A method and system for whole-process traceability management of toxic drugs

Abstract

The application discloses a kind of narcotic drugs whole-process traceability management method and system, specifically related to medical information technology field, the method includes: obtaining the multimodal perception data generated in the circulation process of narcotic drugs, at least including operator identity data, drug object identity data, drug identification data, weight change data and operation image data;Multimodal perception data is input into the dynamic risk score model of edge computing node, the model extracts the behavior fingerprint characteristics of operator based on time series attention mechanism and outputs dynamic risk score;Determine whether dynamic risk score exceeds preset threshold, if exceeds, generate intercept instruction, otherwise generate release instruction and append storage traceability record.The application realizes the intelligent closed-loop traceability management of narcotic drugs whole process by multimodal fusion verification and adaptive risk assessment, effectively improves the real-time nature and data integrity of supervision.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, and more specifically, to a method and system for full-process traceability management of controlled drugs. Background Technology

[0002] Narcotic drugs refer to toxic and anesthetic drugs, which are characterized by strong addictiveness and easy abuse. Their production, circulation, use and disposal are strictly controlled by the state. At present, medical institutions mainly rely on a combination of electronic drug supervision code traceability system and manual ledger to manage narcotic drugs.

[0003] In the prior art, for example, Chinese patent CN117476190A discloses a closed-loop management system and method for the entire process of narcotic and psychotropic drugs, which achieves preliminary traceability of drug flow by establishing an electronic archive to record drug entry and exit information and prescription information.

[0004] For example, Chinese patent CN114580831A discloses a method and device for tracing and monitoring special drugs, which uses blockchain technology to store information on drug circulation nodes to ensure that the data is tamper-proof. In addition, some smart drug cabinets integrate a primary weight sensor to determine drug usage by monitoring changes in the cabinet's weight.

[0005] However, the aforementioned existing technologies still have the following shortcomings: 1. Most solutions are retrospective, meaning that data is recorded after the violation occurs, and abnormal operations cannot be intercepted in real time during the medication collection or administration process. 2. The data perception dimension is singular, relying on the first weight sensor or a single barcode scan, making it difficult to achieve multi-dimensional cross-verification of operator identity, drug identification and operation behavior; 3. Risk assessment uses fixed thresholds, which cannot be dynamically adjusted based on the operator's historical behavior patterns, making it prone to false alarms or missed alarms. 4. Traceability data is limited to a single medical institution, making it difficult to achieve cross-institutional information sharing and integrity verification; 5. The recycling and disposal of empty ampoules still relies on manual verification, which presents a management loophole where residual medicine is not completely destroyed.

[0006] Therefore, how to achieve intelligent and closed-loop traceability and supervision of the entire process of controlled drugs while ensuring the legitimate demand for medication is a technical problem that urgently needs to be solved in this field. In view of this, the present invention provides a method and system for full-process traceability management of controlled drugs. Summary of the Invention

[0007] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a method and system for full-process traceability management of narcotic drugs, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: On the one hand, this invention provides a method for full-process traceability management of controlled substances, including: Acquire multimodal sensing data generated during the circulation of controlled drugs, wherein the multimodal sensing data includes at least operator identification data, drug recipient identification data, drug identification data, weight change data, and operation image data; Multimodal perception data is input into a dynamic risk scoring model deployed on edge computing nodes. The dynamic risk scoring model extracts the behavioral fingerprint features of operators based on a temporal attention mechanism and outputs a dynamic risk score based on the behavioral fingerprint features. Determine whether the dynamic risk score exceeds a preset risk threshold; If the limit is exceeded, an interception command is generated, which is used to prohibit the current operation and report it to the cloud monitoring platform; If the limit is not exceeded, a release order is generated, and traceability records of narcotic drugs are added to the cloud-based monitoring platform based on multimodal perception data.

[0009] Preferably, the dynamic risk scoring model includes: The input layer is used to receive historical behavior sequence data of operators, which includes drug dispensing time sequence, drug dispensing frequency sequence, single drug dispensing dosage sequence, drug use completion rate sequence, and empty ampoule recovery timeliness rate sequence. Long Short-Term Memory (LSTM) network layers are used to extract behavioral fingerprint features from historical behavioral sequence data; The attention layer is used to assign weights to different time steps in the historical behavior sequence data, where the time steps corresponding to abnormal behaviors are assigned higher weights to highlight their contribution to the risk score. The output layer is used to calculate and output a dynamic risk score based on the weighted behavioral fingerprint features.

