A surgical instrument real-time inventory and closed-loop traceability system based on internet of things and multi-modal perception

Abstract

The present application relates to the technical field of surgical instruments, in particular to a surgical instrument real-time inventory and closed-loop tracing system based on Internet of Things and multi-modal sensing, which gives each instrument a unique digital identity through an instrument identity module, and uses a sensing terminal network distributed in key nodes of an operating room to collect instrument position and state data in real time. A central processing server serves as a decision-making center, with built-in dynamic benchmark management, state machine tracking, multi-dimensional inventory verification and intelligent early warning units, which drive instrument state transitions through a finite state machine rule engine, and perform layered inventory verification and intelligent decision-making. The system realizes full-process closed-loop management from preoperative preparation, intraoperative real-time tracking to postoperative tracing and sterilization supply, completely eliminates the monitoring blind area of traditional inventory methods, significantly improves the inventory accuracy and operating room collaborative management efficiency, and fundamentally eliminates the risk of surgical foreign body retention.

Description

Technical Field

[0001] This invention relates to the field of surgical instrument technology, specifically to a real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing. Background Technology

[0002] Intraoperative counting of surgical instruments, gauze, sutures, and other supplies is a crucial and repetitive task in the surgical procedure, aimed at preventing foreign objects from being left inside the patient and ensuring patient safety. Traditional counting procedures heavily rely on manual, visual, and verbal verification by the circulating nurse and scrub nurse. This method is not only inefficient, but also highly susceptible to human error, especially during long surgeries with a wide variety of instruments, due to staff fatigue, emergency situations, or misunderstandings. Particularly when counting intraoperative gauze, small sutures, and similarly shaped instruments, this purely manual method carries a significant risk of omissions and miscounts, posing a potential threat to medical safety.

[0003] To improve the efficiency and accuracy of inventory checks, the industry has attempted to introduce automatic identification technologies, such as solutions based on radio frequency identification (RFID). However, existing RFID inventory systems are mostly limited to static batch inventory checks before and after surgery, failing to achieve continuous tracking of the real-time location and status of instruments (e.g., on the operating table, in use, or discarded) throughout the entire surgical process. This results in the inability to immediately and accurately confirm whether all instruments have been safely retrieved at critical steps (such as before closing the body cavity), creating a significant "monitoring blind spot" in the system. Furthermore, single-dimensional identification methods are susceptible to interference in the complex electromagnetic environment of the operating room, resulting in insufficient read / write reliability and a lack of intelligent fault-tolerance mechanisms for misreading or missed readings.

[0004] Another significant drawback of existing technologies lies in their information isolation. Instrument inventory data is often independent of the hospital's core information management system (such as electronic medical records and operating room management systems), failing to form a closed-loop data flow that runs through preoperative preparation, intraoperative tracking, postoperative traceability, and sterilization supply management. This information fragmentation creates a break in the lifecycle management of instruments, hindering efficient process collaboration and posing significant challenges to subsequent problem tracing and root cause analysis.

[0005] Therefore, there is an urgent need in this field for a system that can achieve real-time, closed-loop, and intelligent traceability of surgical instruments throughout the entire process. It must be able to overcome a series of problems existing in the current technology, such as monitoring blind spots, poor reliability due to environmental interference, information isolation, and lack of intelligent decision support. Summary of the Invention

[0006] The purpose of this invention is to provide a real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing. By creating digital identities for instruments, constructing a distributed sensing network, and utilizing an intelligent decision engine, the system achieves real-time tracking and closed-loop management throughout the entire process, effectively preventing foreign objects from being left behind, ensuring patient safety, and improving the efficiency of operating room management.

[0007] To achieve the above-mentioned technical objectives and effects, the present invention is implemented through the following technical solution: A real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing, comprising a closed-loop control system consisting of a physical sensing layer, a network transmission layer, and an intelligent decision-making layer, including: The equipment identification module is used to create a unique digital twin for each inventoryable object, and includes: The instrument tags attached to reusable surgical instruments are ultra-high frequency passive RFID tags encapsulated in biocompatible materials. The gauze tag woven into medical gauze is an RFID tag made using flexible circuit technology. The needle carrier for suture needles is a carrier embedded with an RFID tag and engraved with an optical code. A distributed sensing and acquisition network constructs a multi-dimensional data acquisition field within the operating room space, comprising: The sterile instrument sensing terminal, located under the sterile instrument table, includes a UHF RFID antenna array and a reader / writer, and is used for inventorying and coarse-grained positioning of instruments. The intelligent waste collection terminal integrated into the waste bin's dispensing port forms an RFID identification tunnel to determine the effective insertion of the equipment. Access control and monitoring terminals deployed at the entrances and exits of the operating room form an electromagnetic monitoring curtain to determine the direction of entry and exit of instruments; Handheld auxiliary terminals for manual operation and inventory management; The central processing server, as the decision-making hub of the system, includes: The dynamic benchmark management unit is used to create and maintain a dynamic benchmark device set B in memory, and can update the set in response to instructions; The state machine tracking unit maintains state variables for each device in set B and drives state transitions based on event streams from the perception network through a rule engine based on finite state machines. The multi-dimensional inventory and verification unit responds to inventory instructions and executes a hierarchical verification process to identify high-risk missing sets, manage abnormal sets, and resolve state logic conflicts. The intelligent decision-making and early warning unit receives the verification results and executes the decision tree to generate alarm instructions of different levels or through instructions; The human-computer interaction and execution module transforms decision commands into multimodal physical outputs, including a main control display screen, multi-level sound and light alarms, and a network voice broadcaster.

[0008] Furthermore, the central processing server also integrates a location analysis and auxiliary search unit, which is configured as follows: RFID signal strength data from a sterile table antenna array is processed, and the two-dimensional coordinates of the instrument are calculated using an improved weighted centroid positioning algorithm; the algorithm is as follows: for instrument i, its coordinates (X... i Y i Calculated using the following formula: , Wherein, weight w ij Signal strength RSSI ij Decision, w ij = (RSSI) ij -RSSI min +ε) k RSSI min ε is the minimum readable signal strength threshold, ε is a small constant to prevent division by zero, and k is the amplification factor. When the instrument status changes from "on stage" to "in use", record its last known coordinates (X). last (i), Y last (i) and logical region Z last (i); When generating a Level 1 alarm, the device i m Z last (i m The information and the circular area highlighted on the layout diagram are encapsulated in the alarm data packet and sent to the main control display screen.

