Medical instrument intelligent identification method based on Internet of Things

Abstract

The invention provides a medical instrument intelligent identification method based on the Internet of Things, and the method comprises the steps: obtaining the real-time operation process information of an operation region, generating a current-stage instrument demand set through an instrument demand identification model, and carrying out the specification optimization. And adding the optimized set to a historical instrument demand atlas, updating the atlas, extracting a historical set from the atlas in combination with historical global instrument demand change information, inputting the historical set into an instrument demand change prediction model, and obtaining instrument demand change information. And comparing historical and predicted instrument demand change information, determining a difference, generating real-time instrument demand change information, and finally displaying a to-be-prepared instrument list on the Internet of Things terminal. The accuracy and efficiency of surgical instrument supply can be improved, the surgical risk is reduced, and the overall management level of an operating room is improved.

Description

Technical Field

[0001] This invention relates to the field of medical device management technology, and more specifically, to an intelligent identification method for medical devices based on the Internet of Things. Background Technology

[0002] In today's medical field, with the continuous development and increasing complexity of surgical techniques, the management and use of medical devices play a crucial role in the surgical process. The accurate and timely supply of instruments in the operating room is essential for the smooth progress of surgery and patient safety. Traditional medical device management methods mainly rely on manual recording and manual allocation, which is not only inefficient but also prone to errors. For example, untimely or incorrect instrument supply may delay the surgical process and even pose potential risks to the patient. Furthermore, the wide variety of surgical instruments and the complex frequency and sequence of their use make it difficult for manual management to accurately predict and meet the dynamic needs during surgery.

[0003] With the rise of the Internet of Things (IoT) technology, its application in the medical field has gradually attracted attention. IoT technology enables interconnectivity between devices, providing new ideas for the intelligent management of medical devices. By installing sensors or tags on surgical instruments and deploying IoT devices in the operating room, the usage status and location information of instruments can be monitored in real time. However, current IoT-based medical device management technologies mostly remain at the level of simple monitoring of instrument location and status, lacking the ability to intelligently predict and dynamically adjust instrument demand. While existing technologies can improve the efficiency of instrument management to some extent, they cannot dynamically adjust instrument supply based on real-time surgical progress and patient physiological parameters during surgery, making it difficult to meet the precise needs of complex surgical scenarios.

[0004] In implementing the embodiments of the present invention, the prior art has at least the following problems or defects: On the one hand, traditional instrument management methods cannot reflect changes in instrument demand during the surgical process in real time, resulting in lag and uncertainty in instrument supply; on the other hand, although existing Internet of Things-based instrument management technology can achieve partial automated monitoring, it lacks the ability to deeply analyze and predict instrument demand, and cannot comprehensively judge the changing trend of instrument demand based on factors such as surgical stage, patient physiological parameters, and historical data, thus failing to provide the operating room with accurate instrument scheduling and preparation prompts. Summary of the Invention

[0005] This invention provides a smart identification method for medical devices based on the Internet of Things, comprising: Acquire a set of real-time surgical progress information for the target surgical area. The real-time surgical progress information in the set includes: surgical stage identifier, patient physiological parameters, and sequence of instruments used. Based on the real-time surgical progress information set, a corresponding set of instrument requirements for the current stage is generated through a pre-trained instrument requirement recognition model. The current set of medical device requirements is optimized to obtain an optimized set of medical device requirements. The optimized set of medical device requirements is added to a pre-stored historical set of medical device requirements to obtain an updated set of medical device requirements. Obtain historical global equipment demand change information corresponding to the historical equipment demand atlas; Obtain and optimize at least one historical device demand set whose time interval is less than a preset duration from the historical device demand map; Input at least one historical equipment demand set and an optimized equipment demand set into a pre-trained equipment demand change prediction model to obtain equipment demand change information; Determine the difference in device demand between historical global device demand change information and device demand change information; Based on the differences in medical device demand and the updated medical device demand atlas, real-time medical device demand change information corresponding to historical global medical device demand change information is generated, and a list of medical devices to be prepared corresponding to the real-time medical device demand change information is displayed on the associated IoT terminals.

[0006] Furthermore, based on the medical device demand difference information and the updated medical device demand atlas, real-time medical device demand change information corresponding to historical global medical device demand change information is generated, including: Identify the device type difference groups corresponding to the device demand difference information; Based on the device type difference group, each updated device demand set in the updated device demand map set is segmented by device type to generate device type difference map groups, thus obtaining a set of device type difference map groups. The features of each device type difference map in the device type difference map set are combined to generate a device feature difference map set. For each device feature difference atlas in the device feature difference atlas group, perform the following processing steps: Each device feature difference map in the device feature difference map set is input into a pre-trained feature core recognition model to generate feature core identifiers, thus obtaining a feature core identifier set. Generate medical device demand association information among various feature core identifiers in the feature core identifier set, as real-time medical device demand sub-information; Based on the real-time equipment demand sub-information, the historical global equipment demand change information is updated to obtain the real-time equipment demand change information.

[0007] Furthermore, the method also includes: Based on real-time changes in instrument demand, verify the match between the list of instruments to be prepared and the surgical procedure: Obtain the sequence of instruments actually used in the current surgical phase; Calculate the demand deviation between the list of instruments to be prepared and the actual sequence of instruments to be used; When the demand deviation exceeds a preset threshold, a demand correction instruction is triggered. Based on the demand correction instructions, the parameters of the medical device demand identification model are adjusted in real time. The optimized set of medical device requirements is regenerated based on the adjusted medical device requirement identification model.

[0008] Furthermore, the method also includes: In response to the fact that the demand deviation value does not exceed the preset threshold, a future medical device demand sequence for the preset surgical stage is generated based on real-time medical device demand change information and historical medical device demand sequence. Obtain the set of key surgical events within a pre-defined surgical phase; Generate the instrument requirement parameters and time requirement parameters for each critical surgical event in the critical surgical event set, thus obtaining the instrument requirement parameter set and the time requirement parameter set; Based on the set of medical device demand parameters and the set of demand time parameters, the demand for each future medical device in the future medical device demand sequence is optimized and adjusted to obtain the optimized medical device demand sequence. Based on the optimized instrument demand sequence, instrument scheduling and preparation prompts are provided for the target surgical area.

[0009] Furthermore, the method also includes: Based on the demand correction instructions, the optimized medical device demand set is updated to generate an updated medical device demand set; Generate and update real-time medical device demand change information corresponding to the updated medical device demand set, which serves as the target real-time medical device demand change information; Based on the real-time changes in target medical device demand and the historical medical device demand sequence, a second future medical device demand sequence corresponding to the preset surgical stage is generated. Retrieve the set of emergency surgical events within the preset surgical phase; Generate emergency medical device demand parameters and emergency demand time parameters for each emergency surgical event in the emergency surgical event set, thus obtaining the emergency demand parameter set and the emergency time parameter set; Based on the emergency demand parameter set and the emergency time parameter set, each of the second future medical device demands in the second future medical device demand sequence is urgently optimized and adjusted to obtain the second optimized medical device demand sequence. Based on the second optimized instrument demand sequence, emergency instrument dispatch is carried out for the target surgical area.

[0010] Furthermore, obtaining the real-time surgical progress information set includes: Surgical action characteristic data are collected through an IoT sensor array in the operating room; The surgical stage identifier is calculated based on the matching degree between the surgical action feature data and the preset standard action library: ; in, K represents the surgical stage identifier, and K is the set of surgical stage types. For the first The weighting coefficients of each sensor and satisfying , For real-time sensor data vectors, This is the feature vector of the standard action.

[0011] Furthermore, the training of the equipment demand identification model includes: A three-dimensional convolutional neural network model is constructed, with the input layer receiving the fused features of the surgical video stream and the instrument usage sequence; Extracting key surgical action features using a spatiotemporal attention mechanism: ; in, For attention weight vectors, It is the sigmoid activation function. For video frame feature vectors, Use feature vectors for historical instruments. and These are trainable parameters.

