs. The interaction layer is the visual interaction module, which realizes status display, instruction sending and receiving, and manual intervention through various terminal devices, serving as the interaction interface between the system and medical staff.

[0022] The multimodal sensing module, serving as the core of the system's data acquisition, is deployed in key locations on the medical device itself. Based on the structure and usage scenarios of different devices, it adopts a modular and lightweight design to ensure that the original functions and ease of operation of the device are not affected. The module as a whole employs a low-power design, with two power supply methods: for continuously powered devices such as ventilators and monitors, it uses the device's built-in power supply; for portable devices such as infusion pumps and mobile ultrasound machines, it uses a rechargeable lithium battery.

[0023] In this embodiment, a UWB (Ultra-Wideband) positioning sensor is used. The sensor's installation position is adjusted according to the type of device: for large devices, such as ventilators and ultrasound diagnostic instruments, it is installed at the top center of the device to ensure unobstructed signal transmission; for small devices, such as infusion pumps and syringe pumps, it is installed on the upper side of the device to avoid interference with the control panel. The positioning sensor operates at a frequency of 3.5GHz-6.5GHz, has a communication distance of up to 100m, and a data update frequency of 10Hz to ensure real-time tracking of the device's position.

[0024] To ensure accurate positioning data, four UWB positioning base stations were deployed on the floor where the smart ward is located, installed in the ceiling at each of the four corners of the floor. The base stations are connected to an IoT gateway module via Ethernet to form a positioning network. The positioning sensors receive signals from multiple base stations and use a Time-of-Flight (TOF) algorithm to calculate their own coordinates, thereby determining the real-time position of the equipment. The coordinate data format is (X, Y, Z), where the Z-axis coordinate is used to distinguish floors.

[0025] In this embodiment, the status monitoring sensor is used to identify the use or idle status of the device. Based on the working principle of the device, a combination of current sensor and pressure sensor is adopted to ensure that the status monitoring needs of different types of devices are covered.

[0026] For electrically controlled medical devices, such as ventilators and infusion pumps, the ACS712ELCTR-05B current sensor is used and installed in the power input circuit of the device to determine its status by detecting the operating current.

[0027] For manually operated devices, such as defibrillators and emergency carts, an FSR402 thin-film pressure sensor is used and installed on key operating parts of the device, such as the defibrillator's operating handle and the emergency cart's drawer handle. When a pressure value of 0.5 kg or more is detected and the duration is 1 second or more, the device is determined to be in use; when the pressure value is less than 0.5 kg and the duration is 30 seconds or more, the device is determined to be in idle state.

[0028] In this embodiment, the disinfection detection sensor is used to monitor the cleaning and disinfection status of medical devices. Based on the disinfection methods commonly used in smart wards, such as ultraviolet disinfection, alcohol disinfection, and chlorine-containing disinfectant disinfection, a multi-parameter integrated sensor is adopted.

[0029] In this embodiment, the maintenance timing unit records the usage time and last maintenance time of the medical device, providing data support for maintenance cycle reminders. It is implemented using a combination of a real-time clock (RTC) module and flash memory. The RTC module uses a DS3231 clock chip. The maintenance timing unit calculates the remaining duration of the maintenance cycle by recording the cumulative usage time of the device and the timestamp of the last maintenance.

[0030] To eliminate outlier data and ensure the accuracy of the sensed data, the multimodal sensing module employs a cross-referencing verification mechanism for sensor data. The specific process is as follows: (1) Verification of multiple sensors in the same state: For the instrument in use state, cross-verification is performed by combining the data from the current sensor and the pressure sensor.

[0031] (2) Positioning data consistency verification: The positioning sensor collects position data every 10 seconds. The core control unit compares the coordinate difference between two adjacent collections. If the difference exceeds the normal movement speed of the instrument, it is judged as abnormal data. The previous valid data is used to replace it, and the number of abnormalities is recorded. If three consecutive abnormalities occur, the positioning sensor fault reminder is triggered.

[0032] (3) Disinfection data logic verification: After the disinfection detection sensor collects the disinfection signal, it is verified in conjunction with the instrument's usage status. If the instrument is in use, the disinfection data is determined to be invalid; if it is idle, the disinfection data is confirmed to be valid, and the disinfection information is recorded.

[0033] The multimodal sensing module employs a combination of periodic and triggered data uploads. Periodic uploads, under normal conditions, upload data at preset time intervals, configurable via a visual interaction module. The default settings are: positioning data every 10 seconds, status monitoring data every 5 seconds, disinfection detection data every minute, and maintenance timing data every 30 seconds. Periodic uploads utilize batch transmission, packaging data from multiple sensors for upload to reduce power consumption. Triggered uploads are triggered immediately upon detecting a sudden change in device status. Conditions for determining a sudden change in status include the device switching from idle to active status or vice versa. During data upload, the core control unit formats the data, encapsulating it in JSON format, including fields such as data type, acquisition timestamp, device ID, sensor ID, data value, and checksum, ensuring data transmission integrity and traceability.

