Surgical instrument integrity detection and closed loop management system and method based on ai vision

Abstract

The application belongs to the technical field of intelligent management, and particularly relates to a surgical instrument integrity detection and closed-loop management method based on AI vision, and the specific steps are as follows: multi-angle imaging is performed on surgical instruments to generate a 3D grid model and establish a structured database; the identity and validity period of the instruments are read in batch preliminary screening, the posture of the instruments is captured, and depth detection and mechanical property testing are performed on the integrity; unqualified instrument packages are intercepted, the instrument package is locked and a pop-up window is prompted in an abnormal state, the remaining life is calculated by analyzing the number of uses and mechanical load, a replacement work order is generated, and is synchronized to a supply chain system, and spare parts are guided according to the replacement frequency; detection data is uploaded to a traceability system, a blockchain ID is generated, audit traceability is supported, and hospital data is integrated to optimize an AI model. The application automatically detects missing, wrong and inconsistent quantities of complete surgical instruments, and is deeply integrated with whole-process traceability, equipment predictive maintenance, intelligent scheduling and automatic quality control.

Description

Technical Field

[0001] This invention belongs to the field of intelligent management technology, specifically relating to an AI vision-based surgical instrument integrity detection and closed-loop management system and method. Background Technology

[0002] With the increasing variety of diseases, the number and types of surgical procedures are also constantly increasing. The number of surgeries performed in the operating room every day is rising sharply, and the number and types of surgical instruments are also increasing daily. Faced with a wide variety of surgical instruments, how to quickly count the instruments after surgery and reduce the loss of instruments is a prerequisite for ensuring surgical safety. Each surgery uses at least one instrument pack, and different instrument packs contain different quantities and types of instruments. After the surgery, the scrub nurse and circulating nurse usually need to repeatedly check the number of instruments, which takes a lot of time.

[0003] Problems with existing technology: Traditional surgical instrument testing still relies heavily on manual inspection, which is prone to omissions and difficulty in detecting missing parts in instrument packs. The omission rate is high and visual inspection errors are frequent. The manual inspection process is not traceable and paper records are easily lost, which can lead to disputes over liability. Traditional quality inspection procedures are relatively passive in the maintenance of surgical instruments, and manual delivery is time-consuming and has low scheduling efficiency. In addition, quality control is overly dependent on manual labor. Summary of the Invention

[0004] The purpose of this invention is to provide an AI vision-based surgical instrument integrity detection and closed-loop management system and method, which can automatically detect missing, incorrect, or quantity discrepancies in complete sets of surgical instruments, and deeply integrate with full-process traceability, predictive maintenance of equipment, intelligent scheduling, and automated quality control to form a practical overall solution.

[0005] The specific technical solution adopted by this invention is as follows: The specific steps of the AI ​​vision-based surgical instrument integrity detection and closed-loop management method are as follows: Digital modeling and standard library construction: Multi-angle imaging of surgical instruments to generate 3D mesh models, establish a structured database, store them according to specialty, integrate 2D-3D data, and generate a visual standard atlas that supports AI comparison; Dynamic detection and intelligent analysis: Batch initial screening reads the identity and expiration date of medical devices, captures the device posture and surface features, and performs surface residue, integrity and mechanical performance testing; Decision execution and interception mechanism: intercept unqualified instrument packages, lock the instrument package and pop up a prompt when there is an anomaly, analyze the number of uses and mechanical load to calculate the remaining life, generate a replacement work order and synchronize it to the supply chain system, and guide the replacement of parts according to the number of replacements. Data closed loop and full-process traceability: Test data is uploaded to the traceability system, generating a blockchain ID to support auditing and traceability, and integrating hospital data to optimize AI models.

[0006] Furthermore, during the construction of the structured database, attribute fields are defined for the grid model, and specialized classifications are performed based on the defined information: The two-dimensional image of the instrument is spatially registered with the 3D model to generate a visual standard map and store the attitude parameters of the standard map. Synchronously connect with the hospital's supplies system to obtain the procurement specifications, batch numbers, and expiration dates of medical devices, and map the device usage sequence to the surgical procedure to optimize placement logic: The instrument model contour is segmented by acquiring images of the calibration instrument and compared with the Hausdorff distance of the standard atlas model in the library. When the deviation threshold is >0.1mm, the data with excessive deviation is removed. When a new device model is added, it is directly entered into the structured database and the modification history is recorded.

[0007] Furthermore, the surgical instruments are retrieved through an RFID channel gate, and images of the instruments are captured using an industrial camera and compared with standard images in the database for posture and integrity. To detect surface wear, deformation, and micro-cracks in instruments, polarized light equipment is used to eliminate interference from metallic reflections. The acquired instrument images are synchronously injected with the YOLOv8 algorithm to enhance defect features and identify defects such as chipped blades, rust spots, and joint jamming.