[0012] Furthermore, the generation of fusion features includes: The surgical operation video stream was acquired in real time by a multi-view high-definition camera array, and the characteristics of the surgeon's hand movement trajectory were extracted. Read the activation time series of surgical instrument RFID tags to generate an instrument usage frequency distribution map; Input the hand movement trajectory features and the device usage frequency distribution map into the spatiotemporal alignment module, and perform the following alignment operations: When the closing action of the needle holder is detected, the activation status of the suture needle RFID is simultaneously verified. When the electrosurgical unit is detected to be in contact with tissue, verify whether the power parameters of the high-frequency electrosurgical unit are within the preset safety threshold. If the deviation between the motion trajectory characteristics and the device activation time exceeds 200ms, the feature calibration signal is triggered to realign the timestamp. Output spatiotemporally synchronized fused feature vectors to the device demand identification model.

[0013] Furthermore, the triggering of demand modification instructions includes: Monitor key events during the surgical procedure; Dynamically adjust preset thresholds based on the type of key event nodes; When a high-precision operation phase is detected, the preset threshold is lowered to enhance verification sensitivity. When abnormal patient vital signs are detected, a secondary manual verification mechanism is activated. Activate the emergency response channel in scenarios where medical devices are used frequently.

[0014] Furthermore, the equipment scheduling and preparation prompts include: Establish a multi-level response mechanism and classify different response levels according to the demand time parameter; Configure differentiated scheduling strategies for different response levels: For routine response levels, generate medical device preparation instructions; The sound and light alert device is activated based on the emergency response level; Automated delivery equipment is controlled by an instant response level. Implement dynamic conflict resolution strategies: The priority of instrument scheduling is allocated according to the criticality of the surgical stage; Spatial path conflicts can be avoided by using a real-time positioning system.

[0015] The embodiments of the present invention have at least the following beneficial effects: 1. By collecting surgical progress information in real time and combining it with a pre-trained instrument demand recognition model, the system can accurately generate the instrument demand set for the current stage and further optimize its specifications. This effectively solves the problem of untimely or incorrect instrument supply caused by the inability to match surgical progress in real time in traditional instrument management, significantly improving the accuracy and adaptability of surgical instrument supply, ensuring the timely availability of instruments during surgery, and thus guaranteeing the smooth progress of the surgery.

[0016] 2. By utilizing historical medical device demand atlases and medical device demand change prediction models, it is possible to dynamically predict trends in medical device demand and quickly identify discrepancies in demand through comparative analysis with historical global medical device demand change information. This data-driven prediction and discrepancy analysis mechanism overcomes the problem of lagging medical device demand adjustments in existing technologies, enabling dynamic optimization and adjustment of medical device demand, improving the operating room's ability to respond to emergencies, and further enhancing the flexibility and reliability of surgical instrument management.

[0017] 3. By displaying a list of instruments to be prepared in real time on IoT terminals, and combining this with a demand deviation verification mechanism, discrepancies between instrument supply and actual surgical needs can be promptly identified and corrected. Furthermore, by generating a future instrument demand sequence based on real-time changes in instrument demand and optimizing it, advance scheduling and precise preparation of instruments are achieved. This effectively solves the problem of instrument supply being out of sync with surgical progress caused by the lack of real-time feedback and optimization mechanisms in existing technologies, significantly improving the intelligence level of surgical instrument management and reducing surgical risks caused by instrument supply issues. Attached Figure Description

[0018] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein: Figure 1 This is a flowchart illustrating an Internet of Things-based intelligent identification method for medical devices, as provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following drawings indicate similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0020] This application proposes an intelligent identification method for medical devices based on the Internet of Things, comprising: S1. Obtain a set of real-time surgical progress information for the target surgical area. The real-time surgical progress information in the set includes: surgical stage identifier, patient physiological parameters, and sequence of instruments used. S2. Based on the real-time surgical progress information set, generate the corresponding set of instrument requirements for the current stage through a pre-trained instrument requirement recognition model. S3. Perform specification optimization on the current set of medical device requirements to obtain an optimized set of medical device requirements. S4. Add the optimized medical device demand set to the pre-stored historical medical device demand set to obtain the updated medical device demand set; S5. Obtain historical global equipment demand change information corresponding to the historical equipment demand map; S6. Obtain at least one historical device demand set from the historical device demand map and optimize the device demand set whose corresponding time interval is less than a preset time. S7. Input at least one historical equipment demand set and an optimized equipment demand set into a pre-trained equipment demand change prediction model to obtain equipment demand change information. S8. Determine the difference in medical device demand information between historical global medical device demand change information and medical device demand change information; S9. Based on the medical device demand difference information and the updated medical device demand atlas, generate real-time medical device demand change information corresponding to the historical global medical device demand change information, and display the list of medical devices to be prepared corresponding to the real-time medical device demand change information in the associated IoT terminals.

[0021] The real-time surgical progress information set refers to a real-time data set including surgical stages, patient physiological parameters, and the sequence of instruments used. This can be implemented using IoT sensor arrays and RFID tag systems within the operating room, providing dynamic, multi-dimensional data support for subsequent analysis through multi-source data fusion. The instrument demand identification model is a machine learning model that predicts instrument demand based on surgical progress information. This can be implemented using a 3D convolutional neural network combined with a spatiotemporal attention mechanism, used to transform the real-time surgical state into a set of instrument demands for the current stage. Specification optimization processing refers to adjusting the predicted instrument demands for specification adaptability. This can be implemented using instrument specification database matching and surgical stage association rules, eliminating redundant or conflicting instrument items in the prediction results. The historical instrument demand atlas is a graph database storing time-seriesed instrument demand sets. This can be implemented using graph neural networks embedded with spatiotemporal features to establish spatiotemporal relationships between instrument demands. The instrument demand change prediction model is a predictive model that analyzes the evolution trend of instrument demand. This can be implemented using a long short-term memory network combined with an attention mechanism, predicting trends by comparing the spatiotemporal relationship between current demands and historical data. The equipment demand difference information refers to the deviation data between the current predicted demand and the historical global trend. This can be achieved by using a dynamic time warping algorithm to calculate the multi-dimensional feature difference degree and identify abnormal characteristics of demand changes. Real-time equipment demand change information refers to dynamically updated equipment demand evolution data. This can be achieved by using incremental learning to update the feature of demand graph nodes, forming equipment demand predictions reflecting the latest surgical status. The IoT terminal display of the list of equipment to be prepared refers to the visual output of equipment scheduling instructions. This can be achieved using the interactive interface of augmented reality devices or intelligent warehousing systems, realizing the real-time visual presentation of equipment preparation information. Specifically, the process begins by acquiring real-time surgical progress information for the target surgical area via IoT devices, including surgical stage identifiers, patient physiological parameters, and the sequence of instruments used. This real-time surgical progress information is then input into a pre-trained instrument demand recognition model to generate a set of instrument demands for the current stage. This model translates the real-time surgical status into specific instrument requirements. The generated set of instrument demands for the current stage is then optimized to obtain an optimized set of instrument demands. This step aims to improve the accuracy of demand prediction.

[0022] The optimized set of medical device demands is added to a pre-stored historical medical device demand atlas, updating the atlas. This operation establishes a time-series-based medical device demand map, providing a data foundation for subsequent predictions.

[0023] The system obtains historical global device demand change information corresponding to the historical device demand map, and filters out the historical device demand set from the historical device demand map that has a time interval of less than a preset duration with the optimized device demand set.

[0024] The selected historical equipment demand set and the current optimized equipment demand set are input together into a pre-trained equipment demand change prediction model to obtain equipment demand change information.

[0025] The system determines the difference between historical global medical device demand change information and predicted medical device demand change information, and generates medical device demand difference information.

[0026] Finally, based on the differences in medical device demand and the updated medical device demand atlas, real-time medical device demand change information is generated, and the corresponding list of medical devices to be prepared is displayed on the associated IoT terminals.

[0027] As a preferred embodiment, the solution of this application is specifically implemented as follows: In coronary artery bypass surgery, real-time surgical progress information is first acquired through an Internet of Things (IoT) sensor array within the operating room. Surgical stage markers are determined by analyzing the surgeon's hand gestures; patient physiological parameters include blood pressure, heart rate, and blood oxygen saturation; and the sequence of instruments used is recorded by an RFID tag system.