[0009] Furthermore, the central processing server also integrates a confidence calculation and fault tolerance unit, which is configured as follows: For each successful RFID read event e, a confidence score C(e) is calculated, and its multiple regression model is as follows:

[0010] in:

[0011] It is a maximum-min normalization of the original signal strength; Is the device i in the most recent sliding time window T? window The standard deviation of the number of reads within 60 seconds (e.g., after normalization); For each device i, maintain dynamically updated average historical confidence level C.avg (i) This value is the exponentially weighted moving average of the confidence levels of its past successful read events; After identifying the high-risk missing set, each device i within it is examined. m C avg (i m If the value is below the empirical threshold, a fault tolerance flag is sent to the early warning unit.

[0012] Furthermore, the central processing server also integrates a time-series behavior analysis unit, which is configured as follows: Listen to the state transition event stream and construct a state transition frequency f for each device i. i Average single usage time t use-avg (i) Multidimensional usage timeline archive; Periodically perform unsupervised clustering analysis on the archive data of all devices to automatically identify device categories with different usage patterns; For device i identified as a "high-frequency used device", the time threshold δ applied in the state machine tracking unit will be... use (i) Dynamically adjusted to:

[0013] Where η is the adjustment coefficient (0 < η < 1), f max The maximum frequency observed; simultaneously, the multi-dimensional inventory verification unit uses an individualized risk threshold when detecting conflict β:

[0014] Where κ is another adjustment coefficient.

[0015] Furthermore, the central processing server also integrates an intelligent report generator, which is configured to automatically aggregate time-series data and generate a structured electronic traceability report at key surgical nodes and at the end of the surgery. This report includes at least: A timeline of inventory consistency presented in an interactive Gantt chart format; Anomaly event tracing log recorded in reverse chronological order; A sequence diagram of microstate transitions generated for each instrument.

[0016] Furthermore, the system also includes a system integration interface gateway, which is configured to enable bidirectional data exchange with the hospital operating room management system, electronic medical record system, and central sterilization supply management system via the HL7 FHIR standard protocol stack.

[0017] Furthermore, the central processing server also integrates a machine learning optimization unit, which is configured as follows: Continuously collect contextual data from historical surgeries to form a feature vector F. n Optimal system parameter set P n and the success rate of counting S n The dataset D constitutes the dataset; A multi-objective regression model M is trained using the random forest regression algorithm on dataset D. The input of the model is the feature vector F, and the output is the prediction P of the optimal parameter set. pred ; At the start of a new surgery, based on surgical characteristics F new Infer the personalized recommendation parameter P new And automatically distribute it to the corresponding units in the system.

[0018] On the other hand, the present invention proposes an operating room instrument pack for use with the above-mentioned system, which contains multiple standardized surgical instruments bearing the above-mentioned RFID tags; The outer packaging of the instrument pack has a unique package barcode; The database of the central processing server pre-stores a strong association mapping between the package barcode and the list of RFID IDs of all devices inside the package; Scanning the package barcode can trigger the dynamic reference management unit to execute a batch initialization script, preloading the ID list of the entire package of instruments into the dynamic reference set B.

[0019] Furthermore, the coarse-grained localization of the sterile table sensing terminal is achieved through a matching algorithm based on signal strength spatial fingerprinting, including: Offline training phase: A virtual grid is established on a sterile table, and at each grid point G... （x，y） The upper record standard reference tag's signal strength vector RSSI ref Construct a location fingerprint database FPDB; Online positioning phase: For device tag i, its signal strength vector RSSI is acquired. i By calculating the similarity with reference vectors in FPDB, the estimated coordinates (X) are calculated using the k-nearest neighbor algorithm. i-est Y i-est ).

[0020] Furthermore, the intelligent waste collection terminal ensures the accuracy of identification through a multi-sensor fusion state machine, which includes "idle", "object enters", "identifying", and "confirm delivery" states. The state transition is jointly driven by events from the infrared photoelectric sensor, RFID reader, and weight sensor. The "confirm delivery" state is reached and the waste event is reported only when the RFID is successfully read, the infrared beam is restored to unobstructed flow, and the weight sensor reading shows a stable step change.

[0021] The beneficial effects of this invention are: Traditional inventory systems only perform static inventory checks before and after surgery, failing to track the dynamic movement of instruments during the procedure. This invention constructs a seamless monitoring field at key nodes in the operating room through a distributed sensing and acquisition network: an antenna array under the sterile operating table performs configurable periodic scanning to continuously monitor the "on-the-table" status of instruments; an RFID identification tunnel formed by the waste bin disposal port, combined with multi-sensor fusion, accurately captures instrument disposal events; and an access control monitoring terminal uses phase sensing technology to determine the direction of instrument entry and exit. These real-time event streams drive the finite state machine in the state machine tracking unit to perform state transitions, maintaining a lifecycle record for each instrument including states such as "on-the-table," "in use," and "discarded." This transforms discrete inventory operations into continuous state monitoring. Through comprehensive coverage in both spatial and temporal dimensions, it ensures that the position and state changes of any instrument at any stage of surgery can be sensed and recorded in real time. This allows for immediate and accurate confirmation of the safe retrieval of all instruments before critical steps such as closing body cavities, fundamentally eliminating the risk of foreign objects being left behind.