[0034] The IoT gateway module, as the core of the system for data aggregation and forwarding, adopts an industrial-grade design to adapt to the complex environment of a smart ward. The IoT gateway module supports multiple wireless communication protocols, including WiFi, Bluetooth, and LoRa, and uses a software-level protocol stack to parse and encapsulate different protocols. For valid sensing data uploaded by the multimodal sensing module, the gateway first identifies the communication protocol type of the data, and then parses it using the corresponding protocol stack to extract the valid sensing data.

[0035] To achieve interconnection between the effectively sensed data and the backend cloud-edge processing modules, all effectively sensed data is uniformly converted to the MQTT (Message Queuing Telemetry Transport) protocol format during the protocol conversion process. The MQTT protocol is lightweight and has low bandwidth consumption, making it suitable for data transmission in IoT devices. The specific conversion process is as follows: (1) Map the fields in the JSON format data uploaded by the multimodal perception module to the payload fields of the MQTT protocol to ensure that the fields correspond one-to-one.

[0036] (2) Define MQTT topics in the format of smart ward / department ID / device ID / data type, such as smart ward / internal medicine ward 3 / ventilator 001 / location data, so that the cloud-edge processing module can subscribe to data by topic and realize the classification and reception of data.

[0037] (3) Set different QoS levels according to the importance of the data. Set QoS 1 for data with high real-time requirements such as location data and status data, and set QoS 0 for disinfection data and maintenance data to ensure that critical data is not lost.

[0038] Data preprocessing is one of the core functions of the IoT gateway module. Its purpose is to remove invalid data and standardize data format, so as to provide high-quality data for subsequent analysis and modeling of cloud-edge processing modules. The preprocessing process includes three steps: noise filtering, data format standardization, and redundant data removal. The noise filtering is performed on random noise in the valid sensed data using a moving average filtering algorithm. For numerical data, the average of five consecutive data collections is taken as valid data, filtering out noise data from single sudden changes. For state-related data, a median filtering algorithm is used, taking the median of three consecutive data collections as valid data to avoid misjudgments of state caused by transient interference.

[0039] The data format standardization involves converting data uploaded from different sensors and using different protocols into a standardized format. Numerical data is uniformly retained to two decimal places, and units are based on international standard units. Timestamps are uniformly formatted using UTC time to ensure time synchronization. Status data is represented by enumerated values, such as usage status = 1, idle status = 0, standby status = 2, disinfection executed = 1, not executed = 0. Device IDs and sensor IDs are formatted as 16-bit strings to ensure uniqueness.

[0040] The redundant data removal process involves the gateway judging duplicate uploaded data based on the data's timestamp and checksum. If two data entries from the same device and of the same data type have a timestamp difference of ≤1 second and the checksums are identical, they are considered redundant data, and only the first valid data entry is retained. Simultaneously, for data with the same status collected multiple times consecutively, the gateway reduces the upload frequency to half of the original, decreasing data transmission volume and reducing network load.

[0041] After data preprocessing is completed, the gateway temporarily stores the processed data in the local storage unit and uploads it in batches to the edge computing nodes of the cloud-edge processing module at a preset frequency. If the network is interrupted, the data will be cached in the SD card and automatically re-uploaded after the network is restored to ensure that the data is not lost.

[0042] The IoT gateway module achieves interconnection and interoperability with the multimodal sensing module and the cloud-edge processing module in two ways. Connection to the multimodal sensing module: Wireless communication is used. After power-on, the multimodal sensing module automatically scans for nearby IoT gateway signals and selects the optimal gateway for connection based on signal strength. AES-128 encryption authentication is used during connection to ensure secure data transmission. The gateway supports simultaneous access for up to 100 multimodal sensing modules, meeting the equipment deployment requirements of large smart wards. When the maximum number of connected devices is reached, the gateway automatically rejects new device access and sends an alarm message to the visual interaction module.

[0043] Connection to the cloud-edge processing module: Edge computing nodes are deployed in the ward equipment room and connected to the IoT gateway module via gigabit Ethernet with a communication rate ≥1Gbps to ensure real-time transmission of high-frequency data; the cloud server is deployed in the hospital data center, and the gateway establishes a connection with the cloud server through the hospital's intranet or 4G / 5G network, using the HTTPS protocol for data transmission to ensure data security. The gateway establishes a heartbeat connection with the cloud-edge processing module. If the heartbeat is interrupted for more than 30 seconds, the gateway will switch to the backup network and send a network anomaly alarm.

[0044] The cloud-edge processing module adopts a cloud-edge collaborative architecture of edge computing nodes and cloud servers. The edge computing nodes are responsible for the real-time processing and immediate feedback of high-frequency data, while the cloud servers are responsible for the storage, modeling and in-depth analysis of all data. The two work together to ensure the real-time performance of data processing and realize the global control of the entire life cycle of medical devices.