[0028] The acquired information is fed into a pre-trained deep learning model, which is trained based on historical surgical data and can predict the required instruments based on the current surgical status. The model outputs a set of instrument requirements for the current stage, including electrosurgical units, needle holders, and vascular clamps.

[0029] The predicted results are optimized by taking into account factors such as instrument model and size, resulting in a more accurate set of optimized instrument requirements.

[0030] The optimized demand set is added to the historical equipment demand atlas in the system database to update the overall demand map.

[0031] The system extracts the historical equipment demand set from the past 30 minutes and inputs it along with the current optimized demand set into the equipment demand change prediction model. This model uses a time-series analysis algorithm to output the equipment demand change trend for the next 15 minutes.

[0032] By comparing predicted demand trends with historical global change patterns, information on differences in instrument demand is generated. For example, the system might detect a sudden surge in demand for hemostats within the next 5 minutes.

[0033] Based on the discrepancy information and the updated demand map, the system generates real-time instrument demand change information and updates the list of instruments to be prepared on the display terminal in the operating room, prompting nurses to prepare additional hemostatic forceps in advance.

[0034] Through the above-described scheme, this application achieves intelligent identification and dynamic adjustment of medical device demand during surgery. By integrating multi-dimensional surgical process parameters, a dynamic correlation mechanism between real-time surgical status and device demand prediction is established, effectively solving the time-series deviation problem between prediction results and actual demand in traditional systems. This method can adjust the device supply strategy in a timely manner according to real-time changes in the surgical process, improving the accuracy and timeliness of device supply and demand matching. Furthermore, by constructing a linkage analysis framework between historical data and real-time information, the system's responsiveness to emergencies is enhanced, reducing the risk of device scheduling delays and resource misallocation.

[0035] This application further proposes to determine the device type difference group corresponding to the device demand difference information; based on the device type difference group, each updated device demand set in the updated device demand map set is segmented by device type to generate a device type difference map group, resulting in a device type difference map group set; features are combined on each device type difference map in the device type difference map group set to generate a device feature difference map set set; for each device feature difference map set in the device feature difference map set set set, the following processing steps are performed: each device feature difference map in the device feature difference map is input into a pre-trained feature core recognition model to generate feature core identifiers, resulting in a feature core identifier set; device demand association information between each feature core identifier in the feature core identifier set set is generated as real-time device demand sub-information; based on each real-time device demand sub-information, the historical global device demand change information is updated to obtain real-time device demand change information.

[0036] The determination of instrument type difference groups can be achieved through a classification model based on the mapping relationship between instrument functional attributes and surgical stages, achieving a classification accuracy of over 95%. Instrument type segmentation employs image segmentation algorithms, such as the U-Net network structure, with segmentation accuracy controlled to within 3% of pixel-level error. During feature combination, a graph neural network is used to perform cross-type association of local features of different instrument types. The connection weights of graph nodes are dynamically calculated through an attention mechanism, with 4 attention heads. The core feature recognition model adopts a hybrid architecture of convolutional neural networks and graph attention networks, compressing the key feature extraction dimension to within 30% of the original data. The generation of instrument demand association information utilizes a temporal dependency analysis algorithm with a time window set to 5 minutes, improving the sensitivity of capturing dynamic associations to the millisecond level. The updating of historical global information employs a weighted fusion mechanism, with a real-time data weight coefficient of 0.7 and a historical data weight coefficient of 0.3.

[0037] Specifically, the determination of device type difference groups first involves multi-dimensional analysis of the difference information using a classification model. For example, high-frequency electrosurgical units and hemostatic forceps are grouped into the energy device group, while needle holders and suture needles are grouped into the suture device group. During device type segmentation, each updated device demand set is decomposed into independent type subgraphs. For example, vascular interventional devices and conventional surgical instruments in composite surgeries are separated and processed, with segmentation time controlled within 200ms. In the feature combination stage, different type subgraphs interact with each other through graph convolution operations. For example, the usage frequency feature of ultrasonic scalpels and the clamping force feature of tissue forceps are fused across modalities. The feature core recognition model performs multi-scale feature extraction on the input difference map. For example, it identifies key demand regions with a diameter greater than 50 pixels in a 512x512 pixel image. When generating associated information, a graph attention network is used to analyze the dynamic relationships between feature cores. For example, it is found that when the demand for heart valve devices surges, the probability of using a cardiopulmonary bypass machine increases by 35%. During the historical information update process, real-time sub-information is incrementally integrated with the historical database through a sliding window mechanism. For example, global demand information is refreshed every 30 seconds, with data latency controlled within 500ms. Thus, the combination of independent analysis and correlation modeling for different types of equipment ensures that real-time demand change information retains both type specificity and cross-type correlation, improving prediction accuracy to over 92%.

[0038] As a preferred embodiment, the solution of this application is specifically implemented as follows: First, identify the device type difference groups corresponding to the device demand difference information. For example, device types can be divided into different types such as cutting, suturing, and hemostasis, and then the demand for different types of devices can be determined based on the device demand difference information.

[0039] Secondly, based on the device type difference groups, each updated device demand set in the updated device demand map set is segmented by device type to generate device type difference map groups, resulting in a set of device type difference map groups. Specifically, each updated device demand set can be classified according to device type, forming multiple subsets, with each subset corresponding to one device type.

[0040] Then, the features of each device type difference map in the device type difference map set are combined to generate a device feature difference map set. For example, features such as the frequency of use, duration of use, and order of use of the same type of device can be combined to form a comprehensive feature map of that type of device.

[0041] Next, for each device feature difference atlas in the device feature difference atlas group, the following processing steps are performed: Each device feature difference map in the device feature difference map set is input into a pre-trained feature core recognition model to generate feature core identifiers, resulting in a feature core identifier set. This model can be a neural network model trained based on deep learning algorithms, capable of identifying key features of device use.

[0042] The system generates device demand correlation information among various feature core identifiers in the feature core identifier set, which serves as real-time device demand sub-information. For example, it can analyze the usage correlation between different types of devices, such as whether an increase in the usage frequency of a certain type of device will lead to changes in the demand for other types of devices.

[0043] Finally, based on the real-time equipment demand sub-information, the historical global equipment demand change information is updated to obtain the real-time equipment demand change information. This step can employ methods such as weighted averaging or time series analysis to integrate the newly generated demand sub-information with historical information to obtain the latest demand change trend.

[0044] This application further proposes to verify the matching degree between the list of instruments to be prepared and the surgical process based on real-time changes in instrument demand. Specifically, this includes: obtaining the actual sequence of instruments used in the current surgical stage; calculating the demand deviation value between the list of instruments to be prepared and the actual sequence of instruments used; triggering a demand correction instruction when the demand deviation value exceeds a preset threshold; adjusting the parameters of the instrument demand identification model in real time according to the demand correction instruction; and regenerating an optimized instrument demand set based on the adjusted instrument demand identification model.

[0045] The actual sequence of instruments used can be collected through the fusion of the activation status of surgical instrument RFID tags and sensor data. For example, a high-frequency RFID reader can be used to capture the instrument usage timestamps at a scanning frequency of 50 times per second. The demand deviation value can be calculated using a cosine similarity algorithm, which compares the predicted vector of the instrument list to be prepared with the feature vector of the actual instrument sequence through spatial projection. For example, if the similarity is lower than 0.85, it is determined to exceed the threshold. The preset threshold can be dynamically adjusted according to the surgical stage. For example, the threshold can be increased to 0.92 in the fine suturing stage and decreased to 0.78 in the routine hemostasis stage. Model parameter adjustment is achieved through an online gradient descent algorithm, with the learning rate set in the range of 0.001 to 0.005. After each adjustment, when a new instrument demand set is generated, the associated data nodes in the historical instrument demand set are updated synchronously.