[0008] Furthermore, the specific process for detecting surface residues is as follows: Industrial cameras are used to capture residues on the surface of instruments, and ultraviolet and infrared light are combined to distinguish the types of contaminants. Instrument images are injected into a U-Net architecture to train a segmentation network, which labels contaminated pixels in the instrument's tooth grooves and joint areas, and calculates the contamination coverage. The instrument image features are also matched with a standard library to automatically label rust spots and calcified scale.

[0009] Furthermore, the specific process for detecting surface residues is as follows: The residual network is used to extract the pixel features of the instrument surface. When the specular reflectance of the instrument surface is less than 95%, it is determined that the residual residue on the instrument surface exceeds the standard. Using ultraviolet light to detect protein fluorescence, when the fluorescence intensity of the instrument surface features is >50 RLU, the instrument surface is considered to have excessive biological contaminants. Using infrared absorbance analysis, when the absorbance difference of the instrument surface features is greater than 0.05, it is determined that the inorganic residue on the instrument surface exceeds the standard.

[0010] Furthermore, the specific process of the integrity detection is as follows: The instrument pack is scanned at the workstation by a 3D scanner on the carrying platform to perform multi-angle imaging. The imaging results are injected into the 3D point cloud matching algorithm. When there is an image of stacked instruments, the instance segmentation model is used to separate the overlaps. The point cloud data topology analysis is combined to confirm the independent individuals. Based on the data information of the independent individuals, it is determined whether there are missing or incorrect parts. SIFT feature points are then extracted from the point cloud data of individual components and matched with a standard model. An alarm is triggered when the calculated structural similarity index SSIM < 0.9. The final analysis results determine whether a single instrument component is missing.

[0011] Furthermore, during the testing of the aforementioned mechanical properties, pressure / torque sensors are integrated to measure the joint flexibility, biting force, and shear force of the instrument. The charge signal from the pressure / torque sensor is converted into a voltage signal by a charge amplifier, and the voltage signal is filtered by a low-pass filter to remove electromyographic noise and power frequency interference; The filtered signal enters the ANSYS simulation platform, which analyzes the stress distribution, peak biting force, and time curve of the surgical instrument, extracts key parameters such as the biting force asymmetry index and fatigue decay coefficient, predicts high wear areas, and outputs mechanical data. When the mechanical data exceeds the preset threshold, the model is determined to be abnormal and a labeling information is output.

[0012] Furthermore, in the calculation of the remaining life of the device, a recurrent neural network and a Weibull distribution model are integrated, and the input parameters include the number of uses, mechanical load, and material fatigue data to calculate the remaining life. Update the life index each time the instrument is inspected; Furthermore, when the remaining lifespan is less than 20%, a replacement work order is automatically triggered and pushed to the procurement system.

[0013] The AI ​​vision-based surgical instrument integrity detection and closed-loop management system includes a detection and management system, which comprises a hardware perception layer, an algorithm engine layer, a data management layer, and a business application layer. The hardware sensing layer integrates a multispectral imaging system and an IoT terminal. The multispectral imaging system integrates a high-precision 3D vision module, anti-interference imaging technology, and pressure / torque sensors. The high-precision 3D vision module uses a binocular camera and a 3D structured light scanner to capture the instrument's three-dimensional posture and surface defects in real time. The anti-interference imaging technology integrates polarized light and annular dark field illumination to eliminate interference from metal reflections and improve the detection rate of microscopic defects such as burrs and cracks. Embedded sensors are used to detect the instrument's joint flexibility and bite force, and are linked with AI for performance evaluation. The IoT terminal includes high-temperature resistant RFID tags and an intelligent shelf system. The RFID tags are embedded in the instrument body to track usage frequency, sterilization cycle and location information in real time. The intelligent shelf system has built-in pressure sensors and indicator lights to automatically alert for abnormal expiration dates and inventory shortages. The algorithm engine layer integrates intelligent recognition core and multimodal fusion technology; The core of intelligent recognition includes an instrument detection model and a defect analysis engine. The instrument detection model is based on an instance segmentation and target detection model, which supports instrument segmentation and classification in chaotic stacking scenarios. The defect analysis engine combines a point cloud semantic segmentation network, point cloud processing and a Weibull life prediction model to identify chipped edges and deformation and calculate the remaining life. Multimodal fusion technology includes a 2D-3D alignment algorithm and an occlusion compensation algorithm. The 2D-3D alignment algorithm is used to match real-time scanned point clouds with a structured database to detect placement angle deviations. The occlusion compensation algorithm solves the recognition blind spots caused by instrument overlap through multi-view dynamic completion technology. The data management layer integrates a digital twin platform and a traceability platform; The digital twin platform is equipped with a 3D structured database and a dynamic knowledge base. The 3D structured database is used to store digital models of medical device packages and supports fast access in the GLTF lightweight format. The dynamic knowledge base integrates data from multiple hospitals based on federated learning and continuously optimizes defect identification rules. The traceability platform is equipped with blockchain evidence storage and full-link mapping. The blockchain evidence storage is used to record the test results on the chain to ensure that the operation records cannot be tampered with. The full-link mapping is used to connect to the HIS / SPD system and dynamically adjust the instrument wear coefficient according to the type of surgery. The business application layer integrates automated quality inspection processes and a resource scheduling hub. The automated quality inspection process includes a real-time interception mechanism and robotic arm sorting. The real-time interception mechanism automatically locks the instrument package after identifying missing / wrong parts based on AI, and pops up a message indicating the name of the missing item and its shelf location. The robotic arm sorting divides the instruments into warehousing / rework / scrap channels according to the quality inspection results. The resource scheduling center is internally equipped with an intelligent replenishment engine and preventive maintenance. The intelligent replenishment engine is used to automatically generate purchase orders and schedule AGV delivery when the inventory is below the safety threshold. The preventive maintenance is based on a life prediction model to trigger maintenance work orders for high-value equipment in advance.