[0045] Edge computing nodes establish a high-speed connection with the cloud server through the hospital's intranet. Data transmission uses the MQTT protocol. Edge computing nodes upload processed abnormal data and key statistical data to the cloud, while the cloud sends model parameters and configuration commands to the edge computing nodes, forming a closed data loop. Simultaneously, edge computing nodes have local caching capabilities. When the network connection to the cloud is interrupted, they can independently complete data processing and command generation. After the network is restored, data is synchronized to the cloud, ensuring system continuity.

[0046] The edge computing nodes utilize industrial-grade edge computing gateways and are deployed in ward equipment rooms or nurse stations. Each node is equipped with a 1TB SATA hard drive for storing recent, high-frequency sensor data and processing results. It also integrates a gigabit Ethernet port, a WiFi 6 module, and a 4G / 5 module, supporting diverse connections with IoT gateway modules, cloud servers, and visual interaction modules. The nodes employ a redundant power supply design, supporting dual 12V power supplies to ensure uninterrupted operation.

[0047] Edge computing nodes perform local processing on high-frequency new sensing data aggregated by IoT gateway modules, specifically including three functions: real-time data noise reduction, abnormal operating condition threshold judgment, and instant feedback signal generation. The high-frequency new sensing data refers to the sensing data that is synchronously collected by the multimodal sensing module for device positioning and status monitoring. After the sensor cross-comparison verification and elimination of abnormal data, the effective sensing data is obtained. Then, the IoT gateway module completes the WiFi protocol to MQTT protocol conversion, noise filtering, data format standardization and redundant data elimination preprocessing. The final updated data has an update frequency of greater than or equal to 5 times / minute and can directly reflect the real-time operating status, position dynamics and current or pressure of the medical device.

[0048] The real-time data denoising addresses environmental interference noise in high-frequency newly sensed data using an adaptive filtering algorithm. Based on the minimum mean square error (LMS) criterion, the adaptive filtering algorithm adjusts the filter coefficients in real time to adapt to the noise characteristics of different data types.

[0049] The abnormal operating condition threshold judgment is based on the standard operating parameters of different types of medical devices. A specific abnormal operating condition threshold range is set for each device, and this range can be configured and adjusted through a visual interactive module. The threshold settings are based on the device's factory parameters, clinical usage guidelines, and hospital management requirements. Edge computing nodes compare the processed new sensing data with the threshold range in real time. If the data exceeds the threshold, it is immediately determined to be an abnormal operating condition, and information such as the time of occurrence, the abnormal data value, and the device ID is recorded. Continuous abnormal data is judged as a serious abnormality; single abnormal data is judged as a minor abnormality, triggering different levels of warnings accordingly.

[0050] The real-time feedback signal generation involves the edge computing node immediately generating a feedback signal upon detecting an abnormal operating condition. This feedback signal is of two types: an early warning signal and a control signal. The early warning signal includes information such as the type of abnormal operating condition, device ID, abnormal location, and abnormal time. It is fed back to the visual interaction module via the IoT gateway module, triggering terminal pop-ups, voice prompts, and other warning methods to remind medical staff to handle the situation promptly. The control signal is generated for severe abnormal operating conditions, such as excessive device current potentially causing a fire or the device being out of range and potentially stolen. It is sent to the electronic control unit of the medical device to implement emergency shutdown or locking functions, preventing accidents. The control signal is transmitted using encryption to ensure the security and uniqueness of the executed commands.

[0051] The cloud server is deployed in the hospital's data center and adopts a cluster architecture design, including 3 application servers, 2 database servers and 1 storage server, with the following specific configuration: The application server uses an Intel Xeon Gold 6330 CPU, 64GB DDR4 memory, and a 2TB SSD to run machine learning algorithms, build full lifecycle models, and process intelligent scheduling requests. The database server uses an Intel Xeon Gold 6348 CPU, 128GB DDR4 memory, a 4TB SSD hard drive, and is equipped with a MySQL cluster database to store information such as new sensing data, model parameters, and scheduling records. The storage server adopts a distributed storage architecture with a total storage capacity of 100TB or more, supports off-site data backup and disaster recovery, and ensures long-term secure data storage.

[0052] The cloud server uses a tiered storage strategy for data storage, including three types: hot data, warm data, and cold data. Hot data is stored on SSDs to ensure high-speed read and write; warm data is stored on SAS drives to balance performance and cost; and cold data is stored on distributed storage servers to reduce storage costs.

[0053] Sensitive data is encrypted using AES-256 encryption technology during data storage, and data access permissions are set so that only authorized personnel can view and modify the data, ensuring data security and compliance.

[0054] Based on preprocessed new sensing data, the cloud server uses machine learning algorithms to construct a full lifecycle model of medical devices. The model covers the entire process from device procurement and warehousing, clinical use, cleaning and disinfection, maintenance and repair, to disposal and recycling, enabling dynamic management and control of the entire device lifecycle. Figure 2 As shown.