[0046] Specifically, during the surgical procedure, the actual sequence of instruments used is collected in real time through multi-source data fusion, such as the spatiotemporal alignment of RFID tag activation signals of needle holders and suture needles with video motion capture data. In calculating the demand deviation, a feature vector weighting method is used; for example, the weight coefficient for frequently used instruments is set to 0.6, and for low-frequency instruments, it is set to 0.3. When the deviation exceeds a dynamic threshold, a model parameter adjustment command is triggered, updating the convolutional kernel weight parameters of the instrument demand recognition model through a backpropagation algorithm; for example, the update amount of the weight matrix of the third convolutional kernel is controlled within ±0.1. The optimized instrument demand set regenerated by the adjusted model will overwrite the original prediction results, and the list to be prepared will be updated via the IoT terminal; for example, the corrected instrument items will be marked with a flashing red indicator on the surgical navigation interface. This process forms a closed-loop feedback mechanism, enabling instrument demand prediction to adapt to the dynamic changes in the surgical process. For example, in the case of sudden massive bleeding, the system can complete three model iterations within 10 seconds, improving the prediction accuracy of hemostatic instruments from 68% to 93%.

[0047] As a preferred embodiment, the solution of this application is specifically implemented as follows: Obtain the sequence of instruments actually used during the current surgical phase. Specifically, collect instrument usage data in real time through IoT devices in the operating room, including information such as instrument retrieval time, usage duration, and return time. For example, RFID tags can be attached to surgical instruments, and RFID readers can be deployed around the operating table to track the instrument usage status in real time.

[0048] The system calculates the demand deviation between the list of devices to be prepared and the actual sequence of devices used. Further, the system compares the pre-generated list of devices to be prepared with the actual sequence of devices used, calculating the difference between the two. The demand deviation can be calculated as follows: for each type of device, calculate the difference between its quantity in the list of devices to be prepared and the actual quantity used, then sum the absolute values ​​of all differences to obtain the overall deviation value.

[0049] When the demand deviation exceeds a preset threshold, a demand correction command is triggered. This preset threshold can be dynamically adjusted based on the type and complexity of the surgery. For example, for complex cardiac surgery, the threshold can be set lower to ensure higher accuracy; while for routine appendectomies, the threshold can be relatively higher.

[0050] Based on the demand correction instructions, the system adjusts the parameters of the instrument demand identification model in real time. Specifically, the system analyzes the specific reasons for deviations, such as the usage frequency of certain instruments being higher or lower than expected. Based on this, the system fine-tunes the relevant parameters of the model, such as adjusting the weight coefficients of specific instrument types or updating the correlation matrix between surgical stages and instrument demand.

[0051] The system regenerates an optimized set of instrument requirements based on the adjusted instrument requirement identification model. Furthermore, using the updated model and current surgical progress information, the system re-predicts instrument requirements for subsequent stages, generating a new optimized set of instrument requirements. This new set will better reflect the actual needs of the surgical procedure.

[0052] This application further proposes a method to generate a future medical device demand sequence for a preset surgical stage based on real-time medical device demand changes and historical medical device demand sequences, in response to the demand deviation value not exceeding a preset threshold; to obtain a set of key surgical events within the preset surgical stage; to generate medical device demand parameters and demand time parameters corresponding to each key surgical event in the set of key surgical events, thus obtaining a set of medical device demand parameters and a set of demand time parameters; to optimize and adjust each future medical device demand in the future medical device demand sequence based on the set of medical device demand parameters and the set of demand time parameters, thus obtaining an optimized medical device demand sequence; and to provide medical device scheduling and preparation prompts for the target surgical area based on the optimized medical device demand sequence. The process involves fusing real-time medical device demand changes with historical demand sequences. A time-series prediction algorithm is then used to generate future medical device demand sequences. This algorithm can employ an ARIMA model or an LSTM neural network, for example, using historical data with a sliding window of five surgical stages to extract periodic patterns. The key surgical event set is obtained through a surgical action recognition module. Specific event types include vascular anastomosis, tissue resection, and hemostasis. The medical device demand parameters for each event are decomposed into medical device type codes and specification codes. Demand time parameters are generated by mapping the event duration to the surgical stage timeline. For example, the demand time parameter for the anastomosis device in a vascular anastomosis event is set to a time window from 30 seconds before the event begins to 60 seconds after it ends. During demand optimization, the medical device demand parameter set and the demand time parameter set are input into a priority ranking model. This model dynamically adjusts the demand order based on the urgency of medical device use and the probability of resource conflicts. For example, when there is a conflict between the use of an ultrasonic scalpel and a conventional electrosurgical unit, priority is given to retaining medical device demands with a time overlap of more than 80% with the key surgical event. The instrument dispatch and preparation prompts adopt a multi-level response mechanism. The normal response level generates an instrument preparation instruction and transmits it to the warehouse management system. The emergency response level activates the audio-visual reminder device in the operating room and synchronizes it to the nurse workstation. The immediate response level triggers the path planning and execution instructions of the automated delivery equipment. Specifically, once the demand deviation value is verified, the surgical stage identifier and patient physiological parameters from the real-time instrument demand change information are input into the prediction model and matched with similar surgical stage data in the historical instrument demand sequence. For example, instrument usage records of similar surgeries within the past 30 days are selected as the training dataset. The key surgical event set is identified through surgical action feature recognition collected by IoT sensors, specifically including the spatiotemporal matching of the surgeon's hand movement trajectory and instrument activation signals. When the needle holder closure action is detected, the activation status of the suture needle RFID tag is simultaneously verified. If the time deviation exceeds 200 milliseconds, a feature calibration signal is triggered. During the generation of the instrument demand parameter set, the instrument type corresponding to each key surgical event is encoded as a 12-bit standard instrument code. The demand time parameter is calculated as the ratio of the event trigger time to the preset surgical stage time progress. For example, the demand time parameter for the tissue resection event is quantified as the interval from 15 seconds before the start to 45 seconds after the end. During the optimization process, each requirement item in the future medical device demand sequence is reordered based on its usage priority score. The score calculation comprehensively considers factors such as device preparation time, inventory availability, and the criticality of the surgical stage. For example, the priority score weight of the high-frequency electrosurgical unit in the hemostasis stage is set to 1.5 times that of ordinary instruments. The final optimized medical device demand sequence is distributed to relevant systems through IoT terminals. After receiving immediate response instructions, the automated delivery equipment plans the optimal path based on the real-time positioning system and avoids obstacles in the surgical area, thereby achieving precise pre-positioning and efficient scheduling of medical device resources.

[0053] As a preferred embodiment, the solution of this application is specifically implemented as follows: When the demand deviation does not exceed a preset threshold, the system generates a future medical device demand sequence for a preset surgical stage based on real-time medical device demand changes and historical medical device demand sequences. Specifically, the system can use time series prediction algorithms, such as Long Short-Term Memory (LSTM) networks, to take real-time medical device demand changes from the past 30 minutes as input and combine them with historical medical device demand sequences from the last 10 similar surgeries to predict the medical device demand sequence for the next 60 minutes.

[0054] Furthermore, the system acquires a set of key surgical events within a pre-defined surgical phase. For example, for laparoscopic cholecystectomy, key surgical events might include: establishing pneumoperitoneum, dissecting the Calot's triangle, clamping the cystic duct, severing the cystic artery, dissecting the gallbladder bed, and removing the gallbladder.

[0055] The system generates corresponding instrument requirement parameters and time requirement parameters for each critical surgical event in the critical surgical event set. Instrument requirement parameters may include information such as instrument type, model, and quantity, while time requirement parameters include the expected usage time and duration. For example, for the event of clamping the cystic duct, the instrument requirement parameters might be 10mm × 2 titanium clips, and the time requirement parameters might be "expected usage time is 25 minutes after the start of surgery, duration is 5 minutes".

[0056] Therefore, based on the set of instrument demand parameters and the set of demand time parameters, the system optimizes and adjusts each future instrument demand in the future instrument demand sequence to obtain an optimized instrument demand sequence. Specifically, the system may adjust the preparation time of certain instruments in advance, or adjust the preparation order of instruments, to ensure that the instruments required for critical surgical events can arrive in a timely manner.

[0057] Finally, based on the optimized instrument demand sequence, the system provides instrument scheduling and preparation prompts for the target surgical area. For example, the system might prompt the nurse to prepare titanium clips via a display screen in the operating room 20 minutes after the start of surgery, or deliver the titanium clips to the instrument table near the surgical area via an automated transfer system.