[0014] A computer device includes: a memory and a processor, the memory and the processor being connected via a bus; the memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, wherein when the computer programs / instructions are executed by the processor, a method for detecting and managing the integrity of surgical instruments is implemented.

[0015] The technical effects achieved by this invention are as follows: (1) The present invention establishes a 3D digital model and structured database of standard instrument packages for each specialty, including instrument name, quantity, specifications, and placement posture. The testing standards are unified and are not affected by personnel experience, fatigue, or emotions.

[0016] (2) In the dynamic detection process, the present invention uses binocular / 3D vision and AI deep learning to automatically scan, identify, count and match the cleaned instrument package. It can identify missing parts, extra parts, wrong parts, wrong model, missing components, wrong front and back, wrong placement, mechanical residue, rust and cracks, replacing manual counting and eliminating surgical risks caused by missing, insufficient or wrong packing.

[0017] (3) In performing dynamic detection, this invention supports robust recognition of random and overlapping placement, enabling the model to maintain high accuracy, stable output and reliable confidence under harsh conditions such as noise, ambiguity, occlusion, drastic changes in lighting, distribution shift and adversarial attacks. It has core advantages such as anti-interference, strong generalization and stable decision-making. It has fast detection speed and is suitable for large-volume, high-turnover medical device package management.

[0018] (4) When an abnormality is detected, the system automatically locks the unqualified package, prohibits it from entering the packaging and sterilization process, and pops up a window to indicate the name and location of the missing instrument; when the instruments used in the instrument package reach the number of replacements required, an alarm can be triggered, and it can be combined with the intelligent workbench to guide the operator to replenish parts in a standardized manner.

[0019] (5) The detection and management method implemented in this invention enables the automatic uploading of detection data to the traceability system to form an integrity detection record. Attached Figure Description

[0020] Figure 1 This is a flowchart of the surgical instrument integrity detection and closed-loop management method provided in the embodiments of the present invention. Detailed Implementation

[0021] To make the objectives and advantages of this invention clearer, the invention will be specifically described below with reference to embodiments. It should be understood that the following text is merely used to describe one or more specific embodiments of the invention and does not strictly limit the scope of protection specifically claimed by the invention.

[0022] Example 1:

[0023] like Figure 1As shown, an AI-based vision-based surgical instrument integrity detection and closed-loop management method is used to automatically detect missing, incorrect, or quantity discrepancies in complete sets of surgical instruments. It is deeply integrated with end-to-end traceability, predictive equipment maintenance, intelligent scheduling, and automated quality control to form a practical overall solution. The specific steps are as follows: Step 1: Digital Modeling and Standard Library Construction Phase An industrial-grade 3D structured light scanner is used to image surgical instruments from multiple angles. Combined with polarized light to suppress metal reflection, sub-millimeter point cloud data is generated. The point cloud data is then transformed into a lightweight 3D mesh model using a Poisson reconstruction algorithm, preserving key features such as instrument teeth and joints. Specifically, the steps for multi-angle imaging of surgical instruments include: During the debugging phase before the system is put into use, an industrial-grade 3D scanner is used to scan the instrument and acquire sub-millimeter point cloud data of the instrument. The data accurately restores the complex curved surface and tooth structure. At the same time, the texture, color and spatial posture of the instrument are captured by a multi-angle depth camera to solve the technical support for the single-view occlusion problem. The acquired images are developed using polarized light scanning or spray development to eliminate the high-gloss interference of metal instruments and ensure data integrity. It is worth noting that the specific steps for processing point cloud data are as follows: Point cloud data is transformed into a lightweight mesh model using algorithms such as Poisson reconstruction and Marching Cubes. The model retains key features, such as jaw teeth and joint axes. The mesh model identifies instrument components, such as scissor blades and needle holder grooves, based on deep neural networks (such as PointNet++), and annotates functional attribute descriptions.

