Medical consumable integrated management system based on deep learning

Abstract

The present application relates to the technical field of medical consumable management, and particularly relates to a medical consumable integrated management system based on deep learning, which comprises a perception sensing module, collects consumable data through array type gravity sensing and self-adaptive radio frequency identification; an edge computing module uses a conflict resolution subunit to perform credibility evaluation and fusion on heterogeneous sensing data, and realizes consumable visual recognition, demand prediction and anomaly detection through a deep learning subunit; a business application module realizes consumable intelligent scheduling and control based on operation scheduling analysis, dynamic expiration date early warning and closed-loop permission control; and a data storage module supports full life cycle data management, and realizes data storage and process automatic settlement in combination with block chain technology. The present application improves the accuracy of consumable management, the efficiency of inventory turnover and the safety of clinical service.

Description

Technical Field

[0001] This invention relates to the field of medical consumables management technology, specifically to an integrated medical consumables management system based on deep learning. Background Technology

[0002] With the rapid development of medical technology, the usage of high-value medical consumables in clinical surgery has been increasing year by year. Their management efficiency is directly related to medical quality and patient safety. Currently, hospital consumable management generally faces problems such as low inventory accuracy, lagging expiration date monitoring, and difficulty in retrieval and traceability. Especially in critical scenarios such as operating rooms, real-time inventory, expiration date warning, and access control of consumables have become core pain points for the refined operation of hospitals. The traditional manual management model is difficult to meet the needs of modern smart hospitals for digital and intelligent management of the entire life cycle of consumables.

[0003] Existing medical consumables management systems primarily rely on single sensor technologies or manual barcode scanning for inventory recording. For example, static identification schemes based on barcodes / QR codes require manual scanning of each item, which is inefficient and prone to omissions. While pure RFID solutions can achieve batch reading, they are susceptible to electromagnetic interference in environments with dense metal consumable storage, leading to missed or misreadings. Gravity sensor-based solutions can monitor weight changes but cannot identify specific consumable categories, and temperature drift errors affect measurement accuracy. Furthermore, existing systems lack the ability to fuse and process multi-source heterogeneous data, and cannot make autonomous decisions when sensor data conflict. Expiry date management often uses fixed threshold warnings without dynamic adjustments based on surgical scheduling, resulting in the dilemma of high-value consumables expiring and being wasted, or having no consumables available for emergency surgeries. Access control is disconnected from surgical information, making it difficult to achieve refined authorization based on actual clinical needs.

[0004] In summary, existing technologies have significant shortcomings in multimodal sensing reliability, edge intelligent decision-making, and clinical business collaboration. There is an urgent need for an integrated medical consumables management system that can integrate gravity sensing and radio frequency identification technologies, has environmental adaptability, and deeply integrates with surgical business processes, in order to achieve accurate perception, intelligent prediction, and closed-loop control of consumable status, thereby improving the safety and operational efficiency of hospital consumables management. Summary of the Invention

[0005] In view of the above-mentioned shortcomings of the existing technology, the present invention provides an integrated medical consumables management system based on deep learning, which can effectively solve the problems mentioned in the existing technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides an integrated medical consumables management system based on deep learning, comprising:

[0008] The sensing module is used to collect physical state data of consumables and generate heterogeneous sensing data streams;

[0009] The sensing module includes an array-type gravity sensing unit and an adaptive radio frequency identification unit. The array-type gravity sensing unit is used to collect real-time weight change data of each storage location, and the adaptive radio frequency identification unit is used to read the electronic tag information of consumables.

[0010] An edge computing module is used for fusing and intelligent reasoning analysis of the heterogeneous sensor data streams;

[0011] The edge computing module includes a heterogeneous data fusion unit and a deep learning inference unit. The heterogeneous data fusion unit is used to perform spatiotemporal alignment of the gravity sensing data and the radio frequency identification data. When the theoretical deviation between the two based on the nominal weight of a single item exceeds a threshold, it resolves the conflict based on environmental interference parameters and adaptively selects the main data source. The deep learning inference unit is used to perform consumable status identification and predictive analysis based on the fused data.

