A medical consumable kit internal component use efficiency analysis and AI optimization system

Abstract

This invention discloses a system for analyzing and optimizing the usage efficiency of components within a medical consumable kit using AI, relating to the field of intelligent management of medical consumables. The system includes a component usage status perception module, a component usage efficiency analysis module, and a component configuration AI optimization module. The status perception module acquires usage status data for each component within the kit under a target surgical procedure. The efficiency analysis module calculates component-level usage efficiency indicators based on the usage status data. The AI ​​optimization module is built based on a reinforcement learning algorithm, using the component-level usage efficiency indicator as the status input, the component configuration scheme as the action output, and a comprehensive score based on usage efficiency feedback results or usage efficiency and low-carbon feedback results as the reward signal. Through continuous iterative learning, it generates and dynamically optimizes the optimal component configuration scheme. Thus, through the collaborative work of the aforementioned modules, a refined analysis of the usage efficiency of medical consumable kit components and an AI-driven continuous optimization closed loop can be achieved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent management technology for medical consumables, and specifically relates to a system for analyzing and optimizing the usage efficiency of components within a medical consumable package using AI. Background Technology

[0002] With the development of medical technology and the continuous increase in surgical volume, medical consumables management plays an increasingly important role in hospital operations. As a major area of ​​resource consumption in the hospital, the efficiency of consumables management in the operating room directly impacts the hospital's operating costs and the quality of medical services. To improve surgical preparation efficiency and the standardization of consumables management, medical institutions generally use pre-assembled and packaged medical consumable kits (also known as surgical kits or procedure kits) to supply the consumables required for surgery. This kitting model, through standardized supply, reduces the time circulating nurses spend managing consumables to a certain extent, thereby improving the efficiency of operating room operations.

[0003] However, existing medical consumable package management solutions have the following shortcomings: (1) Existing technologies have significant limitations in the fine analysis of component usage efficiency. Current management methods mainly rely on manual recording and post-event statistics, which can only obtain the overall consumption of the kit and cannot conduct fine analysis on the usage status of each component in the kit. In clinical practice, a considerable proportion of components in surgical kits are discarded without being actually used during surgery. Studies have shown that optimizing surgical kits can significantly reduce medical waste and related costs. However, in the absence of systematic component-level usage efficiency analysis tools, medical institutions find it difficult to accurately identify which components in the kit are over-configured or under-configured, and it is also difficult to quantify and evaluate the rationality and economy of component configuration schemes. (2) Existing technologies have significant shortcomings in low-carbon management. Surgical waste accounts for 20% to 30% of the total waste in hospitals, and the energy consumption intensity of the operating room is 3 to 6 times that of other departments in the hospital. Disposable surgical consumables are one of the products with the highest carbon intensity in the medical field. Existing studies have achieved significant carbon emission reduction by reasonably simplifying the configuration of surgical tray consumables, but the current management plan does not take carbon emission factors into consideration when optimizing the configuration of the kit. Traditional kit configuration is usually only guided by clinical needs and procurement costs, lacks carbon emission quantification methods based on life cycle assessment, and has not formed a closed-loop management mechanism to feed back carbon emission data to the kit optimization decision. (3) Existing kit configuration optimization schemes are mostly one-time adjustments based on static rules or human experience. For example, some studies have adjusted kit configurations by analyzing unused consumables during surgery, achieving cost savings and waste reduction. However, such schemes lack the ability to continuously track and automatically optimize changes in clinical practice data. When the surgical procedure, doctor team, or patient structure changes, static configuration schemes are difficult to adapt and adjust, resulting in unsustainable optimization effects. Although some hospitals have already transformed and applied medical consumable management systems based on disease-specific consumable recommendation intelligent agents, their core lies in consumable recommendation based on intelligent agents, without introducing a continuous self-learning optimization closed loop with usage efficiency indicators as state input and comprehensive scores as reward signals.