[0010] Preferably, acquiring multimodal sensing data generated during the circulation of controlled substances specifically includes: During the medication dispensing stage, multimodal sensing data is acquired through the intelligent drug cabinet. The intelligent drug cabinet integrates a facial recognition module, an RFID reader, and a first weight sensor. Specifically, the facial recognition module acquires the operator's identity data as the first identity data, the RFID reader acquires the drug's electronic supervision code as the first drug identification data, and the first weight sensor acquires the weight difference before and after the drug is removed as the first weight change data. The first drug identification data is used to identify the unique code of a single drug, and the first weight change data is used to verify the consistency between the number of drugs removed and the RFID reading result. During the medication administration phase, the operator's electronic signature is obtained through the mobile nursing terminal as the second identity data, and the facial features of the medication recipient are obtained through the first image acquisition module as the third identity data. The medication timestamp is also recorded. The mobile nursing terminal also integrates a drug identification function, which is used to read the electronic supervision code of the drug before medication to verify the drug information, and a micro switch, which is used to detect the drug's presence status. When the drug is taken out, a signal is triggered to record the time node of the retrieval action. During the recycling phase, image data of empty ampoules is acquired through an intelligent recycling terminal as the first operational image data, and an appearance image of the empty ampoules is acquired through a second image acquisition module. The weight of the empty ampoules is then verified by a second weight sensor to determine whether the empty ampoules have been emptied and have no residual liquid.

[0011] Preferably, adding traceability records for controlled substances to the cloud-based monitoring platform includes building a cross-agency traceability chain, specifically including: Based on the institution's internal private chain, a fine-grained operation log is recorded to track the flow of controlled drugs within a single medical institution. The fine-grained operation log includes at least the operation time, operator identification, drug batch number, drug quantity, operation type, and operation result. The fine-grained operation log is stored on a cloud-based monitoring platform. Based on a regional consortium blockchain, summary information on the flow of controlled drugs between different medical institutions is recorded. The summary information is obtained by hashing fine-grained operation logs and stored in a cloud-based monitoring platform. When a cross-agency traceability request is initiated, the cloud-based monitoring platform verifies the integrity of the traceability data by comparing the hash value of the summary information with that of the fine-grained operation log.

[0012] Preferably, before generating the release instruction, the method further includes: The first identity data, first drug identification data, and first weight change data obtained during the drug dispensing stage are subjected to three-source fusion verification to determine whether the first identity data matches the pre-stored operation permission list, whether the first drug identification data matches the outbound record of the intelligent drug cabinet, and whether the first weight change data matches the product of the standard weight of a single drug and the number of drugs within the preset error range. If all matches, proceed to the step of generating a release command; If any match is found, an interception command is generated and the exception event is logged.

[0013] Preferably, it also includes an intelligent verification step for empty ampoule destruction: The appearance image of the empty ampoule obtained during the recycling phase is compared with the preset standard image of the intact empty ampoule, and the weight of the empty ampoule measured by the second weight sensor is compared with the weight of the standard empty ampoule. If the image features match and the weight is within the preset error range, a destruction confirmation record is generated. If the image features do not match or the weight exceeds the error range, a destruction failure alarm will be generated.

[0014] On the other hand, the present invention also provides a full-process traceability management system for controlled substances, used to implement the above-mentioned method, including: The intelligent drug cabinet, deployed at the drug dispensing end, integrates a facial recognition module, an RFID reader, a first weight sensor, and a first communication module. A mobile nursing terminal, deployed at the medication dispensing end, integrates a signature entry module, a first image acquisition module, a timing module, and a second communication module. The mobile nursing terminal also integrates a drug recognition function and a micro switch. The intelligent recycling terminal, deployed at the recycling end, integrates a second image acquisition module, a second weight sensor, and a third communication module. The edge computing node is communicatively connected to the intelligent drug cabinet, the mobile nursing terminal, and the intelligent recycling terminal. The edge computing node is equipped with a dynamic risk scoring model, which is used to extract the behavioral fingerprint features of the operator based on the temporal attention mechanism and output a dynamic risk score. The cloud-based monitoring platform communicates with edge computing nodes to receive reports of abnormal events and store traceability records of controlled substances. The multimodal sensing data collected by the intelligent drug cabinet, mobile nursing terminal, and intelligent recycling terminal is sent to the edge computing node. The edge computing node performs three-source fusion verification and dynamic risk scoring on the multimodal sensing data, and generates an interception command or a release command based on the verification results and dynamic risk score. The generated command is then sent to the corresponding terminal to prohibit or allow the current operation.