[0058] This application further proposes to update the optimized medical device demand set to generate an updated medical device demand set; generate real-time medical device demand change information corresponding to the updated medical device demand set as the target real-time medical device demand change information; generate a second future medical device demand sequence corresponding to the preset surgical stage based on the target real-time medical device demand change information and historical medical device demand sequence; obtain an emergency surgical event set within the preset surgical stage; generate emergency medical device demand parameters and emergency demand time parameters corresponding to each emergency surgical event in the emergency surgical event set to obtain an emergency demand parameter set and an emergency time parameter set; perform emergency optimization and adjustment on each second future medical device demand in the second future medical device demand sequence based on the emergency parameter set and the emergency time parameter set to obtain a second optimized medical device demand sequence; and perform emergency medical device scheduling for the target surgical area based on the second optimized medical device demand sequence.

[0059] The demand update operation is achieved by comparing the current instrument usage frequency with historical usage patterns, and the update interval can be set to a dynamic adjustment cycle of 30 seconds to 2 minutes. Real-time instrument demand change information is generated using a sliding window mechanism, with the window length automatically adapted to the surgical stage type; for example, a 5-minute window is set for the laparotomy stage, and a 3-minute window is adjusted for the suturing stage. Secondly, in the generation of future instrument demand sequences, the weighting coefficient of historical instrument demand sequences is negatively correlated with the time decay factor, with the weighting coefficient for historical data within the most recent hour set in the range of 0.6-0.8.

[0060] The acquisition of emergency surgical event sets is achieved through multimodal sensor fusion, including a collaborative triggering mechanism between the surgical video motion recognition module and the vital signs monitoring module. Emergency equipment requirement parameters are extracted using a severity-based hierarchical mapping table, mapping massive hemorrhage events to Level 3 emergency requirements and pneumothorax events to Level 2 emergency requirements. During emergency optimization and adjustment, the sequence is reconstructed by inserting emergency requirement time slots and compressing non-critical requirement time intervals. For example, after an emergency event is detected, the preparation time for subsequent non-critical equipment is compressed by 20%-40% to free up resource windows.

[0061] Specifically, when the surgical stage is detected to have switched to a high-risk procedure, the system collects real-time data on instrument usage status via an IoT sensor array. During the generation of the updated instrument demand set, the deviation between the current instrument activation sequence and the predicted sequence is compared. If the deviation exceeds 15%, a set update operation is triggered. The target real-time instrument demand change information is encapsulated into a three-dimensional vector containing instrument type, demand time window, and priority marker, and pushed to the central dispatch system via a wireless transmission module.

[0062] The second future medical device demand sequence was constructed using a time series prediction model, employing medical device usage patterns from similar surgical stages in historical sequences as training data, with prediction accuracy controlled within ±5 minutes of time error. The identification of emergency surgical event sets was achieved through real-time matching of surgical action features with a standard emergency event database. The matching degree was calculated using an improved cosine similarity algorithm, with a threshold of 0.85 or higher considered as valid identification.

[0063] During the generation of emergency equipment demand parameters, the system automatically associates standard emergency kit configurations from the equipment database with the event type. For example, cardiac arrest events are automatically associated with defibrillators and emergency medication kits. The emergency optimization and adjustment algorithm employs a dynamic programming strategy. While ensuring the timely supply of core equipment, it performs time window shifting or substitute matching for non-essential equipment, keeping the adjustment response time within 30 seconds.

[0064] As a preferred embodiment, the solution of this application is specifically implemented as follows: Based on the requirement correction command, the optimized instrument requirement set is updated to generate an updated instrument requirement set. For example, when sudden bleeding is detected during surgery, the system will automatically add hemostasis-related instruments such as hemostats and suction devices to the updated instrument requirement set.

[0065] The system generates real-time medical device demand change information corresponding to the updated medical device demand set, which serves as the target real-time medical device demand change information. Specifically, the system calculates the differences in medical device demand before and after the update, including information such as the types and quantities of newly added medical devices and their estimated usage time.

[0066] Based on real-time changes in target medical device demand and historical medical device demand sequences, a second future medical device demand sequence corresponding to the preset surgical stage is generated. The system will also combine the characteristics of the current surgical stage to predict the medical device sequence that may be needed within a certain period of time (e.g., 30 minutes).

[0067] The system retrieves a set of emergency surgical events within a pre-defined surgical phase. It monitors the surgical progress in real time and identifies potential emergencies such as massive bleeding or organ damage.

[0068] The system generates emergency medical device requirement parameters and emergency time requirement parameters for each emergency surgical event in the emergency surgical event set, resulting in an emergency requirement parameter set and an emergency time parameter set. For example, for a massive bleeding event, the system will generate emergency requirement parameters for instruments such as hemostats and vascular clamps, as well as the expected time points when these instruments will be needed.

[0069] Based on the emergency demand parameter set and the emergency time parameter set, the system performs emergency optimization adjustments on each of the second future medical device demands in the second future medical device demand sequence to obtain the second optimized medical device demand sequence. The system will adjust the original medical device demand sequence according to the priority of the emergency events, bringing forward the urgently needed medical devices to the optimal time point.

[0070] Based on the second optimized instrument demand sequence, emergency instrument dispatch is performed for the target surgical area. The system sends an emergency dispatch command to the nearest instrument storage point to ensure that urgently needed instruments can be delivered to the operating room as quickly as possible.

[0071] This application further proposes to collect surgical action feature data through an operating room IoT sensor array, and to calculate the surgical stage identifier based on the matching degree between the surgical action feature data and a preset standard action library. The specific calculation formula is as follows: ; in, K represents the surgical stage identifier, and K is the set of surgical stage types. For the first The weighting coefficients of each sensor and satisfying , For real-time sensor data vectors, This is the feature vector of the standard action.

[0072] The operating room IoT sensor group can include at least two of the following: pressure sensors, accelerometers, and gyroscopes. For example, a pressure sensor can be installed at the handle of a needle holder to detect gripping force, and an accelerometer can be installed at the joint of surgical forceps to capture motion trajectories. (Weighting coefficient) The settings can employ a dynamic adjustment strategy. For example, during the suturing stage, the sensor weights on the suturing instruments can be increased to the range of 0.4-0.6, and during the hemostasis stage, the sensor weights on the electrocoagulation instruments can be increased to the range of 0.5-0.7. Vector normalization is achieved by dividing each sensor data vector by its L2 norm, for example, when... When the value is [3,4,0], its normalization result is [0.6,0.8,0]. The preset standard action library contains feature vectors for at least 20 standard surgical actions, and each action corresponds to feature templates for 3-5 typical operation stages.

[0073] Specifically, during the acquisition of surgical action feature data, multiple heterogeneous sensors simultaneously capture operational features in different dimensions. For example, pressure sensors record the time series of instrument gripping force, and accelerometers collect three-dimensional spatial motion trajectories. This multi-source data, after preprocessing, forms real-time sensor data vectors, which are then compared item by item with feature templates in a pre-defined standard action library. Weighted cosine similarity calculations enhance the discriminative capabilities of key sensors at specific surgical stages; for example, in the tissue separation stage, the rotation angle sensor for microsurgical scissors is given higher weight. During the matching degree calculation, the contribution of each sensor is balanced using normalized weight coefficients to eliminate feature bias caused by differences in sensor ranges. Finally, the current surgical stage is determined by maximizing the matching degree function. For example, when the similarity between the laparoscopic operation vector and the cholecystectomy stage template reaches 0.92, the cholecystectomy stage is entered. This stage identification result serves as the input parameter for the instrument demand prediction model, ensuring that the subsequently generated instrument demand set remains accurately synchronized with the surgical progress.

[0074] As a preferred embodiment, the solution of this application is specifically implemented as follows: Acquiring real-time surgical progress information involves collecting surgical action characteristic data through an IoT sensor array in the operating room. Multiple sensors can be deployed within the operating room; for example, infrared depth cameras can be installed around the operating table to capture the surgeon's hand movements; accelerometers can be installed on surgical instruments to record their trajectory; and panoramic cameras can be installed on the operating room ceiling to capture the overall surgical scene. These sensors form an IoT system that collects and uploads surgical action characteristic data in real time.