[0004] In summary, given the key technical problems that existing technologies have not yet solved in the field of medical consumable package management, how to provide a new closed-loop solution that can perform refined usage efficiency analysis of each component in the package and continuously self-optimize based on the analysis results using AI algorithms is a topic that urgently needs to be studied by those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a medical consumable kit component usage efficiency analysis and AI optimization system to address the significant limitations of existing medical consumable kit management schemes in terms of refined analysis of component usage efficiency, significant deficiencies in low-carbon management, and / or the lack of continuous tracking and automatic optimization capabilities for changes in clinical practice data due to the fact that kit configuration optimization schemes are mostly based on static rules or one-time adjustments made by human experience.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a system for analyzing and optimizing the utilization efficiency of components within a medical consumable kit using AI, comprising: The component usage status awareness module is used to obtain the usage status data of each component in the medical consumable kit under the target procedure. The medical consumable kit refers to a collection unit that is required for use under the target procedure and is composed of multiple medical consumable components that are pre-combined and packaged. The usage status data adopts any one of the states determined from the fully used state, the partially used state, and the unused state. The component utilization efficiency analysis module, communicatively connected to the component utilization status sensing module, is used to calculate the component-level utilization efficiency index of the medical consumable kit based on the utilization status data of each component. The component-level utilization efficiency index includes component utilization rate, surgical procedure matching degree, configuration redundancy, and / or waste index. The component utilization rate characterizes the ratio between the number of components actually used and the total number of components in the medical consumable kit. The surgical procedure matching degree characterizes the degree of conformity between the component configuration scheme of the medical consumable kit and the clinical guidelines or historical practices of the target surgical procedure. The configuration redundancy characterizes the ratio between the number of components in the medical consumable kit that are not fully used and the total number of components. The waste index characterizes the ratio between the number of components in the medical consumable kit that are discarded without being used and the total number of components. The component configuration AI optimization module is communicatively connected to the component usage efficiency analysis module. It is used to take component-level usage efficiency indicators as state input, component configuration schemes as action output, and a comprehensive score based on usage efficiency feedback results or usage efficiency and low-carbon feedback results as reward signals. Through continuous iterative learning based on reinforcement learning algorithms, it generates and dynamically optimizes the component configuration scheme of the medical consumable kit corresponding to the target procedure.

[0007] Based on the above-mentioned invention, a novel closed-loop solution is provided that can perform refined usage efficiency analysis on each component within a kit and continuously self-optimize based on the analysis results using AI algorithms. This solution includes a component usage status perception module, a component usage efficiency analysis module, and a component configuration AI optimization module. The component usage status perception module acquires usage status data for each component within the kit under the target surgical procedure. The usage status data adopts any one of the following states: fully used, partially used, and unused. The component usage efficiency analysis module calculates component-level usage efficiency indicators based on the usage status data, including component utilization rate, surgical procedure matching degree, etc. The system configures redundancy and waste indexes. The component configuration AI optimization module is built based on reinforcement learning algorithms. It takes component-level usage efficiency indicators as state inputs, component configuration schemes as action outputs, and a comprehensive score based on usage efficiency feedback results or usage efficiency and low-carbon feedback results as reward signals. Through continuous iterative learning, it generates and dynamically optimizes the optimal component configuration scheme. Thus, through the collaborative work of the aforementioned modules, it can achieve a refined analysis of the usage efficiency of medical consumable kit components and a closed loop of AI-driven continuous optimization. This provides medical institutions with an intelligent consumable management solution that combines clinical safety, operational efficiency, and low-carbon benefits, making it easy to apply and promote in practice.

[0008] In one possible design, the component uses a state-aware module including: A multimodal information acquisition unit is used to acquire preoperative information sets of each component in the medical consumable kit before surgery, and to acquire postoperative information sets of each component after surgery, wherein the preoperative information sets and the postoperative information sets are acquired by at least two of the following methods: RFID tag reading, image recognition, and weight sensing. The component usage status determination unit is communicatively connected to the multimodal information acquisition unit. For each component, based on multidimensional difference parameters between the corresponding preoperative and postoperative information sets, it outputs a corresponding usage status determination result using a component usage status classification model pre-trained using a deep learning algorithm. It then generates corresponding usage status data based on the usage status determination result. The usage status determination result includes three discrete labels: fully used, partially used, and unused. The determination of the partially used state is based on quantitative analysis of at least one feature among component surface morphology change characteristics, weight change characteristics, and packaging integrity change characteristics.