[0024] Furthermore, a structured database is established to store instrument names, specifications, sterilization parameters, and placement postures according to specialty classification, such as needle holders tilted at 30°; 2D-3D data is integrated, high-definition multi-view images are spatially registered with 3D model to generate a visual standard atlas that supports AI comparison. Specifically, the steps for designing a structured database are as follows: Define attribute fields for the mesh model, including information such as instrument name, material, specifications, and sterilization tolerance temperature; classify by specialty based on the defined information, establish hierarchical indexes by specialty such as neurosurgery and orthopedics, and associate with surgical types, such as laparoscopic instrument packs needing to include trocars and dissection forceps; The specific steps for fusing multimodal data from images and model spatial data are as follows: High-resolution 2D images of the instruments (including front / side / top views) and 3D models are spatially registered using 2D-3D alignment technology to generate a visual standard atlas. The posture parameters of the standard atlas are stored, such as the precise placement angle of the instruments in the instrument pack (e.g., needle holder tilted at 30°) and stacking order (hemostat placed on top). By connecting to the hospital's supplies system through the ERP / SCM interface, cross-system data integration is achieved, and information such as the procurement specifications, batch numbers, and expiration dates of medical instruments is obtained; the surgical process is mapped synchronously, and the instrument usage sequence is extracted from the surgical video (such as SurgMotion surgical big model data), and the placement logic of instruments in the instrument pack is optimized based on the usage sequence; Specifically, the steps for AI comparison of images acquired by calibration instruments with standard atlases in the database are as follows: The instrument images acquired for calibration use an instance segmentation and target detection model based on a convolutional neural network (such as YOLOv8) to achieve instrument model contour segmentation. The Hausdorff distance between the new scan model and the standard atlas model in the library is compared, with a deviation threshold of <0.1mm, and large data with excessive deviation is automatically removed. When a new device model is added, the recognition model is updated using Few-shot Learning and directly entered into the structured database to avoid retraining the entire database. Every modification to the graph in the database is recorded based on Git-style management to ensure traceability and compliance.

[0025] Step 2: Dynamic Detection and Intelligent Analysis Stage Using RFID for initial screening, medical device packages pass through RFID access gates, and high-temperature resistant tags read the device identity and expiration date in seconds; using a binocular camera equipped with an AI model (an improved version of YOLOv8) to capture the device posture, identify overlapping / disorderly placed devices, and use a 3D point cloud matching algorithm to solve the occlusion problem, maintaining robustness under harsh conditions such as drastic changes in lighting. Specifically, the steps for initial screening of medical devices in batches are as follows: During the preparation phase before system use, a high-temperature resistant RFID tag is implanted in each instrument. The instrument information is automatically read in batches by a fixed reader or handheld device. The instrument information includes name, specifications, expiration date, frequency of use, sterilization cycle, etc., which can realize inventory and identity binding in seconds. In the initial screening stage of system deployment, all used surgical instruments are collected and passed through an RFID gate. High-resolution industrial cameras capture images of the instruments and compare them with standard images in the database to determine their posture and integrity. This captures surface wear, deformation, and micro-cracks. The RFID gate is equipped with polarized light equipment, and the capture process combined with polarized light can eliminate interference from metal reflections, accurately capturing the surface texture and spatial posture of the instruments and coping with interference from drastic changes in lighting. The captured instrument images are simultaneously injected with the YOLOv8 algorithm to enhance defect features and identify defects such as chipped blades, rust spots, and joint jamming. To improve recognition accuracy, the AI ​​model is trained with over 100,000 labeled samples. The sample training process incorporates an attention mechanism to enhance the perception of local features of the instruments. Furthermore, industrial cameras are used to capture residues on the surface of instruments, and ultraviolet and infrared light are combined to distinguish the types of contaminants by the difference in spectral reflectance. Instrument images are injected into a U-Net architecture to train a segmentation network, and contaminated pixels in areas such as instrument teeth and joints are accurately labeled. At the same time, the contamination coverage is calculated, and the instrument image features are matched with a standard library to automatically mark stubborn stains such as rust and calcified scale. The residual network extracts pixel features from the instrument surface. When the specular reflectance of the instrument surface is less than 95%, the residual residue on the instrument surface is determined to be excessive. Using ultraviolet light to detect protein fluorescence, when the fluorescence intensity of the instrument surface features is greater than 50 RLU, the biological contaminants on the instrument surface are determined to be excessive. Using infrared light to analyze infrared absorbance, when the absorbance difference of the instrument surface features is greater than 0.05, the inorganic residue on the instrument surface is determined to be excessive. Furthermore, the RFID access gate is equipped with 3D scanner scanning stations arranged in a ring. The instrument packs are imaged from multiple angles on the carrying platform. The imaging results are injected into a 3D point cloud matching algorithm to reconstruct the complete 3D model of the obscured instrument. The stacked instrument models are separated by an instance segmentation model (Mask R-CNN), and the individual units are confirmed by topological analysis of the point cloud data. Based on the data information of the individual units, it is determined whether there are missing or incorrect parts. Then, SIFT feature points are extracted from the point cloud data of the individual units and matched with the standard model. When the calculated structural similarity index SSIM < 0.9, an alarm is triggered. The final analysis results determine whether a single instrument component is missing. Adversarial training and noise injection techniques are introduced into the stacked model segmentation process to improve the model's generalization ability against blurring and occlusion, achieving a recognition accuracy of >99.7%. Furthermore, the mechanical property testing steps are as follows: Surgical instruments such as scissors and forceps enter the mechanical testing station. The platform uses a mechanical gripper to perform cutting operations on the surgical instruments. The mechanical gripper is integrated with a pressure / torque sensor to measure the joint flexibility, biting force and shearing force of the instruments in real time during the operation. The charge signal from the pressure / torque sensor is converted into a voltage signal by a charge amplifier. The voltage signal is then filtered by a low-pass filter to remove electromyographic noise and power frequency interference. The filtered signal enters the ANSYS simulation platform, which analyzes the stress distribution, peak biting force, and time curve of the surgical instrument. Key parameters such as the biting force asymmetry index and fatigue decay coefficient are extracted to predict high wear areas. A digital twin of the instrument is built in the Unity engine to simulate shearing / clamping actions to verify the performance degradation threshold. Mechanical data is output based on the test results. Mechanical data is used to determine model anomalies through multi-level thresholds and output anomaly marker information. For example: when the measured value of the bite force is greater than the standard value by 20% (e.g., scissors > 2.4 N·cm), an abnormal bite force is output, triggering a red alarm and locking the instrument; when the bite force asymmetry index is greater than 0.3 (unilateral excessive wear), an abnormal stress distribution is output, and the specific tooth groove location is marked; when the fatigue attenuation coefficient is greater than 15% / 100 uses, an abnormal fatigue failure is output, and a replacement work order is pushed to the SPD system. Furthermore, optical images, mechanical data, and usage records are synchronized to a structured database to generate a unique detection ID.