[0015] Preferably, the dynamic risk scoring model includes: The input layer is used to receive historical behavior sequence data of operators, which includes drug dispensing time sequence, drug dispensing frequency sequence, single drug dispensing dosage sequence, drug use completion rate sequence, and empty ampoule recovery timeliness rate sequence. Long Short-Term Memory (LSTM) network layers are used to extract behavioral fingerprint features from historical behavioral sequence data; The attention layer is used to assign weights to different time steps in the historical behavior sequence data, where the time steps corresponding to abnormal behaviors are assigned higher weights to highlight their contribution to the risk score. The output layer is used to calculate and output a dynamic risk score based on the weighted behavioral fingerprint features; The long short-term memory network layer, attention layer, and output layer are deployed on edge computing nodes.

[0016] Preferably, the edge computing node further includes a three-source fusion verification module, which is connected to the dynamic risk scoring model and is used to verify the real-time data of the medication dispensing stage before the dynamic risk scoring model is calculated; the three-source fusion verification module is specifically used for: Receive the first identity data, the first drug identification data, and the first weight change data from the intelligent drug cabinet; Determine whether the first identity data matches the pre-stored list of operation permissions, whether the first drug identification data matches the outbound record of the intelligent drug cabinet, and whether the first weight change data matches the product of the standard weight of a single drug and the quantity of drugs within the preset error range. When all judgment results are a match, a verification pass signal will be sent to the dynamic risk scoring model to trigger risk score calculation; when any judgment result is a mismatch, an interception command will be sent directly to the intelligent drug cabinet and an abnormal event will be reported to the cloud monitoring platform.

[0017] Preferably, the cloud-based monitoring platform includes a cross-agency traceability chain construction module, which is used for: Construct an internal private chain to record fine-grained operation logs of the circulation of controlled drugs within a single medical institution. The fine-grained operation logs include at least the operation time, operator identification, drug batch number, drug quantity, operation type, and operation result. A regional consortium blockchain is constructed to record summary information on the flow of controlled drugs between different medical institutions. This summary information is obtained by hashing fine-grained operation logs. When a cross-agency traceability request is initiated, the integrity of the traceability data is verified by comparing the hash value of the summary information with that of the fine-grained operation log.

[0018] The technical effects and advantages of this invention are as follows: 1. This invention utilizes a dynamic risk scoring model deployed on edge computing nodes to extract behavioral fingerprint features of operators based on a temporal attention mechanism. Combined with real-time collected multimodal perception data, it performs dynamic risk assessment, enabling the determination of compliance before operation execution and generating interception instructions for abnormal operations in real time. This shifts the regulatory focus to the operation execution stage, effectively preventing drug misuse incidents. 2. On the one hand, this invention performs three-source fusion verification of operator identity, drug identification, and weight changes during the drug dispensing stage, realizing multi-dimensional real-time verification of human-drug-behavior. On the other hand, it extracts individualized behavioral fingerprint features from the operator's historical behavioral sequence data through a long short-term memory network, and strengthens the identification of abnormal behavior patterns by using an attention mechanism. This realizes the transformation of risk judgment standards from static and uniform to dynamic and individualized, significantly reducing the false alarm rate and false negative rate of abnormal behavior identification. 3. This invention adopts a two-layer architecture combining an internal private blockchain and a regional consortium blockchain. The private blockchain is used to record fine-grained operation logs of drugs within a single medical institution, while the consortium blockchain is used to record summary information of drug circulation between different medical institutions. The integrity of traceability data is ensured through hash calculation and comparison verification. At the same time, intelligent verification of empty ampoule destruction is carried out through a combination of image recognition and weight verification, filling management loopholes in the recycling and destruction process and forming a closed-loop supervision of the entire process from drug collection and use to recycling and destruction. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the structure of the full-process traceability management system for controlled substances provided in an embodiment of the present invention.