[0022] This invention significantly improves the robustness and counting accuracy of the system in complex surgical environments through multi-dimensional intelligent verification and adaptive fault tolerance mechanisms. Operating rooms are filled with factors that affect the reliability of RFID reading and writing, such as electromagnetic interference, metal reflection, and fluid obstruction. The confidence calculation and fault tolerance unit integrated into the central processing server calculates a comprehensive confidence score for each reading event. This score is derived through weighted regression of multi-dimensional features such as signal strength normalization, standard deviation of reading counts, and phase stability, and is used to maintain a dynamically updated average historical confidence score for each instrument. When the counting verification unit identifies a high-risk missing set, it checks the average historical confidence score of each instrument within it. If the score is lower than an empirical threshold, the instrument is classified as low-confidence and the fault tolerance mechanism is triggered. A probabilistic model distinguishes between genuine missing instruments and momentary missed readings caused by environmental interference, avoiding unnecessary interference with the surgical process caused by false alarms. Meanwhile, the multi-dimensional inventory and verification unit's hierarchical verification process not only checks for quantity consistency, but also identifies identity consistency anomalies through set theory operations, and detects conflicts that violate medical logic, such as "ghost instruments" (discarded instruments reappearing on the sterile table), through state logic consistency verification, thus maintaining extremely high judgment accuracy even in complex environments.

[0023] This invention achieves adaptive parameter optimization and intelligent decision support based on instrument usage patterns, enabling the system to continuously learn and evolve. Different surgical types and instruments exhibit significant differences in usage patterns, making fixed parameter thresholds unsuitable for diverse clinical scenarios. This invention continuously monitors the state transition event stream through a temporal behavior analysis unit, constructing a multi-dimensional usage timeline profile for each instrument, including features such as state switching frequency and average single-use duration. Unsupervised clustering analysis is performed periodically to automatically identify instrument categories with different usage patterns. Based on this analysis, the system dynamically adjusts the key parameter thresholds in the state machine tracking unit and the multi-dimensional inventory verification unit: for frequently used instruments, their state transition time thresholds are appropriately shortened; for instruments used for extended periods, their high-risk time thresholds are correspondingly extended. Individualized and scenario-based parameter configuration is achieved through behavior pattern recognition, effectively reducing the false alarm rate caused by fixed parameters. The machine learning optimization unit uses a random forest regression algorithm to predict the optimal system parameter set based on the feature vectors of historical surgeries and automatically injects the corresponding functional units at the start of a new surgery, enabling the system to adapt to the characteristics of different surgical scenarios and continuously optimize inventory performance.

[0024] This invention breaks down information barriers throughout the entire medical device management process, constructing a data closed loop deeply integrated with the hospital's information system. Traditional medical device inventory systems often form information silos, unable to effectively collaborate with the hospital's core management system. This invention, through a system integration interface gateway based on the HL7 FHIR international standard protocol, achieves bidirectional data exchange with the hospital's operating room management system, electronic medical record system, and central sterile supply management system. Preoperatively, surgical schedules and patient information are automatically acquired; during surgery, key inventory results and early warning events are pushed in real time and archived to the electronic medical record; postoperatively, structured electronic traceability reports are synchronized to the central sterile supply center. Through standardized data models and interface protocols, the previously isolated medical device inventory data is integrated into the hospital's overall information flow, forming a complete data closed loop from preoperative preparation and intraoperative tracking to postoperative traceability and sterile supply. This not only achieves full traceability of the medical device lifecycle, providing a complete data chain for medical quality audits and root cause analysis, but also optimizes the workflow of the central sterile supply center through data-driven approaches, enabling refined management of medical device resources and efficient cross-departmental collaboration.

[0025] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Detailed Implementation

[0026] 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.

[0027] Example 1

[0028] The embodiment described herein A real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing, comprising a closed-loop control system consisting of a physical sensing layer, a network transmission layer, and an intelligent decision-making layer, specifically including: The equipment identification module creates a unique digital twin for each inventoryable item through both physical and informational means, including: Instrument Tag: Using medical-grade PEEK or ceramic materials for biocompatible encapsulation, an ultra-high frequency passive RFID tag chip (compliant with EPC Global C1G2 standard) is permanently fixed to the non-functional area of ​​the ring, joint, or shaft of reusable instruments such as surgical forceps, scissors, and needle holders; the tag memory stores a 64-bit or 96-bit unique instrument identifier EPC, and reserves storage blocks for writing the instrument model, activation date, and maximum sterilization cycle count; Gauze tag: Using textile-based flexible circuit technology, a dipole antenna made of silver-plated fiber or stainless steel fiber is woven into the yarn of medical gauze and interconnected with a miniature ultra-thin RFID chip. Its overall structure can withstand high-temperature steam sterilization of not less than 135℃ and can still maintain a return signal strength of not less than -10 dBm after the antenna is wetted by blood or saline. Needle carrier: Each suture needle is equipped with a smart insert plate, which contains a UHF RFID tag. The surface of the insert plate is also laser-engraved with a Data Matrix QR code. The data field of the QR code includes the EPC code of the RFID tag bound to it, forming a dual identification redundancy of optical and radio frequency.

[0029] The distributed sensing and acquisition module constructs a multi-dimensional data acquisition field within the physical space of the operating room, including: The sterile instrument table sensing terminal consists of a UHF RFID antenna array embedded under the sterile instrument table surface and a multi-port reader / writer. The antenna array comprises at least four directional antennas arranged in a two-dimensional grid, with the main lobes of each antenna pointing upwards and slightly intersecting. The reader / writer cyclically activates each antenna via a time-division multiple access protocol, with an inventory period T. scan Configurable, ranging from 100ms to 500ms; the sterile table sensing terminal performs coarse-grained positioning of instruments using a spatial spectrum estimation algorithm based on received signal strength indication, specifically: for any instrument tag i, at time t, read the signal strength vector RSSI when it is received by N antennas. i (t) = [RSSI i1 RSSI i2 , ..., RSSI iNBy comparing the relative magnitudes of the elements within the vector, the device i is mapped to K logical regions Z pre-divided by the antenna coverage area. k One of (k=1, 2, ..., K); Intelligent waste collection terminal: Integrated into the disposal opening of sharps containers and medical waste bins, forming a U-shaped RFID identification tunnel; this tunnel has UHF RFID antennas with mutually perpendicular polarization directions deployed at both the entrance and exit, forming a near-field coupling zone; when an object passes through, an infrared photoelectric sensor triggers the reader to complete the process within a time window Δt. waste A dense inventory is performed within the window, and only when the same tag is successfully read n times consecutively (n≥3) within the window is it considered a valid input event; Access control monitoring terminal: Two circularly polarized UHF RFID antennas are symmetrically deployed on both sides of the operating room entrance door frame to form a phase sensing zone; the terminal analyzes the Doppler frequency shift or phase change caused by the tag crossing the phase sensing zone, and combines this with the time stamp difference Δt read by the two antennas. door It determines whether instruments are entering or leaving the operating room and establishes a heartbeat mechanism with the central processing server to report the access control status in real time. Handheld auxiliary terminal: It adopts a built-in UHF RFID reader / writer module and a high-resolution QR code scanner to perform authorized manual inventory, manual addition or removal of instruments in case of system malfunction.