[0055] The data input for the medical device lifecycle model includes the following four categories: (1) Basic information data: procurement information, warehousing information and scrapping information of medical devices, etc.; (2) New sensing data: New sensing data collected by the multimodal sensing module, preprocessed by the IoT gateway, and processed by the edge computing node; (3) Clinical usage data: The departments using the device, patients using it, usage duration and treatment items, etc., are obtained through the connection with the hospital's HIS system and EMR system; (4) Maintenance records and repair records, etc., are entered through the visual interactive module.

[0056] Data input involves data cleaning and feature engineering, including removing missing and outlier values, encoding categorical data, and standardizing numerical data to ensure that the data meets the requirements for model training.

[0057] The medical device lifecycle model adopts a fusion scheme of neural network algorithm and decision tree algorithm. The two algorithms work together to achieve different functions. The neural network algorithm employs a fusion model of convolutional neural network (CNN) and long short-term memory network (LSTM) to extract data features and predict equipment failure risk, remaining service life, and maintenance requirement cycles.

[0058] CNN is responsible for extracting the spatial features of the data, while LSTM is responsible for extracting the temporal features. The model input is a temporal data sequence of length 100, and the output is the failure risk value (0-1, the closer to 1, the higher the failure risk), the remaining service life, and the maintenance requirement cycle.

[0059] The CNN consists of two convolutional layers and one pooling layer. The first convolutional layer uses 32 3×3 convolutional kernels with ReLU activation function, and the output feature map size is 98×32. The second convolutional layer uses 64 3×3 convolutional kernels with ReLU activation function, and the output feature map size is 96×64. The pooling layer uses 2×2 max pooling kernel, and the output feature map size is 48×64.

[0060] The LSTM consists of two LSTM units, each with a hidden layer dimension of 128, an activation function of tanh, and a dropout rate of 0.2 to avoid overfitting. The output of the LSTM layer is mapped to the final prediction result through a fully connected layer, which contains one hidden layer and one output layer.

[0061] The model was trained using the Adam optimizer with a learning rate of 0.001, 100 iterations, a batch size of 32, and mean squared error (MSE) as the loss function. Training data consisted of medical device operation data from the hospital over the past three years, divided into training, validation, and test sets in a 7:2:1 ratio.

[0062] The decision tree algorithm mentioned uses the C4.5 decision tree algorithm to classify and make decisions on data from each process node of medical device procurement and warehousing, clinical use, cleaning and disinfection, maintenance and repair, and disposal and recycling, and to classify the health level of the medical devices.

[0063] In terms of feature selection, 10 key features were selected as inputs to the decision tree, including cumulative usage time, number of failures, disinfection execution frequency, maintenance timeliness rate, current fluctuation amplitude, positioning movement range, sensor working status, instrument service life, clinical satisfaction score, and maintenance cost ratio.

[0064] The decision tree is constructed as follows: using the device health level as the target variable, divided into four levels: excellent, good, fair, and poor, and employing the information gain ratio as the feature selection criterion, a C4.5 decision tree is constructed. The specific process is as follows: (1) Calculate the information gain ratio of all features; (2) Select the feature with the largest information gain ratio as the root node; (3) Divide the dataset according to the different values ​​of the feature and recursively construct subtrees; (4) When the information entropy of the dataset is less than the preset threshold or all features have been used, stop the construction and generate leaf nodes.

[0065] Then, pruning is performed using a post-pruning algorithm to prune the constructed decision tree, removing redundant branches, avoiding overfitting, and improving the model's generalization ability. The depth of the pruned decision tree is controlled to be within 8 layers to ensure efficient classification decisions.

[0066] Finally, the output of the decision tree model is the health level of the device, where "Excellent" indicates that the device is in good condition with no risk of failure; "Good" indicates that the device is in normal condition and requires routine maintenance; "Medium" indicates that the device has a minor risk of failure and requires enhanced monitoring; and "Poor" indicates that the device has a serious risk of failure and requires immediate repair or replacement.

[0067] Once the full lifecycle model of a medical device is constructed, it receives new sensing data uploaded by edge computing nodes in real time. The model calculates and outputs the device's failure risk, remaining service life, maintenance requirement cycle, and health level, providing data support for the intelligent scheduling module.

[0068] The intelligent scheduling module is deployed in the application server of the cloud server. It adopts a distributed architecture design, supports multi-threaded concurrent processing, and can respond to scheduling requests from multiple wards and multiple types of instruments at the same time. The core function of the intelligent scheduling module is based on the output of the medical device lifecycle model and the priority of clinical needs. It enables dynamic tracking of devices, prediction and coordination of usage conflicts, closed-loop management of disinfection processes, and dispatch of maintenance reminders, generating and outputting optimal scheduling instructions. The software architecture of the intelligent scheduling module adopts a layered design, consisting of a data access layer, an algorithm processing layer, an instruction generation layer, and a feedback receiving layer. The data access layer is responsible for receiving full lifecycle model data from the cloud server and clinical demand data from the hospital's HIS and EMR systems, and performing data format conversion and verification. The algorithm processing layer is responsible for running core algorithms such as the clinical demand priority calculation algorithm and the conflict prediction and coordination algorithm to complete scheduling decisions. The instruction generation layer generates standardized scheduling instructions based on the algorithm processing results, including information such as instruction type, target device, executor, execution time, and operation requirements; The feedback receiving layer is used to receive the instruction execution results from the visual interaction module, update the scheduling status, and form a closed-loop control.