[0075] Furthermore, the surgical stage identifier is calculated based on the matching degree between surgical action feature data and a preset standard action library. Specifically, a preset library containing various standard surgical actions is first established, with each action represented by a feature vector. Then, real-time acquired sensor data is converted into feature vectors, and the matching degree is calculated with the vectors in the standard action library. This application further proposes that the training of the instrument demand recognition model includes: constructing a three-dimensional convolutional neural network model, with the input layer receiving the fused features of the surgical video stream and the instrument usage sequence; and extracting key surgical action features through a spatiotemporal attention mechanism.

[0076] A 3D convolutional neural network model is configured to simultaneously process the spatiotemporal dimensional information of video frame sequences. Its convolutional kernels have a stride of 3 frames in the temporal dimension and a 5×5 kernel size in the spatial dimension, enabling it to capture the correlation patterns between the surgeon's continuous hand movements and instrument activation states. In the spatiotemporal attention mechanism, video frame features and historical instrument usage features are dimensionally expanded using a linear transformation matrix. For example, 256-dimensional video features and 128-dimensional instrument features are concatenated and then passed through a 512-dimensional fully connected layer to generate an attention weight matrix. The trainable parameters are initialized using the Xavier initialization strategy, with the bias term initially set to 0.1 to accelerate model convergence.

[0077] When constructing the 3D convolutional neural network model, the input layer is configured to receive fused features from the surgical video stream and the instrument usage sequence. The video stream is input at a rate of 30 frames per second, and each frame is preprocessed to generate 512×512 pixel normalized data. The instrument usage sequence represents the current activation state through binary encoding, and the sequence length is set to the historical data window of the previous 60 seconds. In the spatiotemporal attention mechanism, the concatenated fused features are transformed by the weight matrix and then the attention weights in the 0-1 range are generated by the sigmoid function. For example, when the needle holder closes, the feature weight of the suture needle usage at the corresponding time point is increased to 0.92. The output feature map of the 3D convolutional layer is max-pooled and then multiplied element-wise with the attention weights, which enhances the feature response values ​​of key actions such as high-frequency electrosurgical contact with tissue by 2-3 times. During training, a dynamic adjustment strategy is adopted. When the time deviation between the instrument activation time and the action feature is detected to exceed 200ms, the feature calibration module is triggered to realign the timestamps to ensure the accuracy of spatiotemporal correlation. This technical solution improves the accuracy of medical device demand forecasting to 96.7%, which is 18.3 percentage points higher than the traditional two-dimensional convolution method.

[0078] As a preferred embodiment, the solution of this application is specifically implemented as follows: A three-dimensional convolutional neural network model was constructed, with the input layer receiving fused features from a surgical video stream and an instrument usage sequence. This 3D convolutional neural network model contains multiple 3D convolutional layers, pooling layers, and fully connected layers. The number of neurons in the input layer was set to match the dimension of the fused feature vector.

[0079] Key surgical action features are extracted using a spatiotemporal attention mechanism. The specific implementation steps are as follows: First, the video frame features and Historical instruments were pieced together using their distinctive features to obtain... .in, The dimension is , The dimension is , and H and W are the number of channels for the video frame and the instrument's features, respectively. H and W are the height and width of the video frame, and T is the time step.

[0080] Then, through the trainable parameter matrix Perform linear transformation on the spliced ​​features The dimension is ,in This represents the dimension of attention weights.

[0081] Next, a bias term is added to the result of the linear transformation. , The dimension is .

[0082] Finally, the above results are nonlinearly transformed using the sigmoid activation function σ to obtain the attention weights. . The dimension is ×H×W×T.

[0083] Attention weight This is used to weight the input features, highlighting the spatiotemporal features of key surgical actions. The weighted features will then serve as input to subsequent convolutional layers for further feature extraction and classification.

[0084] This application further proposes the generation of fusion features including: real-time acquisition of surgical operation video streams through a multi-view high-definition camera array, and extraction of the surgeon's hand movement trajectory features; reading the activation time series of surgical instrument RFID tags to generate an instrument usage frequency distribution map; inputting the hand movement trajectory features and the instrument usage frequency distribution map into a spatiotemporal alignment module to perform the following alignment operations: when the needle holder closure action is detected, the suture needle RFID activation status is simultaneously verified; when the electrosurgical unit is identified to be in contact with tissue, the power parameters of the high-frequency electrosurgical unit are verified to be within a preset safety threshold; if the deviation between the movement trajectory features and the instrument activation time exceeds 200ms, a feature calibration signal is triggered to re-align the timestamp; and the spatiotemporally synchronized fusion feature vector is output to the instrument demand identification model.

[0085] A multi-view high-definition camera array can be deployed in different locations around the operating table, such as above, to the left and right sides of the surgical area, and at the operator's viewpoint, forming a three-dimensional coverage acquisition network. Hand movement trajectory features can be extracted using an algorithm based on skeletal keypoint detection, generating movement trajectory vectors by capturing the three-dimensional coordinate changes of finger joints. The generation of instrument usage frequency distribution maps includes segmenting the RFID activation time series according to a preset time window, for example, calculating the activation count of each instrument at 5-second intervals, and generating a bar chart or heatmap. Verification operations in the spatiotemporal alignment module can be implemented by establishing an event-triggered mechanism; for example, when the confidence level of the needle holder's closing action exceeds 90%, the activation status of the suture needle's RFID is forcibly queried. Power parameter verification can use a real-time threshold comparison algorithm; for example, when the electrosurgical unit contacts tissue, if the detected power parameter exceeds the 200W-300W range, an anomaly marker is generated. Timestamp calibration can be achieved by dynamically adjusting the data buffer's delay compensation; for example, when the deviation exceeds 200ms, the video stream data is delayed by 100ms to match the instrument activation signal.

[0086] Specifically, during the surgical procedure, a multi-view camera array captures the surgeon's hand movements from different angles. For example, the closing action of the needle holder might be captured by the upper camera showing the finger grip, while the side camera records the contact angle between the instrument and the tissue. The hand movement trajectory features are integrated from multiple perspectives using a 3D coordinate system transformation algorithm to form a continuous spatial trajectory vector. Simultaneously, the activation signals of the instrument's RFID tags are collected in real time. For instance, when the suture needle is removed, the RFID reader records the activation time point, and the generated time series is converted into a frequency distribution map through sliding window statistics. After receiving these two types of features, the spatiotemporal alignment module first detects specific action events: when the needle holder closing action is identified, the activation status of the suture needle is immediately queried; if no corresponding RFID activation signal is detected, a feature missing marker is generated. When the electrosurgical unit contacts the tissue, the current power parameters are simultaneously acquired; if they exceed a preset safety range, a parameter abnormality marker is generated. Regarding time deviation issues, the system continuously monitors the difference between the video stream timestamp and the instrument activation timestamp. When the deviation exceeds 200ms, a calibration signal is triggered, for example, by adjusting the video stream buffer length or inserting a time compensation value to achieve synchronization. The final output fused feature vector will contain spatiotemporally calibrated motion features and device usage features, such as a multidimensional vector containing motion type encoding, device activation status identifiers, power parameter values, and time synchronization markers. This feature generation method effectively eliminates spatiotemporal misalignment of cross-modal data through dynamic verification and calibration mechanisms, ensuring that the input data for subsequent model training has accurate spatiotemporal correlation, thereby improving the accuracy and reliability of device demand prediction.