[0026] Step 3: Decision Execution and Interception Mechanism Stage The system intercepts defective instrument packs in real time. When the AI ​​model detects an anomaly, the system locks the instrument pack and displays a pop-up window indicating the name and location of the missing item. The robotic arm automatically sorts the defective packs to the rework area and prevents them from entering the sterilization process. Specifically, the process of sorting abnormal packages is as follows: The initial screening and test results enter the sorting system. The robotic arm sorts surgical instruments according to the results data. Qualified products are put into the inventory shelf, products that need repair are transferred to the maintenance area, and scrapped products are put into the recycling channel. The system tracks the circulation status of medical device packages through a dynamic inventory management engine. The status includes recycling, cleaning, sterilization, and storage. The system dynamically updates the inventory data of the inventory shelves based on the consumption of medical devices. When the number of medical devices is lower than the preset safety stock threshold, a replenishment task is automatically triggered and pushed to the procurement system. In addition, when a medical device package nearing its expiration date is detected, a light alarm is triggered, and when an expired medical device is detected, the corresponding shelf is automatically locked. The system synchronously connects to the hospital information system (HIS) to obtain surgical type data, and dynamically adjusts the instrument wear weight based on the surgical type data, such as increasing the wear coefficient of cardiac surgical instruments by 1.5 times.

[0027] Furthermore, based on the analysis of usage frequency and mechanical load using the LSTM model, a replacement work order is triggered when the remaining lifespan is less than 20%. The missing parts information of the work order is synchronized to the supply chain (SPD) system in real time, and the AGV cart delivers the spare instruments to the operating room within 10 minutes. The specific process for predicting remaining useful life is as follows: The system integrates recurrent neural networks (such as Long Short-Term Memory networks LSTM) with the Weibull distribution model. The input layer takes in parameters such as usage count, mechanical load, and material fatigue data, while the output layer outputs the remaining lifespan percentage (RUL%). When the remaining lifespan percentage is less than 20%, it generates commands for locking, pop-up alarms, and replacement work orders. The intelligent shelf system receives the commands and locks the corresponding shelf. The system displays an alarm pop-up on the device and generates a replacement work order, which is then pushed to the purchasing system. In addition, the lifespan index is updated every time the equipment is inspected. It is worth noting that the shelves are equipped with built-in pressure sensors and indicator lights to display the location and status of the instrument packs in real time. The robotic arms automatically sort and replenish the instrument packs to the designated locations according to system instructions. The intelligent shelf system, combined with hospital map navigation, uses AGV carts to accurately transport the replenished instrument packs from the warehouse to the Central Sterile Supply Department (CSSD) or operating room. By integrating the smart shelving system with the hospital information system, and combining the inventory location information provided by the smart shelving system with the surgical scheduling information provided by the hospital information system, the optimal replenishment plan is generated, such as "orthopedic drill bits are in short supply → transfer from warehouse No. 3 → AGV delivery in 10 minutes".