[0012] The business application module is used to implement consumable business logic and clinical process control based on the intelligent reasoning analysis results;

[0013] The business application module includes a surgical scheduling analysis unit, a dynamic expiration date warning unit, and a closed-loop access control unit. The surgical scheduling analysis unit is used to connect to the hospital information system to obtain surgical arrangement data. The dynamic expiration date warning unit is used to calculate the urgency of the expiration date based on surgical scheduling data and consumable consumption prediction. The closed-loop access control unit is used to build an authorization verification mechanism based on surgical information.

[0014] The data storage module is used to store consumable lifecycle data, deep learning model parameters, and system operation logs;

[0015] The heterogeneous data fusion unit is configured with a conflict resolution subunit, which is used to perform the following steps:

[0016] Acquire the actual weight change detected by the array-type gravity sensing unit. The change in the number of tags read by the adaptive RFID unit The formula for calculating the theoretical weight deviation is:

[0017] ;

[0018] Where w0 is the nominal weight of a single item of the corresponding consumable category;

[0019] Obtain the metal interference intensity index collected by the metal detection sensors pre-installed in each storage location. The signal-to-noise ratio of the radio frequency signal output by the adaptive radio frequency identification unit The temperature drift compensation coefficient of the gravity sensing unit is obtained by querying a pre-stored temperature drift compensation coefficient table based on the temperature sensor monitoring data. ;

[0020] Calculate the gravity sensing confidence score by calling the pre-stored confidence evaluation function. and RFID credibility score ;

[0021] Compare theoretical weight deviation values With preset dynamic threshold ,when At that time, compare the gravity sensing reliability scores. and RFID credibility score Select the sensor mode with the higher score as the main data source;

[0022] The sensor parameter adjustment command is generated and fed back to the sensing module. The parameter adjustment command includes the radio frequency antenna power enhancement command, the antenna polarization direction adjustment command, and the gravity sensing unit sampling frequency increase command.

[0023] Furthermore, the conflict resolution subunit is also used for:

[0024] Historical conflict data is input into a deep reinforcement learning network for training to obtain the dynamic threshold. The historical conflict data includes inventory count error records, sensor response delay records, surgical urgency markers, consumable usage frequency statistics, cabinet door opening and closing event records, and environmental temperature and humidity monitoring values.

[0025] Furthermore, the adaptive radio frequency identification unit includes:

[0026] Multi-band antenna sub-units are configured inside each storage location and support multi-frequency switching of high-frequency bands;

[0027] Metal environment adaptive subunit, used to adapt to the metal interference intensity index The pre-stored antenna parameter configuration table is called to dynamically switch the antenna operating frequency, adjust the transmit power, and change the beam direction. The pre-stored antenna parameter configuration table records the antenna parameter combinations corresponding to different metal interference intensity index ranges.

[0028] A near-field focusing antenna subunit is used for precise close-range identification of high-value metal consumables. The near-field focusing antenna subunit adopts a loop antenna structure loaded with magnetic medium. The number of turns, wire diameter, and type of magnetic medium material of the loop antenna are determined by finite element simulation based on the electromagnetic coupling characteristics of the target metal consumable.

[0029] Furthermore, the deep learning inference unit includes:

[0030] The consumables visual recognition subunit is used to activate the cabinet camera to collect images of consumables when the conflict resolution subunit determines that both modes of data are unreliable. It then calls a pre-trained convolutional neural network model to extract image features, compares them with a pre-stored consumables category feature library, and outputs the consumables identity confirmation result.

[0031] The time series prediction subunit is used to call a pre-trained long short-term memory network model, input historical consumable consumption time series, surgical scheduling plan data and seasonal epidemic marker data, and output the predicted value of consumable demand at a future preset time scale.

[0032] The anomaly detection subunit is used to call the pre-trained autoencoder model, input the current sensing module data, reconstruct the normal data pattern, calculate the reconstruction error, and output an anomaly alarm when the reconstruction error exceeds the preset anomaly threshold.