[0020] Figure 2 This is a flowchart illustrating the whole-process traceability management method for controlled substances provided in this embodiment of the invention.

[0021] Figure 3 This is a schematic diagram of the structure of the dynamic risk scoring model provided in an embodiment of the present invention.

[0022] The attached diagram is labeled as follows: 1. Intelligent drug cabinet; 10. Facial recognition module; 11. Radio frequency identification reader; 12. First weight sensor; 13. First communication module; 2. Mobile nursing terminal; 20. Signature input module; 21. First image acquisition module; 22. Timing module; 23. Second communication module; 3. Intelligent recycling terminal; 30. Second image acquisition module; 31. Second weight sensor; 32. Third communication module; 4. Edge computing node; 40. Three-source fusion verification module; 41. Dynamic risk scoring model; 5. Cloud monitoring platform; 50. Cross-institutional traceability chain construction module. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Example 1

[0024] As attached Figure 1 , 3 As shown in the figure, this embodiment provides a full-process traceability management system for controlled substances. The system includes an intelligent controlled substance cabinet 1, a mobile nursing terminal 2, an intelligent recycling terminal 3, an edge computing node 4, and a cloud monitoring platform 5.

[0025] The intelligent drug cabinet 1 is deployed at the drug dispensing end and integrates a face recognition module 10, an RFID reader 11, a first weight sensor 12, and a first communication module 13. The face recognition module 10 is used to collect the face image of the operator to identify the operator's identity, the RFID reader 11 is used to read the electronic supervision code of the drug, the first weight sensor 12 is used to detect the weight change of the drug before and after it is taken out, and the first communication module 13 is used to interact with the edge computing node 4. The mobile nursing terminal 2 is deployed at the medication dispensing end and integrates a signature entry module 20, a first image acquisition module 21, a timing module 22, and a second communication module 23. The signature entry module 20 is used to collect the electronic signature of the operator, the first image acquisition module 21 is used to collect the facial features of the medication recipient, the timing module 22 is used to record the medication timestamp, and the second communication module 23 is used to interact with the edge computing node 4. The mobile nursing terminal 2 also integrates a drug recognition function and a micro switch. The drug recognition function is used to read the electronic supervision code of the drug before medication to verify the drug information, and the micro switch is used to detect the drug's presence status. When the drug is taken out, a signal is triggered to record the time node of the retrieval action. The intelligent recycling terminal 3 is deployed at the recycling end and integrates a second image acquisition module 30, a second weight sensor 31 and a third communication module 32. The second image acquisition module 30 is used to acquire images of the appearance of empty ampoules, the second weight sensor 31 is used to verify the weight of empty ampoules, and the third communication module 32 is used to interact with the edge computing node 4. Edge computing node 4 is connected to intelligent drug cabinet 1, mobile nursing terminal 2, and intelligent recycling terminal 3 respectively. A dynamic risk scoring model 41 is deployed in edge computing node 4. This model is used to extract the behavioral fingerprint features of operators based on the temporal attention mechanism and output dynamic risk scores. Edge computing node 4 also includes a three-source fusion verification module 40, which is connected to the dynamic risk scoring model 41 and is used to verify the real-time data of the drug dispensing stage before the dynamic risk scoring model 41 calculates the risk score. Specifically, the three-source fusion verification module 40 receives first identity data, first drug identification data, and first weight change data from the intelligent drug cabinet 1, and determines whether the first identity data matches the pre-stored operation permission list, whether the first drug identification data matches the outbound record of the intelligent drug cabinet 1, and whether the first weight change data matches the product of the standard weight of a single drug and the number of drugs within a preset error range. The preset error range can be set according to the drug specifications, such as ±5% of the standard weight of a single drug. When all the judgment results are matched, the three-source fusion verification module 40 sends a verification pass signal to the dynamic risk scoring model 41 to trigger the risk score calculation; when any judgment result is a mismatch, the three-source fusion verification module 40 directly sends an interception command to the intelligent drug cabinet 1 and reports the abnormal event to the cloud monitoring platform 5. The dynamic risk scoring model 41 includes an input layer, a long short-term memory network layer, an attention layer, and an output layer. The input layer receives historical behavior sequence data of operators, including drug dispensing time sequence, drug dispensing frequency sequence, single drug dispensing dosage sequence, drug use completion rate sequence, and empty ampoule recycling timeliness rate sequence. This data is extracted from historical traceability records stored on the cloud monitoring platform 5. The long short-term memory network layer extracts behavioral fingerprint features from the historical behavior sequence data and learns the operator's behavior patterns through a gating mechanism, encoding the historical behavior data into feature vectors that reflect individual behavioral habits. The attention layer assigns weights to different time steps in the historical behavior sequence data, with time steps corresponding to abnormal behaviors assigned higher weights to highlight their contribution to the risk score. The output layer calculates and outputs the dynamic risk score based on the weighted behavioral fingerprint features. The long short-term memory network layer, attention layer, and output layer are all deployed on edge computing nodes 4. The cloud-based monitoring platform 5 communicates with the edge computing node 4 to receive reports of abnormal events and store traceability records of controlled drugs; The cloud-based monitoring platform 5 includes a cross-institutional traceability chain construction module 50. This module is used to build an internal private chain to record fine-grained operation logs of the flow of controlled drugs within a single medical institution. The fine-grained operation logs include at least the operation time, operator identification, drug batch number, drug quantity, operation type, and operation result. This module is also used to build a regional consortium chain to record summary information of the flow of controlled drugs between different medical institutions. The summary information is obtained by hashing the fine-grained operation logs. The hashing operation can use the SHA-256 algorithm. When a cross-institutional traceability request is initiated, the cloud-based monitoring platform 5 verifies the integrity of the traceability data by comparing the hash value of the summary information with that of the fine-grained operation logs. The multimodal sensing data collected by the intelligent drug cabinet 1, mobile nursing terminal 2, and intelligent recycling terminal 3 is sent to the edge computing node 4. The edge computing node 4 performs three-source fusion verification and dynamic risk scoring on the multimodal sensing data, and generates an interception command or a release command based on the verification results and dynamic risk score. The generated command is sent to the corresponding terminal to prohibit or allow the current operation. Example 2