[0030] The central processing server, as the decision-making hub of the system, includes the following units: Dynamic reference management unit: Creates and maintains a dynamic reference instrument set B in memory. Its initialization logic is as follows: During the surgical preparation phase, execute B. initial = i | i ∈ Scan Result (Table) Terminal ,t0), where Scan Result (Table) Terminal B(t0) represents the set of instrument IDs obtained by the aseptic table sensing terminal performing a full scan at the initial time t0. initial The initial set of dynamic reference instruments, i.e., the initial value of B; i is the instrument ID, representing the unique identifier of the instrument; Table Terminal This is the identifier for the sterile table sensing terminal. This unit listens for authorization commands from the handheld assistive terminal; when it receives a command to add a new instrument... new When the instruction is given, a set update operation is performed: B current = B previous ∪i new And generate audit log entries including operator ID, timestamp, and device ID; B current For the updated dynamic reference instrument set, B previousThe set of dynamic reference instruments before the update; State Machine Tracking Unit: This unit maintains a state variable S(i) for each device i in set B, with its state space being (on stand, in use, abandoned, suspected to be off-site). The rule engine based on a finite state machine drives state transitions according to the event flow E(t) from the perception network, and its core state transition function δ is defined as follows: δ: In Taiwan → In Use When the state S(i) of device i is on stage, and the event Table Exit (i) When this occurs, perform the assignment operation: S(i) = in use.

[0031] Table Exit (i) The event disappears from the scanning range of instrument i at the end of the sterile table, and the duration is Δt. absent >δ use It is triggered at a typical time (3-5 minutes).

[0032] δ: In use → Obsolete When the state S(i) of device i is in use, and the event Waste Detected (i) When this occurs, perform the assignment operation: S(i) = obsolete.

[0033] Waste Detected (i) The event is generated by direct reporting from the intelligent waste collection terminal.

[0034] δ: In use → On the platform When the state S(i) of device i is in use, and the event Table Reentry (i) When this occurs, perform the assignment operation: S(i) = on the platform.

[0035] Table Reentry (i) The event is triggered when instrument i is re-stable scanned by the sterile station terminal (e.g., read for 3 consecutive cycles).

[0036] δ: (In Taiwan, in use) → Suspected of leaving the room When the state S(i) of device i is in operation or in use, and the event Door Exit (i) When this occurs, perform the assignment operation: S(i) = suspected room leave.

[0037] Door Exit (i) The event is generated when the access control monitoring terminal determines that instrument i leaves the operating room.

[0038] Multi-dimensional inventory verification unit: Responding to inventory commands triggered by the system at scheduled times or manually, it executes a layered verification process, including the following steps: S1: Data Snapshot and Locking: At time tcheck Lock the current dynamic benchmark set B t Real-time sensing set of sterile table T t Collection of historical abandoned archives W t (Including all instruments marked as discarded during this surgery); S2: Global quantity consistency check: Calculate the quantity difference Δ count =|B t |-(|T t |+|W t |); S3: Device Identity Consistency Verification: Identification through set theory operations: High-risk deletion set M highRisk =B t \(T t ∪W t ), \ represents the set difference operation, which is the machine that should be on the table or that is in the scrap but has not been found; Management Exception Set A manage = (T t ∪W t )\B t This refers to equipment that appears in the system but is not in the initial baseline (possibly contaminated equipment that was mistakenly placed on the platform). S4: State logic consistency check: Traverse the state of all instruments and check for the following logical conflicts: Conflict α (Ghost Machine): ∃x, x∈T t ∧S(x) = Discarded. This conflict indicates that a device that has been reported as discarded has reappeared on the sterile table, violating the logic of medical waste management; Conflict β (overtime use of the instrument): ∃y, y∈B t ∧S(y) = ∧(t) in use check -t last use (y))>δ highRisk Among them, t last use (y) represents the moment when the device's y state changes to being in use, δ highRisk For high-risk time thresholds (e.g., 30 minutes), this conflict suggests that the device may have been left in the patient's body cavity; Intelligent Decision-Making and Early Warning Unit: Receives the above verification results and executes a deterministic decision tree. IF(M highRisk ≠∅) OR (Logical conflict detected α or β) THEN generates a Level 1 alarm command; ELSE IF (M highRisk =∅)AND(No logical conflict)AND(Δcount≠0)THEN Generate a level 2 alarm instruction (indicates management anomaly, but no immediate high risk); ELSE generates a count pass command.

[0039] The human-computer interaction and execution module translates the decision instructions from the central processing server into multimodal physical outputs. Main control display screen: Real-time rendering of sterile table layout diagram, instrument status heat map, warning details and handling instructions through a graphical interface; Multi-level sound and light alarm: It has three predefined output modes: Level 1 alarm (red RGB LED rotating and flashing, accompanied by a continuous 85dB buzzing sound), Level 2 alarm (yellow LED constantly on, accompanied by an intermittent 65dB beeping sound), and pass signal (green LED constantly on, accompanied by a crisp ding sound). Networked voice broadcaster: Receives text-to-voice instructions from the server and plays natural language prompts through speakers deployed in the operating room, such as "Count correct, please continue" or "Level 1 alarm, missing instrument detected, please pause operation and verify."