[0069] The intelligent scheduling module supports setting scheduling rules according to dimensions such as department, equipment type, and treatment needs. Medical staff can adjust scheduling parameters through a visual interactive module, balancing the flexibility of intelligent scheduling with manual intervention.

[0070] The device dynamic tracking system is based on location and status data collected by a multimodal perception module. An intelligent scheduling module updates the device's location, usage status, health level, and disinfection status in real time, creating a dynamic tracking ledger. This system primarily includes location tracking, status tracking, and trajectory tracing. Location tracking receives device coordinate data uploaded by positioning sensors in real time and marks the device's real-time location on an electronic map. The electronic map matches the actual hospital layout, including areas such as wards, corridors, nurse stations, and equipment rooms, and supports zooming, panning, and searching functions. Medical staff can query the real-time location of any device through the nurse station terminal or mobile medical terminal of the visual interaction module. Status tracking updates the device's usage status, health level, and disinfection status in real time and displays the device status intuitively on the electronic map using different colors. It also records the device's status change history, forming a status tracking log, which can be queried by time range. Trajectory tracing automatically records the device's movement trajectory, saving location data for the most recent 30 days. Medical staff can query information such as the device's movement path, dwell time, and department of use within any time period, facilitating device management and traceability. Tracking data can be exported to Excel format for equipment usage efficiency analysis and management assessment.

[0071] The priority of clinical needs in the intelligent scheduling module is calculated by weighting three indicators: the urgency of the patient's condition, the time node of the treatment plan, and the department's equipment quota. The weight coefficients can be adjusted by the head nurse through the visual interaction module. The default weights are set as follows: the urgency of the patient's condition, the time node of the treatment plan, and the department's equipment quota.

[0072] The urgency of the patient's condition, the timeline of the treatment plan, and the department's equipment quota are quantified; the quantified values ​​are then prioritized; the priority includes three levels: high priority, medium priority, and low priority.

[0073] The conflict prediction and coordination method uses a greedy algorithm to allocate medical device resources. The core idea is to sort clinical needs from high to low priority, prioritize high-priority needs, and at the same time, take into account factors such as the location, health level and disinfection status of the devices to select the optimal devices for allocation and avoid usage conflicts.

[0074] The specific implementation process is as follows: (1) Demand collection and sorting: The intelligent scheduling module collects the equipment usage requirements of each department in real time, sorts them from high to low according to priority score, and processes high priority requirements first; for requirements with the same priority, they are sorted from shortest to longest according to the remaining time of the treatment plan time node.

[0075] (2) Equipment screening: For each need, screen equipment that meets the following conditions: ① Health level is excellent or good; ② Disinfection status is disinfected; ③ Currently idle; ④ Location is closest to the department that needs it.

[0076] (3) Conflict judgment: If the number of available devices selected is greater than or equal to the number required, there is no conflict of use and the devices are directly allocated; if the number of available devices is less than the number required, a conflict of use is determined and coordination is required.

[0077] (4) Conflict Coordination: A greedy algorithm is used for coordination, prioritizing high-priority needs. For low-priority needs, the usage time is adjusted or idle equipment from other departments is allocated. For example, if department A has a high-priority need for a ventilator and department B also has a low-priority need for a ventilator, but only one is currently available, it will be allocated to department A first. At the same time, the medical staff of department B will be notified to adjust their treatment time or allocate idle ventilators from other wards.

[0078] The closed-loop management of the disinfection process enables the tracking and management of the entire instrument disinfection process, ensuring that disinfection work is carried out in a standardized manner. It includes four steps: The first step is generating a disinfection task. When an instrument switches from a used state to an idle state, or when the disinfection detection sensor detects that disinfection has not been performed for a preset time, the intelligent scheduling module automatically generates a disinfection task, including the instrument ID, current location, disinfection method requirements, and completion time limit, and sends it to the disinfection personnel's mobile medical terminal. The second step is tracking the disinfection process. After receiving the task, the disinfection personnel arrive at the instrument's location and scan the QR code on the instrument using their mobile medical terminal to confirm the start of disinfection. During the disinfection process, the disinfection detection sensor collects disinfection data in real time and uploads it to the intelligent scheduling module, which monitors in real time whether the disinfection meets the requirements. The third step is to confirm the disinfection results. After disinfection, the disinfection personnel enter information such as disinfection duration, disinfectant dosage, and disinfection results on the mobile medical terminal and click submit. The intelligent scheduling module verifies the results by combining the data from the disinfection detection sensors. If the disinfection requirements are met, the disinfection is deemed qualified, and the disinfection status of the equipment is updated to "disinfected." If the requirements are not met, the disinfection is deemed unqualified, a new disinfection task is generated, and the disinfection personnel are notified to disinfect again. The fourth step is to archive the disinfection records. After the disinfection process is completed, the intelligent scheduling module automatically generates disinfection records, including equipment ID, disinfection time, disinfection personnel, disinfection method, disinfection results, and test data. These records are archived to the cloud server, supporting medical personnel and management personnel to query and trace the process, ensuring that the disinfection process is manageable and controllable.