[0026] As attached Figure 2 , 3 As shown, this embodiment provides a method for full-process traceability management of controlled substances. This method can be implemented based on the system of Embodiment 1, and specifically includes the following steps: S1. Multimodal sensing data acquisition During the circulation of controlled substances, multimodal sensing data is acquired. This data includes operator identification data, drug recipient identification data, drug identification data, weight change data, and operational video data. The specific acquisition methods are as follows: S11. During the drug dispensing stage, multimodal perception data is acquired through the intelligent drug cabinet 1. The face recognition module 10 integrated in the intelligent drug cabinet 1 acquires the operator's identity data as the first identity data. The radio frequency identification reader 11 acquires the drug electronic supervision code as the first drug identification data. The first weight sensor 12 acquires the weight difference before and after the drug is taken out as the first weight change data. Among them, the first drug identification data is used to identify the unique code of a single drug, and the first weight change data is used to verify the consistency between the number of drugs taken out and the radio frequency identification reading result. S12. During the medication administration phase, the operator's electronic signature is obtained through the mobile nursing terminal 2 as the second identity data, the first image acquisition module 21 obtains the facial features of the medication recipient as the third identity data, the timing module 22 records the medication timestamp, the mobile nursing terminal 2 also integrates a drug identification function and a micro switch, the drug identification function is used to read the drug electronic supervision code to verify the drug information before medication, the micro switch is used to detect the drug's in-situ status, and when the drug is taken out, a signal is triggered to record the time node of the taking action; S13. During the recycling phase, the image data of the empty ampoule is obtained through the intelligent recycling terminal 3 as the first operation image data, and the appearance image of the empty ampoule is collected through the second image acquisition module 30. The weight of the empty ampoule is checked in conjunction with the second weight sensor 31 to determine whether the empty ampoule has been emptied and has no residual medicine.

[0027] S2, Three-Source Fusion Verification Before generating the release order, the first identity data, the first drug identification data, and the first weight change data obtained during the drug collection stage are subjected to three-source fusion verification. S2 specifically includes: determining whether the first identity data matches the pre-stored list of operation permissions, whether the first drug identification data matches the outbound record of the intelligent drug cabinet 1, and whether the first weight change data matches the product of the standard weight of a single drug multiplied by the number of drugs within a preset error range. The preset error range can be set according to the drug specifications, such as ±5% of the standard weight of a single drug. If all three items match, then execute S3; if any one does not match, then generate an interception instruction and record the exception event.