[0033] Furthermore, the dynamic expiration date early warning unit is used to perform the following steps:

[0034] Obtain the remaining shelf life of consumables in days Calculate the shelf life ratio factor based on the total shelf life days of consumables. ;

[0035] The future output of the timing prediction sub-unit The daily demand forecast for consumables is used as the predicted consumption. Get the current inventory level Calculate the inventory matching factor ;

[0036] The system retrieves surgical scheduling data for the next 72 hours from the surgical scheduling analysis unit, extracts surgical type codes, queries the hospital information system to obtain the emergency level score corresponding to each surgical type, and calculates the surgical emergency adjustment factor. ;

[0037] According to the pre-configured weighting coefficient , , The formula for calculating the expiration date urgency index is as follows:

[0038] ;

[0039] in, It is an expiration date urgency index, and ;

[0040] Determine the range to which the EPI value belongs, when A red emergency alert is triggered and consumables are locked for use in non-emergency surgeries. When a yellow alert is triggered and the weight of the surgical scheduling recommendation is increased, The time marker is green, indicating a normal state.

[0041] Furthermore, the process of querying the hospital information system to obtain the emergency level score corresponding to each surgical type includes:

[0042] Extract surgical type codes from surgical scheduling data;

[0043] The hospital information system is queried to obtain the risk level classification results corresponding to each surgical type code. The risk level classification results are determined in accordance with the surgical classification management system of the National Health Commission.

[0044] Level 1 surgeries are mapped to the first emergency level score, Level 2 surgeries to the second emergency level score, Level 3 surgeries to the third emergency level score, and Level 4 surgeries to the fourth emergency level score.

[0045] The frequency of surgeries at each risk level within the future surgical cycle is statistically analyzed, and a weighted average is calculated as a factor to adjust for surgical urgency. .

[0046] Furthermore, the data storage module includes a first storage sub-unit for storing static information and dynamic records of consumables, a second storage sub-unit for storing deep learning model training parameters and version information, and a third storage sub-unit for storing various event logs of the system in chronological order.

[0047] The first storage subunit records the receipt, issuance, inventory, expiration date, and surgical-related information of consumables;

[0048] The second storage subunit stores the weights and configuration information of the neural network model used in the deep learning inference unit;

[0049] The third storage subunit records events collected by the sensing module, the decision-making process of the conflict resolution subunit, early warning trigger records, and user operation logs.

[0050] Furthermore, the data storage module is also used for:

[0051] The key node data of consumables is encrypted and written into the hospital's internal blockchain node. The key node data includes manufacturer information, sterilization batch information, warehousing and acceptance information, usage record information, patient implantation information and postoperative follow-up information. The hospital's internal blockchain node is deployed using the Hyperledger Fabric framework.

[0052] The system monitors the access confirmation events output by the closed-loop access control unit. When an access confirmation event is detected, it calls the medical insurance settlement interface of the hospital's financial system and the payment interface of the supplier's ERP system to trigger the automatic settlement process.

[0053] Furthermore, the surgical scheduling analysis unit is also used for:

[0054] Historical surgical records are extracted from the surgical anesthesia system database of the hospital information system. These historical surgical records include surgical codes, surgical names, and a list of UDIs for consumables used.

[0055] Statistical analysis was performed by grouping surgical codes, and historical surgical records in which the frequency of UDI (Usage Diary Items) for consumables exceeded a preset threshold were selected.

[0056] The filtered surgical codes, surgical names, and corresponding high-frequency consumable UDI lists are stored as a surgical consumable association index for query and retrieval by the dynamic expiration warning unit.

[0057] The technical solution provided by this invention has the following advantages compared with the known prior art:

[0058] This invention achieves reliability arbitration and dynamic compensation of dual-mode data from gravity sensing and radio frequency identification through the conflict resolution subunit of the heterogeneous data fusion unit. It solves the problem of missed reading and misreading caused by RFID signal shielding in the environment of metal consumables, significantly improves the accuracy and real-time performance of inventory of high-value consumables, and reduces the risk of inventory management errors caused by sensor data conflicts.