[0009] In one possible design, the component uses an efficiency analysis module including: The efficiency index calculation unit is used to calculate the initial component-level usage efficiency index of the medical consumable kit based on the usage status data of each component. The association model construction unit is used to construct an association model of medical consumables consumption based on the medical consumables usage records of different surgical procedures and different doctors in historical surgical data. The association model of medical consumables consumption is used to record the probability of use, quantity and model preference of each doctor for various components under different surgical procedures, so as to identify the differences between the medical consumables usage preferences of different doctors under the same surgical procedure and the best practices. An efficiency index correction unit is communicatively connected to the efficiency index calculation unit and the association model construction unit, respectively. It is used to correct the initial component-level usage efficiency index based on the procedure-doctor-component consumption association model, and to correct the procedure-level usage efficiency index by doctor preference and procedure difference, so as to obtain the corrected component-level usage efficiency index. The doctor preference correction is used to distinguish the difference between doctor's personal habitual use and clinically necessary use, and the procedure difference correction is used to distinguish the difference between different surgical complexities and different procedure subtypes.

[0010] In one possible design, the comprehensive score is calculated based on a reward function that includes a usage efficiency reward, a low-carbon reward, and a completeness reward. The value of the usage efficiency reward is positively correlated with the component-level usage efficiency index from the efficiency feedback results. The value of the low-carbon reward is negatively correlated with the estimated carbon emissions from the low-carbon feedback results and the component configuration scheme corresponding to the medical consumable kit. The value of the completeness reward is positively correlated with the clinical completeness verification results of the component configuration scheme in meeting the surgical requirements.

[0011] In one possible design, the estimated carbon emissions are calculated as follows: Obtain the type, quantity, and material of each component in the component configuration scheme of the medical consumable kit, and also obtain at least one of the following for each component: supplier, transportation distance, and sterilization method; Based on the obtained results, the full-cycle carbon emissions corresponding to the component configuration scheme are estimated using a preset life cycle assessment model as the estimated carbon emissions.

[0012] In one possible design, the component configuration AI optimization module is also used to calculate the actual carbon emissions based on the actual used components and unused returned components after each actual surgery of the target procedure, and to feed back the deviation between the actual carbon emissions and the estimated carbon emissions to the life cycle assessment model and the reward function, so as to continuously improve the accuracy of carbon emission estimation and the accuracy of low-carbon decision-making.

[0013] In one possible design, the component configuration AI optimization module employs a multi-objective reinforcement learning framework to simultaneously optimize at least two objectives, and achieves a dynamic trade-off between the at least two objectives through Pareto front solving or weighted scalarization. The at least two objectives include: maximizing component utilization efficiency, minimizing carbon emissions, maximizing clinical kitting, minimizing consumable costs, and minimizing waste generation.

[0014] In one possible design, an intraoperative real-time monitoring module is also included, which is communicatively connected to the component usage status sensing module. This module is used to identify intraoperative component usage behavior in real time using computer vision technology during the surgery. When at least one wasteful behavior of a consumable component is detected based on the identification results, the module proactively sends a reminder message to the operating room terminal or a medical staff mobile terminal. The at least one wasteful behavior of a consumable component includes abnormal selection of consumable component specifications, improper use, and / or unnecessary unpacking. The reminder message includes a description of the identified wasteful behavior, component information of the wasted component, and / or suggested alternative operating procedures.

[0015] In one possible design, a configuration report automatic generation module is also included, which is communicatively connected to the component configuration AI optimization module and the component usage efficiency analysis module. The configuration report automatic generation module is used to call a large language model, combine the optimization results of the component configuration scheme, component-level usage efficiency indicators, surgical procedure clinical guidelines and doctor operation preference data, and adopt a multi-round inference strategy based on a preset Prompt template to sequentially generate an efficiency comparison analysis paragraph, an optimization suggestion analysis paragraph and a risk warning paragraph, and finally synthesize and generate a structured medical consumable package configuration optimization suggestion report under the target surgical procedure.

[0016] In one possible design, an intelligent sorting and automatic packaging module is also included, which is communicatively connected to the component configuration AI optimization module. The intelligent sorting and automatic packaging module is used to control the intelligent sorting execution structure to sort out the corresponding components from the consumables library according to the optimal component configuration scheme generated and output by the component configuration AI optimization module after receiving the optimal component configuration scheme. After sorting, the module automatically packages the components to form a medical consumable package corresponding to the target procedure.