[0028] The detailed steps for guiding replacement parts based on the number of replacements are as follows: After cleaning, the instruments pass through the RFID channel, automatically updating the sterilization count and synchronizing the mechanical detection data to the edge computing node; The AI ​​model calculates the remaining lifespan in real time. When the lifespan of the device is below the threshold, the system automatically locks the device and displays a pop-up window on the operation interface indicating the reason for replacement. At the same time, it pushes a list of replacement parts suggestions to the CSSD personnel's tablet. Automatic allocation is prioritized. If the hospital has spare inventory, the system generates an allocation instruction, calls an AGV cart to pick up the goods, and delivers them to the disinfection supply center. If inventory is insufficient, the system will automatically connect to the SPD system, trigger emergency procurement, automatically generate a purchase order, push it to the supplier platform, and synchronize logistics tracking information to the hospital dashboard. When new medical devices are received, their RFID tags are scanned, automatically binding them to the original device ID and activating usage rights. The system records the replacement time, operator, and reason for scrapping of the equipment during the transfer and recycling process, and synchronizes the information to the blockchain for evidence storage.

[0029] Step 4: Data Closure and Full-Process Traceability Stage All test data is uploaded to the traceability system in real time, generating a unique blockchain ID to support auditing and traceability. Specifically, the audit tracing process is as follows: The system collects environmental parameter data from cleaning, sterilization, and storage processes and automatically uploads it to the cloud; it scans RFID tags on medical device packages with industrial cameras to automatically link medical device information with operation records; and it adopts a consortium blockchain architecture to distribute and store medical device circulation data, ensuring that it cannot be tampered with. It is worth noting that the above data transmission uses the TLS 1.3 protocol to encrypt the data stream, combined with homomorphic encryption technology to ensure data privacy during cloud analysis; a hierarchical storage architecture is adopted, with hot data (real-time operation logs) stored on edge computing nodes with a response time of <500ms, and cold data (historical records) stored in a distributed database; The graph analysis model constructs a knowledge graph of the entire lifecycle of medical device packages, linking multi-dimensional data such as patients, operating rooms, and sterilization batches, and supports two-way traceability; more than 120 quality control rules are predefined in the AI ​​early warning rule library, such as "the same medical device package is reused within 24 hours" to automatically trigger audits; It is worth noting that it supports forward tracing, reverse tracing, and emergency recall mechanisms simultaneously. The specific process is as follows: Forward traceability (device → patient): Scan the RFID code on the surgical pack → display sterilization record / operator → associate with the surgical patient ID and preoperative examination report; Reverse traceability (patient → device): Enter the patient's medical record number → retrieve the list of device packs used → locate the cleaning video and quality inspection report of suspected problematic devices; Emergency recall mechanism: If a defect is found in a batch of instruments → the system automatically locks 36 associated instrument packages → pushes a pop-up alarm to the operating room and freezes usage permissions.

[0030] Furthermore, based on federated learning, data from multiple hospitals is integrated to continuously optimize the AI ​​model; surgical video analysis is performed to determine instrument usage sequences and optimize placement logic.

[0031] Example 2:

[0032] The AI ​​vision-based surgical instrument integrity detection and closed-loop management system includes a hardware perception layer, an algorithm engine layer, a data management layer, and a business application layer. The hardware sensing layer integrates a multispectral imaging system and an IoT terminal; The multispectral imaging system integrates a high-precision 3D vision module, anti-interference imaging technology, and pressure / torque sensors. The high-precision 3D vision module uses a binocular camera and a 3D structured light scanner to capture the instrument's three-dimensional posture and surface defects in real time. The anti-interference imaging technology integrates polarized light and annular dark field illumination to eliminate interference from metal reflections and improve the detection rate of microscopic defects such as burrs and cracks. Embedded sensors are used to detect the instrument's joint flexibility and bite force, and are linked with AI for performance evaluation. The IoT terminals include high-temperature resistant RFID tags and intelligent shelf systems. The RFID tags are embedded in the medical devices to track usage frequency, sterilization cycle and location information in real time. The intelligent shelf system has built-in pressure sensors and indicator lights to automatically alert users to expiration dates and inventory shortages. The algorithm engine layer integrates intelligent recognition core and multimodal fusion technology; The core of intelligent recognition includes an instrument detection model and a defect analysis engine. The instrument detection model is based on instance segmentation and target detection models, and supports instrument segmentation and classification in chaotic stacking scenarios. The defect analysis engine combines point cloud semantic segmentation network, point cloud processing and Weibull life prediction model to identify chipped edges and deformation and calculate remaining life. Multimodal fusion technology includes 2D-3D alignment algorithms and occlusion compensation algorithms. The 2D-3D alignment algorithm is used to match real-time scanned point clouds with structured databases to detect placement angle deviations. The occlusion compensation algorithm solves the recognition blind spots caused by instrument overlap through multi-view dynamic completion technology. The data management layer integrates the digital twin platform and the traceability platform; The digital twin platform has a 3D structured database and a dynamic knowledge base. The 3D structured database is used to store digital models of medical device packages and supports quick access in the lightweight GLTF format. The dynamic knowledge base integrates data from multiple hospitals based on federated learning and continuously optimizes defect identification rules. The traceability platform is equipped with blockchain evidence storage and full-link mapping. Blockchain evidence storage is used to record test results on the chain to ensure that operation records cannot be tampered with; the full-link mapping is used to connect to the HIS / SPD system and dynamically adjust the instrument wear coefficient according to the type of surgery. The business application layer integrates automated quality inspection processes and a resource scheduling hub; The automated quality inspection process includes a real-time interception mechanism and robotic arm sorting. The real-time interception mechanism automatically locks the instrument package after identifying missing / wrong parts based on AI, and pops up a message indicating the name of the missing item and its shelf location. The robotic arm sorting sorts the instruments into the warehousing / rework / scrap channel according to the quality inspection results. The resource scheduling center is equipped with an intelligent replenishment engine and preventive maintenance. The intelligent replenishment engine automatically generates purchase orders and schedules AGV delivery when inventory falls below a safety threshold. The preventive maintenance is based on a life prediction model and triggers maintenance work orders for high-value equipment in advance.