[0059] This invention deeply integrates consumable expiration management with the hospital's existing surgical grading management system through the linkage of a surgical scheduling analysis unit and a dynamic expiration warning unit. Based on the surgical urgency adjustment factor, it dynamically adjusts the priority of inventory consumption, effectively reducing the loss of high-value consumables due to expiration and optimizing the efficiency of inventory capital turnover. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0061] Figure 1 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0063] The present invention will be further described below with reference to embodiments.

[0064] Example:

[0065] Reference Figure 1 This invention provides an integrated management system for medical consumables based on deep learning. The system includes a sensing module, an edge computing module, a business application module, and a data storage module.

[0066] The sensing module is used to collect physical state data of consumables and generate heterogeneous sensing data streams. Specifically, it includes an array-type gravity sensing unit and an adaptive RFID unit. The array-type gravity sensing unit is deployed in each storage location of the intelligent consumable cabinet to collect real-time weight change data for each location. The adaptive RFID unit is used to read information from the electronic tags attached to the consumables. This adaptive RFID unit further includes a multi-band antenna subunit, a metal environment adaptive subunit, and a near-field focusing antenna subunit. The multi-band antenna subunit supports multi-frequency switching of high-frequency bands to adapt to different environments. The metal environment adaptive subunit can dynamically query a pre-stored antenna parameter configuration table based on the real-time monitored metal interference intensity index Im, and accordingly switch the antenna operating frequency, adjust the transmission power, and change the beam direction to suppress interference. The near-field focusing antenna subunit is specifically designed for high-value metal consumables, employing a magnetically loaded loop antenna structure with parameters determined through finite element simulation to achieve accurate identification at close range.

[0067] The edge computing module is used for fusing and intelligent inference analysis of heterogeneous sensor data streams. It includes a heterogeneous data fusion unit and a deep learning inference unit. The heterogeneous data fusion unit is used for spatiotemporal alignment and fusion of gravity sensor data and RFID data. Its core is a configured conflict resolution subunit. This conflict resolution subunit performs the following steps:

[0068] First, obtain the actual weight change detected by the array-type gravity sensing unit. The change in the number of tags read by the adaptive RFID unit The formula for calculating the theoretical weight deviation is:

[0069] ;

[0070] Where w0 is the nominal weight of a single item of the corresponding consumable category;

[0071] Obtain the metal interference intensity index collected by the metal detection sensors pre-installed in each storage location. The signal-to-noise ratio of the radio frequency signal output by the adaptive radio frequency identification unit The temperature drift compensation coefficient of the gravity sensing unit is obtained by querying a pre-stored temperature drift compensation coefficient table based on the temperature sensor monitoring data. ;

[0072] Calculate the gravity sensing confidence score by calling the pre-stored confidence evaluation function. and RFID credibility score ;

[0073] Compare theoretical weight deviation values With preset dynamic threshold ,when At that time, compare the gravity sensing reliability scores. and RFID credibility score Select the sensor mode with the higher score as the main data source;

[0074] Finally, sensor parameter tuning commands (such as increasing RF antenna power, adjusting antenna polarization direction, and increasing gravity sensor sampling frequency) are generated and fed back to the sensing module to achieve adaptive optimization.

[0075] Furthermore, dynamic thresholds can be obtained by training a deep reinforcement learning network on historical conflict data (such as inventory counting errors, sensor delays, surgical urgency, frequency of consumable retrieval, cabinet door incidents, and environmental temperature and humidity). This enables the system to have adaptive adjustment capabilities.