[0017] The beneficial effects of the above scheme are: (1) This invention provides a novel closed-loop solution capable of performing refined usage efficiency analysis on each component within a kit and continuously self-optimizing based on the analysis results using AI algorithms. This solution includes a component usage status perception module, a component usage efficiency analysis module, and a component configuration AI optimization module. The component usage status perception module acquires usage status data for each component within the kit under the target procedure. The usage status data is determined from any one of the following states: fully used, partially used, and unused. The component usage efficiency analysis module calculates component-level usage efficiency indicators based on the usage status data, including component utilization rate, procedure rate, and other performance metrics. The module measures matching degree, configuration redundancy, and waste index. The component configuration AI optimization module is built based on reinforcement learning algorithm. It takes component-level usage efficiency indicators as state input, component configuration schemes as action output, and a comprehensive score based on usage efficiency feedback results or usage efficiency and low-carbon feedback results as reward signal. Through continuous iterative learning, it generates and dynamically optimizes the optimal component configuration scheme. Thus, through the collaborative work of the aforementioned modules, it can realize a refined analysis of the usage efficiency of medical consumable kit components and a closed loop of AI-driven continuous optimization, providing medical institutions with an intelligent consumable management solution that combines clinical safety, operational efficiency, and low-carbon benefits. (2) The component usage status awareness module can accurately determine the full usage status, partial usage status and unused status of each component in the kit after a single operation by collecting multimodal information and using a deep learning status classification model, providing a refined data foundation for subsequent efficiency analysis; (3) The component utilization efficiency analysis module calculates component-level utilization efficiency indicators such as component utilization rate, surgical procedure matching degree, configuration redundancy and waste index based on status data. It also performs doctor preference correction and surgical procedure difference correction through the surgical procedure-doctor-component consumption correlation model, which can distinguish the difference between doctors' personal habitual use and clinically necessary use, making the efficiency evaluation results more objective and accurate. (4) The component configuration AI optimization module takes the component-level usage efficiency index as the state input, the component configuration scheme as the action output, and the comprehensive score including usage efficiency feedback results or usage efficiency and low carbon feedback results as the reward signal. Through continuous iterative learning based on reinforcement learning algorithm, it can generate and dynamically optimize the optimal component configuration scheme corresponding to the target technique, overcoming the limitations of the one-time adjustment scheme based on static rules or human experience in the existing technology. (5) By using the full-cycle carbon emissions estimated by the life cycle assessment model as the calculation input for low-carbon reward items, and by feeding back the deviation between the actual carbon emissions after surgery and the estimated carbon emissions before surgery to the life cycle assessment model and reward function through the carbon trace feedback mechanism, carbon emission factors can be incorporated into the closed-loop management of package configuration optimization, which can continuously improve the accuracy of carbon emission estimation and the accuracy of low-carbon decision-making. (6) By using a multi-objective reinforcement learning framework, multiple objectives such as component utilization efficiency, carbon emissions, clinical completeness, consumable costs and waste generation can be optimized simultaneously. This enables a scientific balance when multiple objectives conflict, taking into account the comprehensive needs of clinical safety, operational efficiency and green management. (7) Through the real-time computer vision recognition and active reminder of the intraoperative real-time monitoring module, it is also possible to intervene in the waste of consumables during the operation. Furthermore, through the configuration report automatic generation module, it can call the large language model to automatically generate a structured optimization suggestion report, which can improve the efficiency and scientific nature of management decisions. In addition, through the intelligent sorting and automatic packaging module, the AI ​​optimization scheme can be automatically transformed into a physical package entity, which can realize end-to-end intelligent closed-loop management from perception, analysis, optimization to execution, which is convenient for practical application and promotion. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the structure of the medical consumable kit component usage efficiency analysis and AI optimization system provided in an embodiment of the present invention. Detailed Implementation

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these embodiments without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0021] It should be understood that although the terms "first" and "second", etc., may be used herein to describe various objects, these objects should not be limited by these terms. These terms are only used to distinguish one object from another. For example, the first object may be referred to as the second object, and similarly, the second object may be referred to as the first object, without departing from the scope of the exemplary embodiments of the invention.

[0022] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, or A and B exist simultaneously. Another example is A, B and / or C, which can mean that any one of A, B, and C or any combination thereof exists. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone or A and B exist simultaneously. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0023] Example like Figure 1 As shown, the medical consumable kit component usage efficiency analysis and AI optimization system provided in this embodiment includes, but is not limited to, a component usage status perception module, a component usage efficiency analysis module, and a component configuration AI optimization module.