[0033] A computing device includes a memory and a processor connected via a bus; the memory stores computer programs / instructions, and the processor executes the computer programs / instructions, which, when executed by the processor, implement a method for detecting and managing the integrity of surgical instruments.

[0034] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A method for surgical instrument integrity detection and closed-loop management based on AI vision, characterized in that, The specific steps are as follows: Digital modeling and standard library construction: Multi-angle imaging of surgical instruments to generate 3D mesh models, establish a structured database, store them according to specialty, integrate 2D-3D data, and generate a visual standard atlas that supports AI comparison; Dynamic detection and intelligent analysis: Batch initial screening reads the identity and expiration date of medical devices, captures the device posture and surface features, and performs surface residue, integrity and mechanical performance testing; Decision execution and interception mechanism: intercept unqualified instrument packages, lock the instrument package and pop up a prompt when there is an anomaly, analyze the number of uses and mechanical load to calculate the remaining life, generate a replacement work order and synchronize it to the supply chain system, and guide the replacement of parts according to the number of replacements. Data closed loop and full-process traceability: Test data is uploaded to the traceability system, generating a blockchain ID to support auditing and traceability, and integrating hospital data to optimize AI models.

2. The surgical instrument integrity detection and closed-loop management method according to claim 1, characterized in that: During the construction of the structured database, attribute fields are defined for the grid model, and specialized classifications are performed based on the defined information. The two-dimensional image of the instrument is spatially registered with the 3D model to generate a visual standard map and store the attitude parameters of the standard map. Synchronously connect with the hospital's supplies system to obtain the procurement specifications, batch numbers, and expiration dates of medical devices, and map the device usage sequence to the surgical procedure to optimize placement logic: The instrument model contour is segmented by acquiring images of the calibration instrument and compared with the Hausdorff distance of the standard atlas model in the library. When the deviation threshold is >0.1mm, the data with excessive deviation is removed. When a new device model is added, it is directly entered into the structured database and the modification history is recorded.

3. The surgical instrument integrity detection and closed-loop management method according to claim 1, characterized in that: The surgical instruments are recovered through an RFID channel gate, and images of the instruments are captured by an industrial camera and compared with standard images in the database for posture and integrity. To detect surface wear, deformation, and micro-cracks in instruments, polarized light equipment is used to eliminate interference from metallic reflections. The acquired instrument images are synchronously injected with the YOLOv8 algorithm to enhance defect features and identify defects such as chipped blades, rust spots, and joint jamming.

4. The surgical instrument integrity detection and closed-loop management method according to claim 1, characterized in that: The specific process for detecting surface residues is as follows: Industrial cameras are used to capture residues on the surface of instruments, and ultraviolet and infrared light are combined to distinguish the types of contaminants. Instrument images are injected into a U-Net architecture to train a segmentation network, which labels contaminated pixels in the instrument's tooth grooves and joint areas, and calculates the contamination coverage. The instrument image features are also matched with a standard library to automatically label rust spots and calcified scale.

5. The surgical instrument integrity detection and closed-loop management method according to claim 4, characterized in that: The specific process for detecting surface residues is as follows: The residual network is used to extract the pixel features of the instrument surface. When the specular reflectance of the instrument surface is less than 95%, it is determined that the residual residue on the instrument surface exceeds the standard. Using ultraviolet light to detect protein fluorescence, when the fluorescence intensity of the instrument surface features is >50 RLU, the instrument surface is considered to have excessive biological contaminants. Using infrared absorbance analysis, when the absorbance difference of the instrument surface features is greater than 0.05, it is determined that the inorganic residue on the instrument surface exceeds the standard.

6. The surgical instrument integrity detection and closed-loop management method according to claim 1, characterized in that: The specific process of the integrity detection is as follows: The instrument pack is scanned at the workstation by a 3D scanner on the carrying platform to perform multi-angle imaging. The imaging results are injected into the 3D point cloud matching algorithm. When there is an image of stacked instruments, the instance segmentation model is used to separate the overlaps. The point cloud data topology analysis is combined to confirm the independent individuals. Based on the data information of the independent individuals, it is determined whether there are missing or incorrect parts. SIFT feature points are then extracted from the point cloud data of individual devices and matched with a standard model. An alarm is triggered when the calculated structural similarity index SSIM < 0.