[0076] The deep learning inference unit performs consumable status identification and predictive analysis based on fused reliable data. It comprises three sub-units. The consumable visual recognition sub-unit is activated when both dual-mode data are deemed untrustworthy. It acquires images by calling the camera inside the cabinet and uses a pre-trained convolutional neural network model for feature extraction and comparison, outputting the consumable identification result. The time series prediction sub-unit calls a pre-trained long short-term memory network model, inputting historical consumption sequences, surgical schedules, and epidemiological markers, and outputs a predicted value for future consumable demand. The anomaly detection sub-unit calls a pre-trained autoencoder model, calculates the reconstruction error of real-time sensor data to determine abnormal states, and triggers an alarm when the error exceeds a threshold.

[0077] The business application module uses intelligent reasoning and analysis results to implement consumable business logic and clinical process control. It includes a surgical scheduling analysis unit, a dynamic expiration date early warning unit, and a closed-loop access control unit. The surgical scheduling analysis unit connects to the hospital information system to obtain surgical arrangement data and can analyze historical surgical records to establish an association index between high-frequency consumables and surgical types for other units to access.

[0078] The dynamic expiration date early warning unit is used to calculate the urgency of the expiration date, and its execution steps are as follows:

[0079] Obtain the remaining shelf life days Tr and the total shelf life days of the consumables, and calculate the shelf life ratio factor TrTt;

[0080] The predicted consumption amount Cp is obtained by calling the consumable demand forecast value of the future Tr days output by the time series forecasting subunit, obtaining the current inventory amount Cc, and calculating the inventory matching factor CpCc.

[0081] The surgical scheduling data for the next 72 hours output by the surgical scheduling parsing unit is called, the surgical type code is extracted, the emergency level score corresponding to each surgical type is obtained by querying the hospital information system, and the surgical emergency level adjustment factor Rs is calculated.

[0082] The urgency index is calculated using pre-configured weighting coefficients α, β, and γ, using the following formula:

[0083] EPI=α×(TrTt)+β×(CpCc)+γ×Rs;

[0084] Among them, EPI is the expiration date urgency index, and α+β+γ=1;

[0085] Determine the range of EPI value. When EPI ≥ 0.8, trigger a red emergency warning and lock consumables for non-emergency surgery. When 0.5 ≤ EPI < 0.8, trigger a yellow attention warning and increase the recommended weight of surgery scheduling. When EPI < 0.5, mark it as green normal status.

[0086] The closed-loop access control unit establishes a dynamic authorization and verification mechanism based on surgical information to ensure that the use of consumables is linked to surgical tasks, thereby achieving closed-loop management.

[0087] The data storage module stores various types of system data, including a first storage sub-unit for storing static and dynamic information throughout the entire lifecycle of consumables, a second storage sub-unit for storing deep learning model parameters and version information, and a third storage sub-unit for storing system operation event logs in sequence. Furthermore, this module is responsible for encrypting key consumable node data (production, sterilization, warehousing, retrieval, implantation, etc.) and writing it to the hospital's internal blockchain node deployed based on the Hyperledger Fabric framework, ensuring data immutability and traceability. Simultaneously, it monitors consumable retrieval confirmation events and automatically calls the financial and supplier system interfaces to trigger the settlement process, achieving business-finance linkage.

[0088] Furthermore, the conflict resolution subunit is also used for:

[0089] Historical conflict data is input into a deep reinforcement learning network for training to obtain dynamic thresholds. Historical conflict data includes inventory count error records, sensor response delay records, surgical urgency markers, consumable usage frequency statistics, cabinet door opening and closing event records, and environmental temperature and humidity monitoring values.

[0090] Furthermore, the adaptive radio frequency identification unit includes:

[0091] Multi-band antenna sub-units are configured inside each storage location and support multi-frequency switching of high-frequency bands;

[0092] Metal environment adaptive subunit, used to adapt to the metal interference intensity index Call the pre-stored antenna parameter configuration table to dynamically switch the antenna operating frequency, adjust the transmit power and change the beam direction. The pre-stored antenna parameter configuration table records the antenna parameter combinations corresponding to different metal interference intensity index ranges.

[0093] The near-field focusing antenna subunit is used for precise close-range identification of high-value metal consumables. The near-field focusing antenna subunit adopts a loop antenna structure loaded with magnetic medium. The number of turns, wire diameter and type of magnetic medium material of the loop antenna are determined by finite element simulation based on the electromagnetic coupling characteristics of the target metal consumable.