[0028] S3, Dynamic Risk Score Calculation Multimodal perception data is input into the dynamic risk scoring model 41 deployed on the edge computing node 4. The dynamic risk scoring model 41 extracts the behavioral fingerprint features of the operator based on the temporal attention mechanism and outputs a dynamic risk score based on the behavioral fingerprint features. The dynamic risk scoring model 41 includes an input layer, a long short-term memory network layer, an attention layer, and an output layer. The input layer receives historical behavior sequence data of the operator, which includes drug dispensing time sequence, drug dispensing frequency sequence, single drug dispensing dosage sequence, drug use completion rate sequence, and empty ampoule return timeliness rate sequence. The long short-term memory network layer extracts behavioral fingerprint features from the historical behavior sequence data. The attention layer assigns weights to different time steps in the historical behavior sequence data, with the time steps corresponding to abnormal behaviors being assigned higher weights to highlight their contribution to the risk score. The output layer calculates and outputs the dynamic risk score based on the weighted behavioral fingerprint features.

[0029] S4. Risk Threshold Assessment and Instruction Generation Determine whether the dynamic risk score exceeds the preset risk threshold. The risk threshold can be dynamically set based on the percentile method or standard deviation method of historical operation data. If the threshold is exceeded, an interception command is generated to prohibit the current operation and report it to the cloud monitoring platform. If the threshold is not exceeded, a release order is generated, and traceability records of narcotic drugs are added to the cloud-based monitoring platform 5 based on multimodal perception data.

[0030] S5. Establish cross-agency traceability chains Adding storage traceability records to the cloud-based regulatory platform S5 includes building a cross-agency traceability chain. Specifically, S5 includes the following sub-steps: S51. Based on the internal private chain of the institution, record the fine-grained operation log of the circulation of controlled drugs within a single medical institution. The fine-grained operation log shall include at least the operation time, operator identification, drug batch number, drug quantity, operation type and operation result, and store the fine-grained operation log on the cloud supervision platform 5. S52. Based on the regional alliance chain, record the summary information of the flow of controlled drugs between different medical institutions. The summary information is obtained by hashing the fine-grained operation log. The hashing operation can use the SHA-256 algorithm. The summary information is stored in the cloud supervision platform 5. S53. When a cross-agency traceability request is initiated, the cloud-based monitoring platform 5 verifies the integrity of the traceability data by comparing the hash values ​​of the summary information and the fine-grained operation log. If the hash values ​​match, it proves that the traceability data has not been tampered with; if they do not match, it indicates that the data integrity has been compromised.

[0031] S6, Intelligent Verification of Empty Ampoule Disposal The recycling phase also includes a smart verification step for empty ampoule destruction. S6 specifically includes the following sub-steps: S61. Compare the appearance image of the empty ampoule obtained during the recycling stage with the preset standard image of the intact empty ampoule, and at the same time compare the weight of the empty ampoule measured by the second weight sensor 31 with the weight of the standard empty ampoule. S62. If the image features match and the weight is within the preset error range, a destruction confirmation record is generated. S63. If the image features do not match or the weight exceeds the error range, a destruction failure alarm will be generated.

[0032] In summary: This invention achieves intelligent traceability management of narcotic drugs throughout the entire process from drug collection and use to recycling and destruction by combining real-time acquisition and fusion verification of multimodal sensing data with dynamic risk assessment based on behavioral fingerprint features extracted from historical behavioral sequence data. The system can perform real-time risk judgment and interception before the operation occurs, effectively preventing drug misuse incidents. At the same time, it ensures the integrity and immutability of traceability data through cross-institutional traceability chains.