9. The final analysis results determine whether a single device component is missing.

7. The surgical instrument integrity detection and closed-loop management method according to claim 1, characterized in that: During the testing of the mechanical properties, pressure / torque sensors are integrated to measure the joint flexibility, biting force, and shear force of the instrument. The charge signal from the pressure / torque sensor is converted into a voltage signal by a charge amplifier, and the voltage signal is filtered by a low-pass filter to remove electromyographic noise and power frequency interference. The filtered signal enters the ANSYS simulation platform, which analyzes the stress distribution, peak biting force, and time curve of the surgical instrument, extracts key parameters such as the biting force asymmetry index and fatigue decay coefficient, predicts high wear areas, and outputs mechanical data. When the mechanical data exceeds the preset threshold, the model is determined to be abnormal and a labeling information is output.

8. The surgical instrument integrity detection and closed-loop management method according to claim 1, characterized in that: In the calculation of the remaining life of the instrument, a recurrent neural network and a Weibull distribution model are integrated. The input parameters include the number of uses, mechanical load and material fatigue data to calculate the remaining life. Update the life index each time the instrument is inspected; Furthermore, when the remaining lifespan is less than 20%, a replacement work order is automatically triggered and pushed to the procurement system.

9. A surgical instrument integrity detection and closed-loop management system based on AI vision, comprising a detection and management system, wherein the detection and management system is used to execute the surgical instrument integrity detection and closed-loop management method according to any one of claims 1-8, characterized in that: The detection and management system includes a hardware perception layer, an algorithm engine layer, a data management layer, and a business application layer. The hardware sensing layer integrates a multispectral imaging system and an IoT terminal. The multispectral imaging system integrates a high-precision 3D vision module, anti-interference imaging technology, and pressure / torque sensors. The high-precision 3D vision module uses a binocular camera and a 3D structured light scanner to capture the instrument's three-dimensional posture and surface defects in real time. The anti-interference imaging technology integrates polarized light and annular dark field illumination to eliminate interference from metal reflections and improve the detection rate of microscopic defects such as burrs and cracks. Embedded sensors are used to detect the instrument's joint flexibility and bite force, and are linked with AI for performance evaluation. The IoT terminal includes high-temperature resistant RFID tags and an intelligent shelf system. The RFID tags are embedded in the instrument body to track usage frequency, sterilization cycle and location information in real time. The intelligent shelf system has built-in pressure sensors and indicator lights to automatically alert for abnormal expiration dates and inventory shortages. The algorithm engine layer integrates intelligent recognition core and multimodal fusion technology; The core of intelligent recognition includes an instrument detection model and a defect analysis engine. The instrument detection model is based on an instance segmentation and target detection model, which supports instrument segmentation and classification in chaotic stacking scenarios. The defect analysis engine combines a point cloud semantic segmentation network, point cloud processing and a Weibull life prediction model to identify chipped edges and deformation and calculate the remaining life. Multimodal fusion technology includes a 2D-3D alignment algorithm and an occlusion compensation algorithm. The 2D-3D alignment algorithm is used to match real-time scanned point clouds with a structured database to detect placement angle deviations. The occlusion compensation algorithm solves the recognition blind spots caused by instrument overlap through multi-view dynamic completion technology. The data management layer integrates a digital twin platform and a traceability platform; The digital twin platform is equipped with a 3D structured database and a dynamic knowledge base. The 3D structured database is used to store digital models of medical device packages and supports fast access in the GLTF lightweight format. The dynamic knowledge base integrates data from multiple hospitals based on federated learning and continuously optimizes defect identification rules. The traceability platform is equipped with blockchain evidence storage and full-link mapping. The blockchain evidence storage is used to record the test results on the chain to ensure that the operation records cannot be tampered with. The full-link mapping is used to connect to the HIS / SPD system and dynamically adjust the instrument wear coefficient according to the type of surgery. The business application layer integrates automated quality inspection processes and a resource scheduling hub. The automated quality inspection process includes a real-time interception mechanism and robotic arm sorting. The real-time interception mechanism automatically locks the instrument package after identifying missing / wrong parts based on AI, and pops up a message indicating the name of the missing item and its shelf location. The robotic arm sorting divides the instruments into warehousing / rework / scrap channels according to the quality inspection results. The resource scheduling center is internally equipped with an intelligent replenishment engine and preventive maintenance. The intelligent replenishment engine is used to automatically generate purchase orders and schedule AGV delivery when the inventory is below the safety threshold. The preventive maintenance is based on a life prediction model to trigger maintenance work orders for high-value equipment in advance.

10. A computer device, comprising: A memory and a processor, the memory and the processor being connected via a bus; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the surgical instrument integrity detection and closed-loop management method according to any one of claims 1-8.

Images