[0024] The component usage status sensing module is used to acquire usage status data of each component within the medical consumable kit under the target surgical procedure. The medical consumable kit refers to a collection unit consisting of multiple pre-assembled and packaged medical consumable components required for the target surgical procedure. The usage status data is determined from any one of the following states: fully used, partially used, and unused. The target surgical procedure refers to the specific surgical type targeted by the system of this invention for analysis and optimization. For example, the target surgical procedure can be any standardized surgical type such as anterior cervical fusion, total knee replacement, laparoscopic cholecystectomy, percutaneous coronary intervention, hip replacement, or posterior lumbar interbody fusion. Different surgical procedures require different medical consumable kits with varying component configurations. For example, a medical consumable kit for a total knee replacement typically includes more than ten components such as bone cement, a pulse irrigation device, sutures, surgical gloves, an electrosurgical pen, a suction tube, and sterile dressings. Conversely, a medical consumable kit for a laparoscopic cholecystectomy typically includes laparoscopic-specific components such as a pneumoperitoneum needle, a trocar, titanium clips, and an electrocautery hook. This invention's system can acquire usage status data for each component within the corresponding medical consumable kit for any given surgical procedure, enabling efficiency analysis and configuration optimization. Furthermore, within the medical consumable kit, each component is pre-operatively assembled and packaged in a sterile package by a sterilization supply center or SPD (Supply Processing Distribution) service provider according to the clinical needs of the target procedure, forming a standard kit corresponding to that procedure.

[0025] In one specific embodiment, the three usage states described above have the following meanings: the fully used state refers to a state in which the component has been unsealed and completely consumed during surgery (such as a screw implanted in the patient's body or a completely injected bone cement, etc.) or is fully used; the partially used state refers to a state in which the component has been unsealed but only partially consumed, with the remaining part unused and non-returnable (such as only partially taken from a package containing multiple gauze pads or only partially injected into a multi-dose drug, etc.); and the unused state refers to a state in which the component has not been unsealed, has not been used, and can be completely returned.

[0026] In one embodiment of the present invention, the component usage status perception module includes, but is not limited to, a multimodal information acquisition unit and a component usage status determination unit; these two units work together to automatically determine the usage status of each component in the kit by collecting information twice before and after the operation and comparing the differences based on a deep learning model. In detail, the multimodal information acquisition unit is used to acquire preoperative information sets of each component in the medical consumable kit before surgery, and postoperative information sets of each component after surgery. The preoperative and postoperative information sets are obtained through at least two methods, such as RFID (Radio Frequency Identification) tag reading, image recognition, and weight sensing. The component usage status determination unit is communicatively connected to the multimodal information acquisition unit. For each component, based on multidimensional difference parameters between the corresponding preoperative and postoperative information sets, it outputs a corresponding usage status determination result using a component usage status classification model pre-trained based on a deep learning algorithm. It also generates corresponding usage status data based on the usage status determination result. The usage status determination result includes, but is not limited to, three types of discrete tags: fully used, partially used, and unused. The determination of the partially used state is based on quantitative analysis of at least one feature, such as component surface morphology change characteristics, weight change characteristics, and packaging integrity change characteristics.

[0027] The timing and purpose of the two data collections differ: Preoperative data collection is conducted before the surgery begins, when the kit arrives at the operating room but has not yet been unpacked or used—at this time, all components are in a sterile and sealed state. The preoperative information set records the initial baseline state of each component, including the component's identification information (e.g., RFID tag code), initial image features (e.g., component appearance and packaging integrity), initial weight value, and structured text information such as batch number and expiration date obtained from the certificate of conformity OCR (Optical Character Recognition). Postoperative data collection is conducted after the surgery, when the remaining components in the kit (including unused and partially used components) have not yet been returned to the SPD warehouse—at this time, each component has undergone the use of a complete surgery, and its physical state may have changed. The postoperative information set is used to compare the differences with the preoperative information set to determine the actual usage status of each component. Specifically, the methods for collecting the preoperative and postoperative information sets are as follows: The RFID tag reading method involves automatically reading the RFID tag information attached to each component using RFID readers deployed in the package unpacking area and component recycling area to obtain the unique identification code, batch number, and specifications of each component; the image recognition method involves taking multi-angle photos of each component using an industrial camera or high-definition camera to obtain the component's appearance image and extract visual features such as surface morphology, edge contours, and packaging status (where surface morphology features reflect minor scratches, indentations, and distortions on the component surface; packaging integrity features reflect the degree of unpacking of the component packaging, such as whether it has been torn, has wrinkles, or is damaged); and the weight sensing method involves weighing each component individually using a high-precision electronic weighing module to obtain the weight value of each component (weight change is one of the key physical indicators for determining whether a component has been used or partially used).