[0094] Furthermore, the deep learning inference unit includes:

[0095] The consumables visual recognition subunit is used to activate the cabinet camera to collect images of consumables when the conflict resolution subunit determines that both modes of data are unreliable. It then calls a pre-trained convolutional neural network model to extract image features, compares them with a pre-stored consumables category feature library, and outputs the consumables identity confirmation result.

[0096] The time series prediction subunit is used to call a pre-trained long short-term memory network model, input historical consumable consumption time series, surgical scheduling plan data and seasonal epidemic marker data, and output the predicted value of consumable demand at a future preset time scale.

[0097] The anomaly detection subunit is used to call the pre-trained autoencoder model, input the current sensing module data, reconstruct the normal data pattern, calculate the reconstruction error, and output an anomaly alarm when the reconstruction error exceeds the preset anomaly threshold.

[0098] Furthermore, querying the hospital information system to obtain the emergency level score corresponding to each surgical type includes:

[0099] Extract surgical type codes from surgical scheduling data;

[0100] Query the hospital information system to obtain the risk level classification results corresponding to each surgical type code. The risk level classification results are determined in accordance with the surgical classification management system of the National Health Commission.

[0101] Level 1 surgeries are mapped to the first emergency level score, Level 2 surgeries to the second emergency level score, Level 3 surgeries to the third emergency level score, and Level 4 surgeries to the fourth emergency level score.

[0102] The frequency of surgeries at each risk level within the future surgical cycle is statistically analyzed, and a weighted average is calculated as a factor to adjust for surgical urgency. .

[0103] Furthermore, the data storage module includes a first storage sub-unit for storing static information and dynamic records of consumables, a second storage sub-unit for storing deep learning model training parameters and version information, and a third storage sub-unit for storing various event logs of the system in chronological order.

[0104] The first storage sub-unit records the receipt, issuance, inventory, expiration date, and surgical-related information of consumables;

[0105] The second storage subunit stores the weights and configuration information of the neural network model used in the deep learning inference unit;

[0106] The third storage subunit records events collected by the sensing module, the decision-making process of the conflict resolution subunit, early warning trigger records, and user operation logs.

[0107] Furthermore, the data storage module is also used for:

[0108] The key node data of consumables is encrypted and written into the hospital's internal blockchain node. The key node data includes manufacturer information, sterilization batch information, warehousing and acceptance information, usage record information, patient implantation information and postoperative follow-up information. The hospital's internal blockchain node is deployed using the Hyperledger Fabric framework.

[0109] The system monitors the access confirmation events output by the closed-loop access control unit. When an access confirmation event is detected, it calls the medical insurance settlement interface of the hospital's financial system and the payment interface of the supplier's ERP system to trigger the automatic settlement process.

[0110] Furthermore, the surgical scheduling analysis unit is also used for:

[0111] Historical surgical records were extracted from the surgical anesthesia system database of the hospital information system. The historical surgical records include surgical codes, surgical names, and a list of UDIs for consumables used.

[0112] Statistical analysis was performed by grouping surgical codes, and historical surgical records in which the frequency of UDI (Usage Diary Items) for consumables exceeded a preset threshold were selected.

[0113] The filtered surgical codes, surgical names, and corresponding high-frequency consumable UDI lists are stored as a surgical consumable association index for query and retrieval by the dynamic expiration warning unit.