[0033] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for full-process traceability management of controlled substances, characterized in that: include: Acquire multimodal sensing data generated during the circulation of controlled drugs, wherein the multimodal sensing data includes at least operator identification data, drug recipient identification data, drug identification data, weight change data, and operation image data; Multimodal perception data is input into a dynamic risk scoring model (41) deployed on an edge computing node (4). The dynamic risk scoring model (41) extracts the behavioral fingerprint features of the operator based on a temporal attention mechanism and outputs a dynamic risk score based on the behavioral fingerprint features. Determine whether the dynamic risk score exceeds a preset risk threshold; If the limit is exceeded, an interception instruction is generated, which is used to prohibit the current operation and report it to the cloud monitoring platform (5). If the limit is not exceeded, a release order is generated, and traceability records of narcotic drugs are added to the cloud-based monitoring platform (5) based on the multimodal perception data.

2. The method according to claim 1, characterized in that: The dynamic risk scoring model (41) includes: The input layer is used to receive historical behavior sequence data of operators, which includes drug dispensing time sequence, drug dispensing frequency sequence, single drug dispensing dosage sequence, drug use completion rate sequence, and empty ampoule recovery timeliness rate sequence. Long Short-Term Memory (LSTM) network layers are used to extract behavioral fingerprint features from historical behavioral sequence data; The attention layer is used to assign weights to different time steps in the historical behavior sequence data, where the time steps corresponding to abnormal behaviors are assigned higher weights to highlight their contribution to the risk score. The output layer is used to calculate and output a dynamic risk score based on the weighted behavioral fingerprint features.

3. The method according to claim 1, characterized in that: Acquire multimodal sensing data generated during the circulation of controlled substances, specifically including: During the drug dispensing stage, multimodal perception data is obtained through the intelligent drug cabinet (1). The intelligent drug cabinet (1) integrates a face recognition module (10), a radio frequency identification reader (11), and a first weight sensor (12). Specifically, the face recognition module (10) obtains the operator's identity data as the first identity data, the radio frequency identification reader (11) obtains the drug electronic supervision code as the first drug identification data, and the first weight sensor (12) obtains the weight difference before and after the drug is taken out as the first weight change data. The first drug identification data is used to identify the unique code of a single drug, and the first weight change data is used to verify the consistency between the number of drugs taken out and the radio frequency identification reading result. During the medication administration phase, the operator's electronic signature is obtained through the mobile nursing terminal (2) as the second identity data, and the facial features of the drug recipient are obtained through the first image acquisition module (21) as the third identity data. The medication timestamp is also recorded. The mobile nursing terminal (2) also integrates a drug identification function, which is used to read the drug electronic supervision code before medication to verify the drug information, and a micro switch, which is used to detect the drug's presence status. When the drug is taken out, a signal is triggered to record the time node of the taking action. During the recycling phase, the image data of the empty ampoule is obtained by the intelligent recycling terminal (3) as the first operation image data, and the appearance image of the empty ampoule is collected by the second image acquisition module (30). The weight of the empty ampoule is checked by the second weight sensor (31) to determine whether the empty ampoule has been emptied and has no residual medicine.

4. The method according to claim 3, characterized in that: Adding traceability records for controlled substances to the cloud-based regulatory platform (5) includes building a cross-agency traceability chain, specifically including: Based on the internal private chain of the institution, a fine-grained operation log is recorded to record the circulation of controlled drugs within a single medical institution. The fine-grained operation log includes at least the operation time, operator identification, drug batch number, drug quantity, operation type and operation result, and the fine-grained operation log is stored in the cloud supervision platform (5). Based on the regional alliance chain, the summary information of the flow of controlled drugs between different medical institutions is recorded. The summary information is obtained by hashing the fine-grained operation log and stored in the cloud supervision platform (5). When a cross-agency traceability request is initiated, the cloud-based monitoring platform (5) verifies the integrity of the traceability data by comparing the hash value of the summary information with that of the fine-grained operation log.

5. The method according to claim 3, characterized in that: Before generating the release command, the following is also included: The first identity data, the first drug identification data, and the first weight change data obtained during the drug dispensing stage are subjected to three-source fusion verification to determine whether the first identity data matches the pre-stored operation permission list, whether the first drug identification data matches the outbound record of the intelligent drug cabinet (1), and whether the first weight change data matches the product of the standard weight of a single drug multiplied by the number of drugs within the preset error range. If all matches, proceed to the step of generating a release command; If any match is found, an interception command is generated and the exception event is logged.