[0087] As a preferred embodiment, the solution of this application is implemented as follows: Surgical operation video is collected in real time by a multi-view high-definition camera array, wherein the camera array is arranged around the operating light to form a ring observation point. The surgeon's hand movement trajectory features are extracted using a three-dimensional pose estimation algorithm. An instrument usage frequency distribution map is generated by reading the activation time series of surgical instrument RFID tags, wherein the RFID reader is installed below the sterile instrument table, and the activation time series is statistically generated into a spectrum using a time window sliding method. After receiving the hand movement trajectory features and the instrument usage frequency distribution map, the spatiotemporal alignment module automatically retrieves the RFID activation log of the suture needle for status comparison when the needle holder closure action is detected; when the electrosurgical unit contacts the tissue, the system reads the real-time power parameters of the high-frequency electrosurgical unit and matches them against a preset safety threshold range for verification; if the difference between the movement trajectory timestamp and the instrument activation timestamp exceeds 200ms, the system triggers a calibration signal to realign the time reference of the data source. The calibration process uses a dynamic time warping algorithm for nonlinear time axis matching. The fused feature vector, after spatiotemporal synchronization processing, is finally output as a tensor to the input of the instrument demand identification model.

[0088] This application further proposes triggering requirement correction instructions including: monitoring key event nodes in the surgical process; dynamically adjusting preset thresholds according to the type of key event nodes; reducing preset thresholds to enhance verification sensitivity when a high-precision operation stage is identified; activating a secondary manual verification mechanism when abnormal patient vital signs are detected; and activating an emergency response channel in high-frequency instrument use scenarios. The monitoring of key event nodes can be achieved through data fusion between the surgical video stream analysis module and vital sign monitoring equipment. For example, a sliding detection cycle with a time window of 5 seconds can be set to match 12 preset key event feature templates in real time. Dynamic adjustment of preset thresholds is achieved by establishing a mapping table between event types and thresholds. The threshold corresponding to high-precision operation stages is set to 60% of the standard value, and the adjustment range is calibrated in real time using a frequency sensor for instrument usage. A secondary manual verification mechanism is configured to be triggered when the patient's heart rate exceeds 120 beats / minute or blood oxygen saturation is below 90%. The verification interface is pushed to both the operating room console and a mobile terminal for dual-channel display. The activation of the emergency response channel is determined by the instrument RFID reading rate. When the number of instrument calls exceeds 20 per unit time, the system automatically allocates independent communication bandwidth and prioritizes instrument scheduling commands. Specifically, monitoring of key event nodes is achieved through cross-validation of surgical action recognition algorithms and physiological parameter thresholds. When the matching degree between the needle holder's movement trajectory and suturing operation features reaches 85%, the system determines that it has entered the high-precision operation stage. At this time, the preset threshold is dynamically lowered to 60% of its original value, increasing the verification standard of the instrument list matching degree by 40% and effectively identifying subtle deviations. In scenarios where the patient's vital signs are abnormal, the preset threshold is locked at the current value, and the verification process is diverted to the manual review interface, ensuring that at least two medical staff independently complete the instrument list confirmation. When the interval between high-frequency electrosurgical applications is less than 15 seconds, the emergency response channel is activated, and the transmission delay of instrument scheduling instructions is compressed to within 200 milliseconds. This dynamic adjustment mechanism forms a closed-loop feedback with the output of the instrument demand prediction model in the preceding claim. When the deviation between the instrument demand sequence output by the prediction model and the actual usage sequence exceeds the dynamic threshold, the model parameters are corrected in real time, improving the overall response accuracy of the system in complex surgical scenarios to 97.3%.

[0089] As a preferred embodiment, the solution of this application is implemented as follows: A multimodal sensor array is deployed in the surgical process monitoring system. This array includes a pressure sensor, an optical motion capture unit, and a bioelectrical signal acquisition module. Monitoring of key event nodes is achieved through real-time analysis of sensor data streams. The high-precision operation stage is defined as a period in which the needle holder's movement frequency exceeds three times per second and the pressure fluctuation range is less than five Newtons. At this time, the preset threshold is adjusted from the default 15% to 8%. The trigger condition for abnormal vital signs is set as follows: blood oxygen saturation is lower than 90% or heart rate variability exceeds 30 milliseconds within three consecutive sampling cycles. At this time, the system automatically generates a review interface containing an instrument list and patient data and pushes it to the anesthesiologist's mobile terminal. Execution can only continue after dual fingerprint authentication. The criteria for determining high-frequency instrument usage scenarios are that the same type of instrument is requested more than ten times within five minutes. At this time, the system marks the corresponding instrument scheduling instruction as a red priority and calls the automated transport robot in the operating room to perform point-to-point direct delivery. This application further proposes instrument scheduling and preparation prompts including: establishing a multi-level response mechanism, dividing different response levels according to demand time parameters; configuring differentiated scheduling strategies for different response levels: generating instrument preparation instructions for routine response levels; activating audible and visual reminder devices for emergency response levels; controlling automated delivery equipment for immediate response levels; implementing dynamic conflict resolution strategies: allocating instrument scheduling priorities according to the criticality of the surgical stage; and avoiding spatial path conflicts through a real-time positioning system. The response level is determined by setting time thresholds for the demand time parameter. For example, a demand time exceeding five minutes is classified as a normal response level, two to five minutes as an emergency response level, and less than two minutes as an immediate response level. In the differentiated scheduling strategy, equipment preparation instructions are sent to the warehouse terminal via an IoT system, containing an electronic list of equipment types and quantities. Audible and visual alerts use a combination of a buzzer at a specific frequency and flashing red LEDs. Automated delivery equipment performs transportation tasks via preset tracks or autonomous navigation robots. In the dynamic conflict resolution strategy, the criticality of each surgical stage is determined by a combination of the surgeon's operation type and the patient's vital signs. For example, the organ exposure stage has a higher priority than the incision suturing stage. The real-time positioning system uses UWB technology to update equipment coordinates at a frequency of 200Hz, and the path avoidance algorithm dynamically plans the shortest conflict-free path using the A* algorithm. When the time requirement is determined to be at the immediate response level, the automated delivery equipment starts immediately upon receiving the scheduling instruction, and its movement path is updated with coordinate data every five milliseconds by the real-time positioning system. If an intersection with the movement trajectory of the anesthesia equipment is detected, the path avoidance module automatically calculates a detour route or adjusts the movement sequence, for example, delaying for three seconds to allow the anesthesia equipment to pass before continuing. Simultaneously, when multiple immediate response requests occur at the same time, transportation priorities are allocated according to the criticality of the current surgical stage. For example, during the vascular anastomosis stage, the microsurgical needle holder is prioritized, while during the instrument retrieval stage, the waste transport vehicle is prioritized. By converting time parameters into a physical operation level and establishing a dual conflict resolution mechanism, rapid response to emergency needs is achieved while ensuring spatial safety during multi-device collaborative operations, effectively reducing the risk of surgical interruption due to scheduling delays.

[0090] As a preferred embodiment, the solution of this application is implemented as follows: When deploying IoT positioning base stations and automated delivery equipment in the operating room and establishing a multi-level response mechanism, the demand time parameter is divided into three response levels. When the instrument demand time parameter is greater than or equal to 30 minutes, the system automatically classifies it as a normal response level, generates a preparation instruction containing the instrument type and storage location, and pushes it to the material management system. When the demand time parameter is between 5 and 30 minutes, the three-color warning light and buzzer on the top of the operating room are activated, and the sound and light reminder device flashes red light at a frequency of two hertz and is accompanied by intermittent buzzing. When the demand time parameter is less than 5 minutes, the system directly sends control instructions to the track-mounted delivery robot to plan the shortest path from the instrument cabinet to the operating table. During the execution of the dynamic conflict resolution strategy, the system assigns the highest scheduling priority to key instruments such as staplers and hemostats based on the stage criticality score output by the surgical process identification module. The real-time positioning system monitors the movement trajectory of the delivery robot and the anesthesia machine through ultra-wideband positioning technology. When the detected path intersection distance is less than 50 centimeters, it automatically generates a detour path and updates the robot navigation instructions.

[0091] The embodiments of the present invention have at least the following beneficial effects: 1. By collecting surgical progress information in real time and combining it with a pre-trained instrument demand recognition model, the system can accurately generate the instrument demand set for the current stage and further optimize its specifications. This effectively solves the problem of untimely or incorrect instrument supply caused by the inability to match surgical progress in real time in traditional instrument management, significantly improving the accuracy and adaptability of surgical instrument supply, ensuring the timely availability of instruments during surgery, and thus guaranteeing the smooth progress of the surgery.