[0079] The maintenance reminder dispatch supports tiered reminders, which push reminder information to different responsible persons based on the remaining maintenance cycle and health level of the medical device, to ensure that maintenance work is carried out in a timely manner; The tiered alerts include three alert levels; Level 1 Reminder: If the remaining maintenance time is less than or equal to 24 hours or the health level is "poor", the reminder will be given to the equipment administrator and head nurse. The reminder methods include terminal pop-up window, SMS and voice prompt, and the reminder frequency is once every 1 hour until the maintenance is completed. Level 2 Reminder: If the remaining maintenance cycle time is greater than 24 hours and less than or equal to 72 hours or the health level is "medium", the reminder is given to the equipment administrator. The reminder methods include terminal pop-up window and SMS, and the reminder frequency is once every 6 hours. Level 3 Reminder: If the remaining maintenance period is more than 72 hours and the health level is "Excellent" or "Good", the reminder is given to the equipment administrator via a terminal pop-up window and is given once a day.

[0080] The reminder message includes the device ID, device name, current location, remaining maintenance period, health level, and maintenance requirements, which helps the responsible person quickly understand the situation and arrange maintenance work.

[0081] Based on the processing results of the aforementioned core functions, the intelligent scheduling module generates optimal scheduling instructions. The instructions adopt a standardized format and include fields such as instruction ID, instruction type, target device ID, executor ID, execution time, operation requirements, and priority.

[0082] After a dispatch instruction is generated, it is sent to the corresponding visual interactive terminal via the IoT gateway module and simultaneously stored in the dispatch log on the cloud server. If the instruction fails to be sent, the intelligent dispatch module will continuously retry sending it. If it still fails, the instruction will be forwarded to the nurse station terminal, where nurse station staff will manually notify the executor to ensure that the instruction is not lost.

[0083] After receiving the instructions, the personnel perform the operations as required. After the operations are completed, they report the execution results and related information through the terminal. The intelligent scheduling module receives the feedback and updates the status of the scheduling task, thus forming a closed-loop management system.

[0084] like Figure 3 As shown, the visual interaction module serves as the interface between the system and medical staff, and includes a nurse station terminal, a mobile medical terminal, and an instrument control interface. The module's software interface features a user-friendly design, with simple and intuitive operation, adaptable to the usage habits of different medical staff.

[0085] The communication of the visualization interaction module is encrypted, and all data interactions are encrypted using HTTPS and AES-128 to ensure information security. It also supports hierarchical permission management, with medical staff in different positions having different operating permissions to prevent misoperation and data leakage.

[0086] The nurse station terminals are deployed at nurse stations in each ward, featuring industrial-grade touchscreens that support multi-touch. The terminal hardware uses an Intel Core i7 processor, 16GB of RAM, a 512GB SSD, and runs Windows 11 to ensure smooth interface operation.

[0087] The nurse station terminal includes five main functions. The first function is to display an electronic map of the ward, marking the real-time location and status of all medical devices. Users can click on icons to view device details and a search function is also supported. The second function is to display statistical information on the status of ward devices in chart form, including the percentage of devices in use, idle, and faulty, the percentage of devices that have passed disinfection but not, and the percentage of devices at each health level. The charts are updated in real time, allowing medical staff to quickly grasp the overall status of the devices. The third function is to display abnormal device warning information, sorted by severity, including warning time, device ID, abnormality type, and handling suggestions. Users can click on warning information to view detailed data and handling records, and manually confirm the warning handling results. The fourth function is... The first function supports displaying a list of currently pending scheduling tasks, sorted by priority, including task ID, instruction type, target device, executor, execution time, and status. Medical staff can click on a task to view details or manually assign tasks and adjust task priorities. The fifth function provides system parameter configuration capabilities, including upload frequency, anomaly thresholds, scheduling rules, and access control for the multimodal sensing module. Only nurses with head nurse or higher privileges can operate on these parameters to ensure system configuration security. The nurse station terminal supports data export, allowing the export of device status statistics, scheduling task records, and anomaly warning records to Excel or PDF formats for departmental management and performance evaluation. It also supports log query functionality, recording all operational behaviors for easy traceability.