[0114] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A deep learning-based integrated management system for medical consumables, characterized in that, include: The sensing module is used to collect physical state data of consumables and generate heterogeneous sensing data streams; The sensing module includes an array-type gravity sensing unit and an adaptive radio frequency identification unit. The array-type gravity sensing unit is used to collect real-time weight change data of each storage location, and the adaptive radio frequency identification unit is used to read the electronic tag information of consumables. An edge computing module is used for fusing and intelligent reasoning analysis of the heterogeneous sensor data streams; The edge computing module includes a heterogeneous data fusion unit and a deep learning inference unit; The heterogeneous data fusion unit is used to perform spatiotemporal alignment of the gravity sensing data and the radio frequency identification data, and when the theoretical deviation between the two based on the nominal weight of a single item exceeds a threshold, it resolves the conflict based on environmental interference parameters and adaptively selects the main data source; the deep learning inference unit is used to perform consumable status identification and predictive analysis based on the fused data. The business application module is used to implement consumable business logic and clinical process control based on the intelligent reasoning analysis results; The business application module includes a surgical scheduling analysis unit, a dynamic efficacy period early warning unit, and a closed-loop permission control unit. The surgical scheduling parsing unit is used to connect to the hospital information system to obtain surgical arrangement data; the dynamic expiration date early warning unit is used to calculate the urgency of expiration date based on surgical scheduling data and consumable consumption prediction; the closed-loop access control unit is used to build an authorization verification mechanism based on surgical information. The data storage module is used to store consumable lifecycle data, deep learning model parameters, and system operation logs; The heterogeneous data fusion unit is configured with a conflict resolution subunit, which is used to perform the following steps: Acquire the actual weight change detected by the array-type gravity sensing unit. The change in the number of tags read by the adaptive RFID unit The formula for calculating the theoretical weight deviation is: ； Where w0 is the nominal weight of a single item of the corresponding consumable category; Obtain the metal interference intensity index collected by the metal detection sensors pre-installed in each storage location. The signal-to-noise ratio of the radio frequency signal output by the adaptive radio frequency identification unit The temperature drift compensation coefficient of the gravity sensing unit is obtained by querying a pre-stored temperature drift compensation coefficient table based on the temperature sensor monitoring data. ; Calculate the gravity sensing confidence score by calling the pre-stored confidence evaluation function. and RFID credibility score ; Compare theoretical weight deviation values With preset dynamic threshold ,when At that time, compare the gravity sensing reliability scores. and RFID credibility score Select the sensor mode with the higher score as the main data source; The sensor parameter adjustment command is generated and fed back to the sensing module. The parameter adjustment command includes the radio frequency antenna power enhancement command, the antenna polarization direction adjustment command, and the gravity sensing unit sampling frequency increase command.

2. The integrated medical consumables management system based on deep learning according to claim 1, characterized in that, The conflict resolution subunit is also used for: Historical conflict data is input into a deep reinforcement learning network for training to obtain the dynamic threshold. The historical conflict data includes inventory count error records, sensor response delay records, surgical urgency markers, consumable usage frequency statistics, cabinet door opening and closing event records, and environmental temperature and humidity monitoring values.

3. The integrated medical consumables management system based on deep learning according to claim 1, characterized in that, The adaptive radio frequency identification unit includes: Multi-band antenna sub-units are configured inside each storage location and support multi-frequency switching of high-frequency bands; Metal environment adaptive subunit, used to adapt to the metal interference intensity index The pre-stored antenna parameter configuration table is called to dynamically switch the antenna operating frequency, adjust the transmit power, and change the beam direction. The pre-stored antenna parameter configuration table records the antenna parameter combinations corresponding to different metal interference intensity index ranges. A near-field focusing antenna subunit is used for precise close-range identification of high-value metal consumables. The near-field focusing antenna subunit adopts a loop antenna structure loaded with magnetic medium. The number of turns, wire diameter, and type of magnetic medium material of the loop antenna are determined by finite element simulation based on the electromagnetic coupling characteristics of the target metal consumable.

4. The integrated medical consumables management system based on deep learning according to claim 1, characterized in that, The deep learning inference unit includes: The consumables visual recognition subunit is used to activate the cabinet camera to collect images of consumables when the conflict resolution subunit determines that both modes of data are unreliable. It then calls a pre-trained convolutional neural network model to extract image features, compares them with a pre-stored consumables category feature library, and outputs the consumables identity confirmation result. The time series prediction subunit is used to call a pre-trained long short-term memory network model, input historical consumable consumption time series, surgical scheduling plan data and seasonal epidemic marker data, and output the predicted value of consumable demand at a future preset time scale. The anomaly detection subunit is used to call the pre-trained autoencoder model, input the current sensing module data, reconstruct the normal data pattern, calculate the reconstruction error, and output an anomaly alarm when the reconstruction error exceeds the preset anomaly threshold.