6. The method according to claim 3, characterized in that: It also includes a smart verification process for empty ampoule disposal: The appearance image of the empty ampoule obtained during the recycling phase is compared with the preset standard image of the intact empty ampoule, and the weight of the empty ampoule measured by the second weight sensor is compared with the weight of the standard empty ampoule. If the image features match and the weight is within the preset error range, a destruction confirmation record is generated. If the image features do not match or the weight exceeds the error range, a destruction failure alarm will be generated.

7. A full-process traceability management system for controlled substances, used to implement the method described in any one of claims 1-6, characterized in that: include: The intelligent drug cabinet (1) is deployed at the drug dispensing end and integrates a face recognition module (10), a radio frequency identification reader (11), a first weight sensor (12) and a first communication module (13). The mobile nursing terminal (2) is deployed at the medication terminal and integrates a signature entry module (20), a first image acquisition module (21), a timing module (22) and a second communication module (23). The mobile nursing terminal (2) also integrates a drug identification function and a micro switch. The intelligent recycling terminal (3) is deployed at the recycling end and integrates a second image acquisition module (30), a second weight sensor (31) and a third communication module (32). The edge computing node (4) is connected to the intelligent drug cabinet (1), the mobile nursing terminal (2), and the intelligent recycling terminal (3) respectively. The edge computing node (4) is equipped with a dynamic risk scoring model (41). The dynamic risk scoring model (41) is used to extract the behavioral fingerprint features of the operator based on the temporal attention mechanism and output the dynamic risk score. The cloud-based monitoring platform (5) is connected to the edge computing node (4) to receive abnormal event reports and store traceability records of narcotic drugs; The multimodal perception data collected by the intelligent drug cabinet (1), mobile nursing terminal (2), and intelligent recycling terminal (3) is sent to the edge computing node (4). The edge computing node (4) performs three-source fusion verification and dynamic risk scoring on the multimodal perception data, and generates an interception command or a release command based on the verification results and dynamic risk scoring, and sends the generated command to the corresponding terminal to prohibit or allow the current operation.

8. The system according to claim 7, characterized in that: The dynamic risk scoring model (41) includes: The input layer is used to receive historical behavior sequence data of operators, which includes drug dispensing time sequence, drug dispensing frequency sequence, single drug dispensing dosage sequence, drug use completion rate sequence, and empty ampoule recovery timeliness rate sequence. Long Short-Term Memory (LSTM) network layers are used to extract behavioral fingerprint features from historical behavioral sequence data; The attention layer is used to assign weights to different time steps in the historical behavior sequence data, where the time steps corresponding to abnormal behaviors are assigned higher weights to highlight their contribution to the risk score. The output layer is used to calculate and output a dynamic risk score based on the weighted behavioral fingerprint features; The long short-term memory network layer, attention layer and output layer are deployed on edge computing nodes (4).

9. The system according to claim 7, characterized in that: The edge computing node (4) also includes a three-source fusion verification module (40), which is connected to the dynamic risk scoring model (41) and is used to verify the real-time data of the drug dispensing stage before the dynamic risk scoring model (41) is calculated; the three-source fusion verification module (40) is specifically used for: Receive first identity data, first drug identification data and first weight change data from the intelligent drug cabinet (1); Determine whether the first identity data matches the pre-stored list of operation permissions, whether the first drug identification data matches the outbound record of the intelligent drug cabinet (1), and whether the first weight change data matches the product of the standard weight of a single drug multiplied by the number of drugs within the preset error range. When all judgment results are a match, a verification pass signal is sent to the dynamic risk scoring model (41) to trigger risk scoring calculation; when any judgment result is a mismatch, an interception command is sent directly to the intelligent drug cabinet (1) and an abnormal event is reported to the cloud monitoring platform (5).

10. The system according to claim 7, characterized in that: The cloud-based regulatory platform (5) includes a cross-agency traceability chain construction module (50), which is used for: Construct an internal private chain to record fine-grained operation logs of the circulation of controlled drugs within a single medical institution. The fine-grained operation logs include at least the operation time, operator identification, drug batch number, drug quantity, operation type, and operation result. A regional consortium blockchain is constructed to record summary information on the flow of controlled drugs between different medical institutions. This summary information is obtained by hashing fine-grained operation logs. When a cross-agency traceability request is initiated, the integrity of the traceability data is verified by comparing the hash value of the summary information with that of the fine-grained operation log.

Images