[0088] The mobile medical terminal uses an industrial-grade handheld tablet computer, suitable for use in complex environments such as wards and corridors. The terminal hardware features a Qualcomm Snapdragon 870 processor, 8GB of RAM, 256GB of storage, and supports fast charging. The mobile medical terminal includes five main functions. The first function is to receive dispatch instructions, anomaly alerts, and maintenance reminders from the intelligent dispatch module, with push notifications including pop-up windows, voice prompts, and vibration, ensuring medical staff receive information promptly and do not miss critical tasks. The second function supports querying real-time location, status, health level, disinfection status, and maintenance cycle information of medical devices, synchronized with data from the nurse station terminal, allowing medical staff to check device status at any time in the ward. The third function allows medical staff to receive and confirm dispatch tasks through the terminal, enter the execution results after completion, upload them to the intelligent dispatch module, and also supports taking photos of the scene as evidence. The fourth function supports manual intervention by medical staff according to their job authority. Nurses can adjust the temporary allocation of equipment and enter the reasons for the allocation; head nurses can modify the priority weight of clinical needs and scheduling rule parameters, and the adjustments take effect in real time. The fifth function supports medical staff to manually enter information such as equipment maintenance records, repair records and disinfection records to supplement the deficiencies of automatically collected data. The entered data is automatically synchronized to the cloud server to form a complete equipment file. The mobile medical terminal supports offline operation. When the network is interrupted, medical staff can still enter data and perform tasks normally. After the network is restored, the data is automatically synchronized to the system to ensure work continuity.

[0089] The device control interface is used to interact with the electronic control unit (ECU) of medical devices to achieve remote start / stop control of some devices. The interface adopts a standardized design, supporting compatibility with ECUs from mainstream medical device manufacturers. In terms of hardware interfaces, it uses both RS485 and CAN bus interfaces, which can be switched to adapt to the communication needs of different device ECUs. The interface module adopts an isolated design with overvoltage and overcurrent protection to ensure safe connection with the device ECU and not affect the original control functions of the device. For communication protocols, it uses Modbus RTU and CANopen protocols, which can be selected through software configuration. The protocol data frames include address codes, function codes, data segments, and checksums to ensure the accuracy of command transmission. In terms of control functions, it supports remote start and remote stop control, only for medical devices with remote control capabilities. Control commands require authorization verification, and the current status of the device must be confirmed before sending to avoid misoperation. The device control interface supports the recording and traceability of control commands. All remote control operations are recorded in the scheduling log, including information such as operator, operation time, command content, and execution result, facilitating management and auditing.

[0090] The manual intervention permissions of the visual interaction module are set according to job positions, and a role-based access control (RBAC) mechanism is adopted to ensure that medical staff in different positions can only perform operations corresponding to their permissions. The specific permission division is as follows: Nurses have the following permissions: they can query the status, location, and scheduling tasks of all medical devices; they can adjust the temporary allocation of devices, but must enter the reason for the allocation. The allocation is valid for a maximum of 24 hours, after which it will automatically revert to the original allocation; they can receive and execute scheduling instructions, disinfection tasks, and maintenance tasks, and provide feedback on the execution results; they cannot modify system configurations such as clinical demand priority weights, scheduling rule parameters, and abnormal thresholds.

[0091] The head nurse's permissions include all the permissions of nurses; they can modify the priority weight of clinical needs; modify scheduling rule parameters; review nurses' temporary allocation requests, extend the validity period of allocation or reject the requests; and view the department's scheduling logs, abnormal warning records and data statistical reports for management and analysis.

[0092] The device administrator has the following privileges: they can query all medical device maintenance records, repair records, and health status information; they can enter and modify device maintenance records, repair records, and scrapping information; they can receive maintenance reminders and provide feedback on maintenance results; they can apply for device procurement and scrapping and submit them to the head nurse for review; however, they cannot participate in device scheduling or clinical-related operations.

[0093] System administrator privileges, responsible for the overall configuration, maintenance and upgrade of the system; can modify all system parameters; can manage user accounts and role permissions; can view system logs, data backup and recovery; cannot participate in clinical scheduling and device management operations.

[0094] This embodiment achieves accurate collection and verification of medical device data across all dimensions through a multimodal perception module, data aggregation and preprocessing through an IoT gateway module, real-time processing of high-frequency data and full lifecycle modeling through a cloud-edge processing module, optimal scheduling instructions based on models and clinical needs through an intelligent scheduling module, and status display, instruction execution, and tiered manual intervention through a visual interaction module. These modules work collaboratively to form a closed-loop management system, solving problems in existing technologies such as single-dimensional medical device perception, insufficient data accuracy, delayed response, inefficient scheduling, and untimely maintenance. It achieves intelligent management of the entire lifecycle of medical devices, improves scheduling accuracy and efficiency, ensures clinical diagnostic and treatment needs, and is applicable to various smart ward scenarios, possessing broad application prospects and promotional value.