5. The integrated medical consumables management system based on deep learning according to claim 1, characterized in that, The dynamic validity period early warning unit is used to perform the following steps: Obtain the remaining shelf life of consumables in days Calculate the shelf life ratio factor based on the total shelf life days of consumables. ; The future output of the timing prediction sub-unit The daily demand forecast for consumables is used as the predicted consumption. Get the current inventory level Calculate the inventory matching factor ; The system retrieves surgical scheduling data for the next 72 hours from the surgical scheduling analysis unit, extracts surgical type codes, queries the hospital information system to obtain the emergency level score corresponding to each surgical type, and calculates the surgical emergency adjustment factor. ; According to the pre-configured weighting coefficient , , The formula for calculating the expiration date urgency index is as follows: ； in, It is an expiration date urgency index, and ; Determine the range to which the EPI value belongs, when A red emergency alert is triggered and consumables are locked for use in non-emergency surgeries. When a yellow alert is triggered and the weight of the surgical scheduling recommendation is increased, The time marker is green, indicating a normal state.

6. The integrated medical consumables management system based on deep learning according to claim 5, characterized in that, The emergency level score obtained from the hospital information system for each type of surgery includes: Extract surgical type codes from surgical scheduling data; The hospital information system is queried to obtain the risk level classification results corresponding to each surgical type code. The risk level classification results are determined in accordance with the surgical classification management system of the National Health Commission. Level 1 surgeries are mapped to the first emergency level score, Level 2 surgeries to the second emergency level score, Level 3 surgeries to the third emergency level score, and Level 4 surgeries to the fourth emergency level score. The frequency of surgeries at each risk level within the future surgical cycle is statistically analyzed, and a weighted average is calculated as a factor to adjust for surgical urgency. .

7. The integrated medical consumables management system based on deep learning according to claim 1, characterized in that, The data storage module includes a first storage sub-unit for storing static information and dynamic records of consumables, a second storage sub-unit for storing training parameters and version information of deep learning models, and a third storage sub-unit for storing various event logs of the system in chronological order. The first storage subunit records the receipt, issuance, inventory, expiration date, and surgical-related information of consumables; The second storage subunit stores the weights and configuration information of the neural network model used in the deep learning inference unit; The third storage subunit records events collected by the sensing module, the decision-making process of the conflict resolution subunit, early warning trigger records, and user operation logs.

8. The integrated medical consumables management system based on deep learning according to claim 1, characterized in that, The data storage module is also used for: The key node data of consumables is encrypted and written into the hospital's internal blockchain node. The key node data includes manufacturer information, sterilization batch information, warehousing and acceptance information, usage record information, patient implantation information and postoperative follow-up information. The hospital's internal blockchain node is deployed using the Hyperledger Fabric framework. The system monitors the access confirmation events output by the closed-loop access control unit. When an access confirmation event is detected, it calls the medical insurance settlement interface of the hospital's financial system and the payment interface of the supplier's ERP system to trigger the automatic settlement process.

9. The integrated medical consumables management system based on deep learning according to claim 1, characterized in that, The surgical scheduling analysis unit is also used for: Historical surgical records are extracted from the surgical anesthesia system database of the hospital information system. These historical surgical records include surgical codes, surgical names, and a list of UDIs for consumables used. Statistical analysis was performed by grouping surgical codes, and historical surgical records in which the frequency of UDI (Usage Diary Items) for consumables exceeded a preset threshold were selected. The filtered surgical codes, surgical names, and corresponding high-frequency consumable UDI lists are stored as a surgical consumable association index for query and retrieval by the dynamic expiration warning unit.

Images