[0040] In this embodiment, the central processing server also integrates a positioning analysis and auxiliary search unit, which works in conjunction with the sterile table sensing terminal to achieve refined location services for the instruments. The system processes RFID signal strength data from the sterile table antenna array and calculates the instrument's two-dimensional logical coordinates (X) using an improved weighted centroid positioning algorithm. i Y i The specific algorithm is as follows: Assume there are M reference antennas, whose position coordinates are known as (x...). j y j (j=1,2,...,M), for instrument i, its coordinates are calculated using the following formula: , w ij = (RSSI) ij -RSSI min +ε) k in, X i Y i The estimated two-dimensional logical coordinates of device i; M represents the total number of reference antennas; x j y j The known position coordinates of the j-th reference antenna w ij : Signal strength weight between device i and antenna j; RSSI ij : The signal strength received by device i by antenna j; RSSI min The minimum signal strength threshold that the system can read; weight w ij Signal strength RSSI ij Decision, RSSI min ε is the minimum readable signal strength threshold, ε is a small constant to prevent division by zero, and k is an amplification factor, usually taken as 2, used to enhance the weighted difference of signal strength. When the state machine tracking unit changes the state of device i from "on stage" to "in use" according to the rules, it will simultaneously record its last known precise coordinates (X). last (i), Y last (i) and its logical region Z last (i); When generating a Level 1 alarm command, the intelligent decision-making and early warning unit will search for M. highRisk Each device i m Z last (i m The information will include the message "Last appeared in [area name]" and a highlighted circular area on the aseptic table layout diagram (marked with (X)). last (i m ), Y last (i m The data is encapsulated in an alarm data packet with the historical average positioning error σ as the radius and sent to the main control display screen for visual display to assist medical staff in directional searching.

[0041] In this embodiment, the central processing server also integrates a confidence calculation and fault tolerance unit, which improves the system's robustness in complex electromagnetic environments through a probabilistic model: Instantaneous confidence model: A confidence score C(e) is calculated for each successful RFID read event e (corresponding to device i being read by antenna j at time t). Its multiple regression model is as follows:

[0042] in:

[0043] It is a maximum-min normalization of the original signal strength; Is the device i in the most recent sliding time window T? window The standard deviation of the number of reads within 60 seconds (e.g., after normalization); Based on phase ij The calculated stability index scores high if the phase value fluctuates less than the threshold in several consecutive readings. α, β, and γ are weighting coefficients, and α + β + γ = 1, which are obtained through training with historical data; Device profile maintenance: Maintain dynamically updated average historical confidence level C for each device i. avg (i), which is the exponentially weighted moving average of the confidence C(e) of its past L successful read events (e.g., L=100): C avg (i) new = λ·C(e) + (1-λ)·C avg (i) old , where λ is the attenuation factor; C avg (i) new The average historical confidence level of the updated device i; C avg (i) old The average historical confidence level before the update; C(e) is the confidence score of the current read event.

[0044] Fault-tolerant decision-making mechanism: Identifying M in a multi-dimensional inventory and verification unit. highRisk Then, the fault-tolerant unit will check each of the devices i. m C avg (i m If C avg (i m (Below the empirical threshold θ derived from a large amount of clinical data) low If the device is identified as a "low-confidence device," a fault-tolerant flag is sent to the intelligent decision-making and early warning unit. When generating a Level 1 alarm command, the early warning unit adds explanatory text to these devices, such as "low confidence, suspected to be due to signal obstruction causing missed reading, manual verification of this area is recommended first," thereby optimizing the priority of the alarm and the accuracy of the guidance.

[0045] Furthermore, the central processing server also integrates a time-series behavior analysis unit, which achieves adaptive optimization of system parameters by learning the usage patterns of the equipment. Behavioral pattern modeling: Continuously monitor the state transition event stream of the state machine tracking unit, and construct a multi-dimensional usage time sequence profile for each device i, excluding the state transition frequency f. i In addition, it also records the average duration of a single use, t. use-avg (i) The most frequently used logical region Z mode (i); Dynamic clustering and pattern recognition: This unit periodically performs unsupervised clustering analysis (such as K-means clustering) on ​​the archive data of all instruments, and automatically identifies categories such as "high-frequency used instruments", "long-term used instruments" and "instruments for specific areas". Parameter adaptive adjustment: This unit works in conjunction with the state machine tracking unit and the multi-dimensional inventory verification unit. For device i identified as a "high-frequency used device," the state machine tracking unit adjusts its time threshold δ when applying rule R1. use (i) Dynamically adjusted to:

[0046] Where η is the adjustment coefficient (0 < η < 1); δ use-default The default time threshold; f i The state switching frequency of device i; f max The maximum state switching frequency among all observed instruments; Where η is the adjustment coefficient (0 < η < 1), f max The maximum frequency observed; simultaneously, the multi-dimensional inventory verification unit uses an individualized risk threshold when detecting conflict β:

[0047] δ highRisk (i) represents the adjusted high-risk time threshold for device i; δ highRisk-default The default high-risk time threshold; t use-avg (i): The average duration of a single use of device i; t use-max The maximum average usage time among all devices; κ is another adjustment coefficient. This fine-tuning of parameters effectively reduces the false alarm rate of devices used for high-frequency and long-term periods.

[0048] In this embodiment, the central processing server also integrates an intelligent report generator, which is configured to automatically perform the following operations at critical surgical points (such as before closing the body cavity and after suturing the skin) and at the end of the surgery: Aggregate time-series data from various units of the system, including the change history of the dynamic baseline set B, the complete transition sequence of the state machine S(i), and the results R of all inventory checks. check (t), and all warning events E alert (t); Based on the aggregated data, generate a structured, auditable electronic traceability report that conforms to medical record standards. This report should include at least: Inventory Consistency Timeline: Visualizes the status of each predefined inventory node throughout the entire surgical procedure (pass / Level 1 Alert / Level 2 Alert) in an interactive Gantt chart format, and supports clicking to view a detailed snapshot of the inventory; Anomaly traceability log: Records detailed metadata for each alarm in reverse chronological order, including the trigger timestamp, the EPC list of the triggering device, and the root cause (such as the specific difference set M). highRisk (or logical conflict type), system-recommended treatment suggestions, nurse's actual handling actions, and final confirmation results; Individual medical device lifecycle map: Generate a micro-state transition sequence map for each medical device in the baseline set B. The horizontal axis represents the operation time, and the vertical axis represents the state (on the operating table, in use, abandoned). Mark the key event points captured by each sensing terminal on the timeline to form an immutable electronic evidence chain.