[0095] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A medical device IoT sensing and intelligent scheduling system for smart wards, characterized in that, The system includes: a multimodal perception module, an IoT gateway module, a cloud-edge processing module, an intelligent scheduling module, and a visual interaction module; The multimodal sensing module is deployed on the medical device body, synchronously collects the sensing data of the device and uploads it to the Internet of Things gateway module, and obtains effective sensing data by cross-comparison and verification of sensor data and elimination of abnormal data. The IoT gateway module and the multimodal sensing module establish a connection via wireless communication, aggregate the effective sensing data, complete protocol conversion and data preprocessing, obtain new sensing data, and interconnect with the cloud-edge processing module. The cloud-edge processing module includes edge computing nodes and a cloud server; the edge computing nodes process frequently occurring new sensing data after aggregation; the cloud server is used to store the new sensing data and to build a full life cycle model of medical devices based on the new sensing data using machine learning algorithms. The intelligent scheduling module is deployed on the cloud server. Based on the medical device life cycle model and clinical demand priority, it realizes dynamic tracking of devices, prediction and coordination of usage conflicts, closed-loop management of disinfection process and dispatch of maintenance reminders, and generates and outputs the optimal scheduling instructions. The visualization interaction module includes a nurse station terminal, a mobile medical terminal, and a device control interface, which is used to present the full-dimensional status of medical devices, send and receive scheduling instructions, and provide feedback on execution results.

2. The medical device IoT sensing and intelligent scheduling system for smart wards according to claim 1, characterized in that, The multimodal sensing module includes: a positioning sensor, a status monitoring sensor, a disinfection detection sensor, and a maintenance timing unit; the positioning sensor uses UWB ultra-wideband positioning technology to collect the real-time position of the instrument; the status monitoring sensor identifies the instrument's use or idle status through current and pressure feedback, and the data collected by each sensor are cross-checked to eliminate abnormal data.

3. The medical device IoT sensing and intelligent scheduling system for smart wards according to claim 1 or 2, characterized in that, The multimodal sensing module uploads sensing data using a combination of periodic and triggered uploads. Under normal conditions, data is uploaded at preset time intervals, while triggered uploads are used when a sudden change in the device's state is detected.

4. The medical device IoT sensing and intelligent scheduling system for smart wards according to claim 1, characterized in that, The wireless communication supports WiFi, Bluetooth and LoRa wireless communication protocol adaptation. During the protocol conversion process, the effective sensed data is uniformly converted into the MQTT protocol format. Data preprocessing includes noise filtering, data format standardization and redundant data removal.

5. The medical device IoT sensing and intelligent scheduling system for smart wards according to claim 1, characterized in that, The edge computing node performs localized processing of frequently occurring new sensing data after aggregation, specifically including: real-time data noise reduction, abnormal operating condition threshold judgment, and instant feedback signal generation; the real-time data noise reduction uses an adaptive filtering algorithm to remove environmental interference signals; the abnormal operating condition threshold judgment and instant feedback signal generation are based on setting threshold ranges according to the standard operating parameters of different types of medical devices, and when the threshold range is exceeded, an abnormal warning signal is generated and instantly fed back to the IoT gateway module and the visualization interaction module; the machine learning algorithm includes neural network algorithm and decision tree algorithm.

6. The medical device IoT sensing and intelligent scheduling system for smart wards according to claim 5, characterized in that, The neural network algorithm employs a fusion model of convolutional neural network and long short-term memory network, trained based on newly perceived data, to extract data features and predict device failure risk, remaining service life, and maintenance requirement cycle; the decision tree algorithm uses the C4.5 decision tree algorithm to classify and make decisions on process node data of device procurement and warehousing, clinical use, cleaning and disinfection, maintenance and repair, and scrapping and recycling, and to classify device health levels.

7. The medical device IoT sensing and intelligent scheduling system for smart wards according to claim 1, characterized in that, The priority of clinical needs in the intelligent scheduling module is calculated by weighting three indicators: the urgency of the patient's condition, the time node of the treatment plan, and the department's equipment quota; the conflict prediction and coordination is to allocate equipment resources using a greedy algorithm.

8. The medical device IoT sensing and intelligent scheduling system for smart wards according to claim 1, characterized in that, The intelligent scheduling module supports tiered reminders for maintenance reminders, pushing reminder information to the equipment administrator and head nurse respectively based on the remaining time of the maintenance cycle. The reminder methods include terminal pop-ups, SMS, and voice prompts.

9. The medical device IoT sensing and intelligent scheduling system for smart wards according to claim 1, characterized in that, The nurse station terminal adopts a large-screen display mode, showing the distribution of ward equipment, status statistics, and scheduling task progress in sections; the mobile medical terminal supports handheld operation and pushes scheduling instructions and abnormal reminders in real time; the equipment control interface is linked with the medical device electronic control unit to realize remote start and stop control of some equipment.

10. The medical device IoT sensing and intelligent scheduling system for smart wards according to claim 1, characterized in that, The visual interaction module supports manual intervention in the scheduling process by medical staff. The manual intervention permissions are set according to job level. Nurses are responsible for adjusting the temporary use allocation of equipment, and head nurses are responsible for modifying the priority weight of clinical needs and scheduling rule parameters. The full-dimensional status of the medical device includes: device location, real-time use, cleaning and disinfection, maintenance cycle information, abnormal operating conditions, and health level.

Images