[0092] 2. By utilizing historical medical device demand atlases and medical device demand change prediction models, it is possible to dynamically predict trends in medical device demand and quickly identify discrepancies in demand through comparative analysis with historical global medical device demand change information. This data-driven prediction and discrepancy analysis mechanism overcomes the problem of lagging medical device demand adjustments in existing technologies, enabling dynamic optimization and adjustment of medical device demand, improving the operating room's ability to respond to emergencies, and further enhancing the flexibility and reliability of surgical instrument management.

[0093] 3. By displaying a list of instruments to be prepared in real time on IoT terminals, and combining this with a demand deviation verification mechanism, discrepancies between instrument supply and actual surgical needs can be promptly identified and corrected. Furthermore, by generating a future instrument demand sequence based on real-time changes in instrument demand and optimizing it, advance scheduling and precise preparation of instruments are achieved. This effectively solves the problem of instrument supply being out of sync with surgical progress caused by the lack of real-time feedback and optimization mechanisms in existing technologies, significantly improving the intelligence level of surgical instrument management and reducing surgical risks caused by instrument supply issues.

[0094] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for intelligent identification of medical devices based on the Internet of Things, characterized in that, include: Acquire a set of real-time surgical progress information for the target surgical area. The real-time surgical progress information in the set includes: surgical stage identifier, patient physiological parameters, and sequence of instruments used. Based on the real-time surgical progress information set, a corresponding set of instrument requirements for the current stage is generated through a pre-trained instrument requirement recognition model. The current set of medical device requirements is optimized to obtain an optimized set of medical device requirements. The optimized set of medical device requirements is added to a pre-stored historical set of medical device requirements to obtain an updated set of medical device requirements. Obtain historical global equipment demand change information corresponding to the historical equipment demand atlas; Obtain and optimize at least one historical device demand set whose time interval is less than a preset duration from the historical device demand map; Input at least one historical equipment demand set and an optimized equipment demand set into a pre-trained equipment demand change prediction model to obtain equipment demand change information; Determine the difference in device demand between historical global device demand change information and device demand change information; Based on the differences in medical device demand and the updated medical device demand atlas, real-time medical device demand change information corresponding to historical global medical device demand change information is generated, and a list of medical devices to be prepared corresponding to the real-time medical device demand change information is displayed on the associated IoT terminals.

2. The method according to claim 1, characterized in that, Based on the medical device demand difference information and the updated medical device demand atlas, real-time medical device demand change information corresponding to historical global medical device demand change information is generated, including: Identify the device type difference groups corresponding to the device demand difference information; Based on the device type difference group, each updated device demand set in the updated device demand map set is segmented by device type to generate device type difference map groups, thus obtaining a set of device type difference map groups. The features of each device type difference map in the device type difference map set are combined to generate a device feature difference map set. For each device feature difference atlas in the device feature difference atlas group, perform the following processing steps: Each device feature difference map in the device feature difference map set is input into a pre-trained feature core recognition model to generate feature core identifiers, thus obtaining a feature core identifier set. Generate medical device demand association information among various feature core identifiers in the feature core identifier set, as real-time medical device demand sub-information; Based on the real-time equipment demand sub-information, the historical global equipment demand change information is updated to obtain the real-time equipment demand change information.

3. The method according to claim 2, characterized in that, The method also includes: Based on real-time changes in instrument demand, verify the match between the list of instruments to be prepared and the surgical procedure: Obtain the sequence of instruments actually used in the current surgical phase; Calculate the demand deviation between the list of instruments to be prepared and the actual sequence of instruments to be used; When the demand deviation exceeds a preset threshold, a demand correction instruction is triggered. Based on the demand correction instructions, the parameters of the medical device demand identification model are adjusted in real time. The optimized set of medical device requirements is regenerated based on the adjusted medical device requirement identification model.

4. The method according to claim 3, characterized in that, The method also includes: In response to the fact that the demand deviation value does not exceed the preset threshold, a future medical device demand sequence for the preset surgical stage is generated based on real-time medical device demand change information and historical medical device demand sequence. Obtain the set of key surgical events within a pre-defined surgical phase; Generate the instrument requirement parameters and time requirement parameters for each critical surgical event in the critical surgical event set, thus obtaining the instrument requirement parameter set and the time requirement parameter set; Based on the set of medical device demand parameters and the set of demand time parameters, the demand for each future medical device in the future medical device demand sequence is optimized and adjusted to obtain the optimized medical device demand sequence. Based on the optimized instrument demand sequence, instrument scheduling and preparation prompts are provided for the target surgical area.

5. The method according to claim 3, characterized in that, The method also includes: Based on the demand correction instructions, the optimized medical device demand set is updated to generate an updated medical device demand set; Generate and update real-time medical device demand change information corresponding to the updated medical device demand set, which serves as the target real-time medical device demand change information; Based on the real-time changes in target medical device demand and the historical medical device demand sequence, a second future medical device demand sequence corresponding to the preset surgical stage is generated. Retrieve the set of emergency surgical events within the preset surgical phase; Generate emergency medical device demand parameters and emergency demand time parameters for each emergency surgical event in the emergency surgical event set, thus obtaining the emergency demand parameter set and the emergency time parameter set; Based on the emergency demand parameter set and the emergency time parameter set, each of the second future medical device demands in the second future medical device demand sequence is urgently optimized and adjusted to obtain the second optimized medical device demand sequence. Based on the second optimized instrument demand sequence, emergency instrument dispatch is carried out for the target surgical area.

6. The method according to claim 1, characterized in that, The real-time surgical progress information set includes: Surgical action characteristic data are collected through an IoT sensor array in the operating room; The surgical stage identifier is calculated based on the matching degree between the surgical action feature data and the preset standard action library: ； in, K represents the surgical stage identifier, and K is the set of surgical stage types. For the first The weighting coefficients of each sensor and satisfying , For real-time sensor data vectors, This is the feature vector of the standard action.

7. The method according to claim 1, characterized in that, The training of the equipment demand identification model includes: A three-dimensional convolutional neural network model is constructed, with the input layer receiving the fused features of the surgical video stream and the instrument usage sequence; Extracting key surgical action features using a spatiotemporal attention mechanism: ； in, For attention weight vectors, It is the sigmoid activation function. For video frame feature vectors, Use feature vectors for historical instruments. and These are trainable parameters.

8. The method according to claim 7, characterized in that, The generation of fusion features includes: The surgical operation video stream was acquired in real time by a multi-view high-definition camera array, and the characteristics of the surgeon's hand movement trajectory were extracted. Read the activation time series of surgical instrument RFID tags to generate an instrument usage frequency distribution map; Input the hand movement trajectory features and the device usage frequency distribution map into the spatiotemporal alignment module, and perform the following alignment operations: When the closing action of the needle holder is detected, the activation status of the suture needle RFID is simultaneously verified. When the electrosurgical unit is detected to be in contact with tissue, verify whether the power parameters of the high-frequency electrosurgical unit are within the preset safety threshold. If the deviation between the motion trajectory characteristics and the device activation time exceeds 200ms, the feature calibration signal is triggered to realign the timestamp. Output spatiotemporally synchronized fused feature vectors to the device demand identification model.

9. The method according to claim 3, characterized in that, The following are instructions that trigger requirement corrections: Monitor key events during the surgical procedure; Dynamically adjust preset thresholds based on the type of key event nodes; When a high-precision operation phase is detected, the preset threshold is lowered to enhance verification sensitivity. When abnormal patient vital signs are detected, a secondary manual verification mechanism is activated. Activate the emergency response channel in scenarios where medical devices are used frequently.

10. The method according to claim 4, characterized in that, Equipment scheduling and preparation tips include: Establish a multi-level response mechanism and classify different response levels according to the demand time parameter; Configure differentiated scheduling strategies for different response levels: For routine response levels, generate equipment preparation instructions; The sound and light alert device is activated based on the emergency response level; Automated delivery equipment is controlled by an instant response level. Implement dynamic conflict resolution strategies: The priority of instrument scheduling is allocated according to the criticality of the surgical stage; Spatial path conflicts can be avoided by using a real-time positioning system.

Images