[0049] In this embodiment, the system further includes a system integration interface gateway, which serves as a standardized bridge for bidirectional data exchange with the hospital information ecosystem. The gateway is a standalone embedded server that complies with the electromagnetic compatibility standards for medical devices in terms of hardware, and runs a protocol stack that supports the HL7 FHIR R4 international standard in software. It achieves automatic, asynchronous data interaction with external systems through a secure HTTPS / VPN channel within the hospital. The specific interaction process includes: Preoperative information retrieval: After the surgical schedule is determined, the gateway actively initiates a FHIRRESTful API call to the hospital operating room management system to query and obtain the Patient resources (patient information), Procedure resources (surgical information), and Practitioner resources (medical staff information) for the upcoming surgery, and uses this information to create a surgical context in the central processing server; Intraoperative event push: All key count results (including count passed, secondary alarm, and primary alarm) and warning events generated by the intelligent decision-making and early warning unit are encapsulated as FHIR Observation resources and attached to the corresponding surgical context. They are then pushed to the hospital's electronic medical record system in real time and archived as structured nursing process documents. Postoperative report synchronization: At the end of the surgery, the final electronic traceability report generated by the intelligent report generator is packaged into an FHIR DocumentReference resource and automatically pushed to: The hospital's electronic medical record system serves as a permanent component of the patient's surgical record. The Central Sterile Supply Department (CSSD) management system provides data-driven support for the accurate collection, cleaning, sterilization, and redistribution of medical devices. The data packets pushed to the CSSD include a list of discarded medical devices, enabling the CSSD to anticipate the number of devices to be collected and to initiate a traceability process for missing devices.

[0050] In this embodiment, the central processing server also integrates a machine learning optimization unit, which enables the system to learn and evolve from historical data. Training Data Pool Construction: This unit continuously collects and stores contextual data from historical surgeries, fully complying with data anonymization regulations, to form a multidimensional feature dataset D = (F... n P n S n ),in: F n It is the feature vector of the nth surgery, including the surgery type (such as CPT code), the surgeon's identifier, the estimated duration, and the distribution of the types of instruments actually used; P n It is the set of system parameters that was subsequently proven to be optimal in this operation, namely (δ use δ highRisk θ low This was determined through retrospective analysis; S n It is the final success rate of the surgery (binary label); Model Trainer: This unit periodically (e.g., weekly) trains a multi-objective regression model M using a dataset D via a random forest regression algorithm. The model's input is the feature vector F, and its output is a prediction P of the optimal parameter set. pred =(δ use-pred δ highRisk-pred θ low-pred The training process employs k-fold cross-validation to evaluate the model's generalization ability and calculates feature importance rankings. Online inference and parameter injection: At the start of a new operation, the unit extracts its known features F. new The trained model M is invoked for inference to obtain the personalized recommendation parameters P. new Subsequently, it transmits δ via the internal message bus. use-pred Send it to the state machine tracking unit, and δ highRisk-pred The data is sent to the multi-dimensional inventory verification unit, and θ is... low-pred The parameters are then sent to the confidence calculation and fault tolerance unit to enable adaptive configuration of system parameters for specific surgical scenarios.

[0051] Example 2

[0052] An operating room instrument bag, as described in this embodiment, is used in conjunction with the system described in Embodiment 1. It includes multiple standardized surgical instruments with RFID tags. The surgical instruments are registered by an RFID reader before being placed in the bag. In this embodiment, a unique barcode is affixed to the outer packaging of the instrument pack. The barcode adopts the GS1-128 standard and encodes the unique ID, type and expiration information of the instrument pack. In the database of the central processing server, a strong association mapping Map between the package barcode and the RFID ID list of all instruments inside the package is pre-established and stored. Barcode to EPCList; EPCList refers to the list of RFID tag IDs of all individual instruments within the same instrument package; Scanning the package barcode using a handheld auxiliary terminal or a dedicated fixed barcode scanner can trigger the system's dynamic baseline management unit to execute a batch initialization script: querying the database based on the scanned barcode to obtain the corresponding EPC list, and then executing B... initial = B current ∪ (EPCList) allows for the rapid and accurate preloading of the entire instrument pack into the dynamic reference set B, greatly improving efficiency in the surgical preparation phase and avoiding errors caused by manual scanning one by one.

[0053] In this embodiment, the coarse-grained two-dimensional spatial positioning function of the sterile station sensing terminal is implemented through a matching algorithm based on signal strength spatial fingerprinting: Offline training phase: When the sterile table is unloaded, a virtual X×Y grid (e.g., a 10cm×10cm grid) is established on the table surface using an antenna array. At each grid point G... （x，y） Above, the reader records the signal strength vector from a standard reference tag at a known location: RSSI ref = [RSSI ref1 RSSI ref2 , ..., RSSI refN ], the vectors of all grid points constitute the location fingerprint database FPDB = (G （x，y） RSSI ref ); RSSI ref For grid point G (x，y) The reference tag signal strength vector recorded above; RSSI i Let be the real-time signal strength vector of the instrument i to be positioned; N is the number of antennas; FPDB is a location fingerprint database that stores reference vectors for all grid points; Online positioning phase: For any device tag i that needs to be positioned, the system acquires its signal strength vector RSSI in real time. i By calculating RSSI i With all RSSIs in FPDB ref Using Euclidean distance or cosine similarity, find the k most similar reference vectors (e.g., k=3), and then calculate the estimated coordinates (X, X) of device i by weighted averaging using the k-nearest neighbor algorithm. i-est Y i-est ), and map it to the corresponding logical region Z k .

[0054] In this embodiment, the instantaneous association mechanism of the intelligent waste collection terminal ensures the accuracy of identification through a state machine fused from multiple sensors: In addition to the UHF RFID reader, the terminal also integrates an infrared photoelectric sensor and a miniature weight sensor. Its internal state machine is defined as follows: "Idle", "Object enters", "Identifying", "Delivery confirmed"; State transitions are driven by sensor events: From "idle" to "object enters": triggered by the infrared beam being blocked; From "object enters" to "identification in progress": Simultaneously activate the RFID reader for intensive inventory checks and monitor minute changes in the weight sensor; From "Identifying" to "Delivery Confirmed": Three conditions must be met simultaneously: 1) RFID successfully reads at least one tag ID; 2) The infrared beam is clear again (indicating that the object has completely fallen); 3) The weight sensor reading undergoes a step change and stabilizes. Only when the status reaches "delivery confirmed" will the list of identified tag IDs be bound to the "discarded" event and reported to the central processing server, thereby effectively preventing misidentification caused by objects briefly stopping or shaking at the entrance.

[0055] In this embodiment, the state machine tracking unit is built into the physical memory of the central processing server and is configured to work in conjunction with the continuous data stream of the sterile table sensing terminal through a hardware timer module to achieve precise monitoring of instrument state changes. The state machine tracking unit allocates an independent state record block in memory for each device i in the dynamic reference set B. This block stores the following data fields: Current state value S(i) The timestamp t that was finally read by the aseptic table sensing terminal last-seen (i) The time t when the status is marked as "in use" marked-in-use (i) Device type identifier Type(i) The system initializes the hardware timer with a fixed period Δt. check (Configured to 1-5 seconds) Trigger status check routine; The specific execution process of rule R1 is achieved through the following consecutive physical steps: Each time the timer is triggered, the state machine tracking unit obtains the latest scan result set T from the sterile table sensing terminal. current ; For each device i∈B, if S(i) = "on stage", then execute: a. Verify whether device i exists in set T. current middle; b. If it does not exist, then calculate the current time and t. last-seen (i) the difference Δt absent = t now - t last-seen (i); c. Δt absent The threshold δ obtained by querying the preset device type-time threshold mapping table according to Type(i) use (Type(i)) is compared; d. If Δt absent >δ use If (Type(i)), then the following physical operations are performed: 1. Update the value of S(i) from "in operation" to "in use"; 2. Set the current time t now Write t marked-in-use (i) Fields; 3. Generate a table of state transition events. Exit (i) and write it to the system event queue of the central processing server; For those appearing in T current Any device i, immediately using the current time t now Update its t last-seen (i) Field.

[0056] The device type-time threshold mapping table is physically stored in the non-volatile memory of the central processing server, associating the physical characteristics of the device with the operational logic. For example, a longer threshold δ is configured for devices such as "retractors" and "spreaders". use-long (Range 5-8 minutes) A ​​shorter threshold δ is configured for surgical scissors and forceps. use-short (Range 2-4 minutes).

[0057] In this embodiment, the multi-level sound and light alarm of the human-computer interaction and execution module is an embedded hardware device, including: The microcontroller unit is responsible for protocol parsing and logic control; The tri-color RGB LED light ring has a concentric circle structure in its physical layout. The piezoelectric ceramic buzzer's drive circuit supports frequency and duty cycle modulation. The wireless communication module is used to receive instructions from the central processing server; A real-time clock module for local timing; The alarm's operating logic is implemented through its internal firmware, specifically executing the following physical processes: Instruction parsing and arbitration: The microcontroller unit continuously listens for instruction data packets from the communication module; when multiple alarm instructions are received simultaneously, hardware interrupt priority processing is performed, and only the highest priority instruction is responded to. The priority order is: Level 1 alarm > Level 2 alarm > pass signal; Multimodal output control: Based on the parsed instruction type, the microcontroller unit directly drives the LED ring and buzzer through its GPIO pins, generating three distinct physical signal modes: Level 1 Alarm Mode: The microcontroller outputs a PWM signal to make the red segment of the LED ring rotate and flash at a frequency of 0.5Hz, while simultaneously outputting a continuous square wave signal with a frequency of 2.5kHz and a duty cycle of 50% to the buzzer. Level 2 Alarm Mode: The microcontroller outputs a high-level signal to keep the yellow segment of the LED ring constantly lit, while simultaneously outputting an intermittent square wave signal with a frequency of 1kHz and a period of 1 second (0.5 seconds of sound, 0.5 seconds of silence) to the buzzer. Signal mode: The microcontroller outputs a high-level signal to keep the green segment of the LED ring constantly lit, while simultaneously outputting a short, single-tone signal with a duration of 0.2 seconds and a frequency of 3kHz to the buzzer; Progressive alarm escalation: Maintaining an alarm duration counter t in the microcontroller's firmware. alert When in Level 1 Alert Mode and t alert When the preset threshold T1 is exceeded (e.g., 2 minutes), the physical parameters of the output signal are automatically modified: the LED rotation frequency is increased to 1Hz, and the buzzer drive frequency is increased to 3kHz; when t alert When the threshold T2 is exceeded (e.g., 5 minutes), the microcontroller will send a trigger signal to the network voice broadcaster through the communication module to start periodic voice broadcasting; Hardware self-test function: When the alarm is powered on each time, the microcontroller automatically executes a self-test sequence: sequentially driving the red, yellow and green LEDs to light up, and driving the buzzer to emit a test tone with a frequency sweep from 1kHz to 3kHz, thereby verifying the physical integrity of all output units.

[0058] In summary, this invention proposes a real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things (IoT) and multimodal sensing. Each instrument is assigned a unique digital identity through an instrument identification module, and the location and status data of instruments are collected in real time using a network of sensing terminals distributed at key nodes in the operating room. The central processing server acts as the decision-making hub, incorporating dynamic benchmark management, state machine tracking, multi-dimensional inventory verification, and intelligent early warning units. It drives instrument state transitions through a finite state machine rule engine and performs hierarchical inventory verification and intelligent decision-making. This system achieves closed-loop management of the entire process from preoperative preparation and real-time intraoperative tracking to postoperative traceability and sterilization supply, completely eliminating the monitoring blind spots of traditional inventory methods, significantly improving inventory accuracy and operating room collaborative management efficiency, and fundamentally eliminating the risk of foreign body retention during surgery.

[0059] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing, characterized in that, include: The equipment identification module is used to create a unique digital twin for each inventoryable object, and includes: The instrument tags attached to reusable surgical instruments are ultra-high frequency passive RFID tags encapsulated in biocompatible materials. The gauze tag woven into medical gauze is an RFID tag made using flexible circuit technology. The needle carrier for suture needles is a carrier embedded with an RFID tag and engraved with an optical code. A distributed sensing and acquisition network constructs a multi-dimensional data acquisition field within the operating room space, comprising: The sterile instrument sensing terminal, located under the sterile instrument table, includes a UHF RFID antenna array and a reader / writer, and is used for inventorying and coarse-grained positioning of instruments. The intelligent waste collection terminal integrated into the waste bin's dispensing port forms an RFID identification tunnel to determine the effective insertion of the equipment. Access control and monitoring terminals deployed at the entrances and exits of the operating room form an electromagnetic monitoring curtain to determine the direction of entry and exit of instruments; Handheld auxiliary terminals for manual operation and inventory management; The central processing server, as the decision-making hub of the system, includes: The dynamic benchmark management unit is used to create and maintain a dynamic benchmark device set B in memory, and can update the set in response to instructions; The state machine tracking unit maintains state variables for each device in set B and drives state transitions based on event streams from the perception network through a rule engine based on finite state machines. The multi-dimensional inventory and verification unit responds to inventory instructions and executes a hierarchical verification process to identify high-risk missing sets, manage abnormal sets, and resolve state logic conflicts. The intelligent decision-making and early warning unit receives the verification results and executes the decision tree to generate alarm instructions of different levels or through instructions; The human-computer interaction and execution module transforms decision commands into multimodal physical outputs, including a main control display screen, multi-level sound and light alarms, and a network voice broadcaster.

2. The real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing as described in claim 1, characterized in that: The central processing server also integrates a location analysis and auxiliary search unit, which is configured as follows: RFID signal strength data from a sterile table antenna array is processed, and the two-dimensional coordinates of the instrument are calculated using an improved weighted centroid positioning algorithm; the algorithm is as follows: for instrument i, its coordinates (X... i Y i Calculated using the following formula: ， Wherein, weight w ij Signal strength RSSI ij Decision, w ij = (RSSI) ij -RSSI min +ε) k RSSI min ε is the minimum readable signal strength threshold, ε is a small constant to prevent division by zero, and k is the amplification factor. When the instrument status changes from "on stand" to "in use", record its last known coordinates (X). last (i), Y last (i) and logical region Z last (i); When generating a Level 1 alarm, the device i m Z last (i m The information and the circular area highlighted on the layout diagram are encapsulated in the alarm data packet and sent to the main control display screen.

3. The real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing as described in claim 1, characterized in that: The central processing server also integrates a confidence calculation and fault tolerance unit, which is configured as follows: For each successful RFID read event e, a confidence score C(e) is calculated, and its multiple regression model is as follows: in: It is a maximum-min normalization of the original signal strength; Is the device i in the most recent sliding time window T? window The standard deviation of the number of reads within 60 seconds (e.g., after normalization); For each device i, maintain dynamically updated average historical confidence level C. avg (i) This value is the exponentially weighted moving average of the confidence levels of its past successful read events; After identifying the high-risk missing set, each device i within it is examined. m C avg (i m If the value is below the empirical threshold, a fault tolerance flag is sent to the early warning unit.

4. The real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing as described in claim 1, characterized in that: The central processing server also integrates a time-series behavior analysis unit, which is configured as follows: Listen to the state transition event stream and construct a state transition frequency f for each device i. i Average single usage time t use-avg (i) Multidimensional usage timeline archive; Periodically perform unsupervised clustering analysis on the archive data of all devices to automatically identify device categories with different usage patterns; For device i identified as a "high-frequency used device", the time threshold δ applied in the state machine tracking unit will be... use (i) Dynamically adjusted to: Where η is the adjustment coefficient (0 < η < 1), f max The maximum frequency observed; simultaneously, the multi-dimensional inventory verification unit uses an individualized risk threshold when detecting conflict β: Where κ is another adjustment coefficient.

5. The real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing as described in claim 1, characterized in that: The central processing server also integrates an intelligent report generator, which is configured to automatically aggregate time-series data and generate a structured electronic traceability report at key surgical points and at the end of the surgery. This report includes at least the following: A timeline of inventory consistency presented in an interactive Gantt chart format; Anomaly event tracing log recorded in reverse chronological order; A sequence diagram of microstate transitions generated for each instrument.

6. The real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing as described in claim 1, characterized in that: The system also includes a system integration interface gateway, which is configured to enable bidirectional data exchange with the hospital operating room management system, electronic medical record system, and central sterilization supply management system via the HL7 FHIR standard protocol stack.

7. The real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing as described in claim 1, characterized in that: The central processing server also integrates a machine learning optimization unit, which is configured as follows: Continuously collect contextual data from historical surgeries to form a feature vector F. n Optimal system parameter set P n and the success rate of counting S n The dataset D constitutes the dataset; A multi-objective regression model M is trained using the random forest regression algorithm on dataset D. The input of the model is the feature vector F, and the output is the prediction P of the optimal parameter set. pred ; At the start of a new surgery, based on surgical characteristics F new Infer the personalized recommendation parameter P new And automatically distribute it to the corresponding units in the system.

8. The operating room instrument pack used in conjunction with the system according to any one of claims 1-7, characterized in that: The instrument package contains multiple standardized surgical instruments with the aforementioned RFID tags; a unique package barcode is affixed to the outer packaging of the instrument package; a strong association mapping between the package barcode and the RFID ID list of all instruments in the package is pre-stored in the database of the central processing server; by scanning the package barcode, the dynamic reference management unit can be triggered to execute a batch initialization script, preloading the ID list of the entire package of instruments into the dynamic reference set B.

9. The real-time inventory and closed-loop traceability system for surgical instruments based on the Internet of Things and multimodal sensing as described in claim 1, characterized in that: The coarse-grained localization of the sterile table sensing terminal is achieved through a matching algorithm based on signal strength spatial fingerprinting, including: Offline training phase: A virtual grid is established on a sterile table, and at each grid point G... （x，y） The upper record standard reference tag's signal strength vector RSSI ref Construct a location fingerprint database FPDB; Online positioning phase: For device tag i, its signal strength vector RSSI is acquired. i By calculating the similarity with reference vectors in FPDB, the estimated coordinates (X) are calculated using the k-nearest neighbor algorithm. i-est Y i-est ).

Images