[0028] In the process of determining the usage status of components, both high-value and low-value consumables can be determined in this way: High-value consumables (such as orthopedic implants or cardiac stents) usually have high unit prices and strict traceability requirements, and their usage status determination requires extremely high accuracy. After surgery, it is necessary to distinguish between three statuses: "fully used (implanted in the patient's body)," "unused (unopened and can be returned)," and "partially used (although opened, it was not implanted due to temporary size adjustment during the operation, and special traceability marking is required)." Low-value consumables (such as gauze, gloves, or syringes) have lower unit prices but large usage. Their usage status determination focuses on batch identification and efficiency assessment. After surgery, it is necessary to distinguish between three statuses: "fully used (consumed)," "unused (unopened, can be resterilized or returned)," and "partially used (opened but not used up, needs to be treated as medical waste)," and calculate the overall waste index of the kit accordingly. For example, regarding RFID information contained in preoperative and postoperative information sets, if the RFID tag of a certain component can be read before the operation but cannot be read after the operation, it indicates that the component has been completely consumed during the operation, corresponding to a fully used state; if the RFID tag of a certain component can be read both before and after the operation, it indicates that the component still exists after the operation (it may be completely unused or partially unused), and further judgment needs to be made in combination with other dimensions of difference characteristics.

[0029] Specifically, the determination of the usage status is based on quantitative analysis of at least one of the following features: component surface morphology change characteristics, weight change characteristics, and packaging integrity change characteristics. In a preferred embodiment, the above three features can be used in combination to improve the accuracy of the determination. Specifically: Surface morphology change features are obtained by using a deep learning model to analyze the morphological differences between pre- and post-operative component images—quantitative parameters such as surface roughness, scratch density, and edge deformation are extracted from the two images. When the difference exceeds a preset threshold, it indicates that the component may have been used. Weight change features are one of the core indicators for determining whether a component has been partially used—the weight values ​​of the component are directly compared before and after surgery. When the weight reduction exceeds a preset weighing error tolerance range but does not reach the weight reduction threshold corresponding to "full use," it is determined to be partially used. For example, a kit containing 10 gauze pads weighs 50 grams before surgery and 30 grams after surgery, a reduction of 20 grams (corresponding to approximately 4 gauze pads being used). At the same time, surface morphology analysis shows that the packaging has been opened. The two features complement each other and can be used to determine that it has been partially used. Packaging integrity change features are obtained by detecting the unsealing status of the component packaging through image recognition technology—for cases where the packaging is intact in the pre-operative image but has been opened in the post-operative image but the contents have not been completely removed, the above weight change features and surface morphology change features are combined for comprehensive judgment, which can accurately quantify the degree of unsealing. In a specific implementation, the state classification model used by the above components can be conventionally trained using a deep convolutional neural network structure. It can use pre- and post-operative image pairs, weight difference values, and RFID reading status as input features of the model, output a three-class probability distribution, and take the class with the highest probability as the final usage status determination result.

[0030] The component utilization efficiency analysis module, communicatively connected to the component utilization status perception module, is used to calculate the component-level utilization efficiency index of the medical consumable kit based on the utilization status data of each component. The component-level utilization efficiency index includes, but is not limited to, component utilization rate, surgical procedure matching degree, configuration redundancy, and / or waste index. The component utilization rate characterizes the ratio between the number of components actually used and the total number of components in the medical consumable kit. The surgical procedure matching degree characterizes the degree of conformity between the component configuration scheme of the medical consumable kit and the clinical guidelines or historical practices of the target surgical procedure. The configuration redundancy characterizes the ratio between the number of components in the medical consumable kit that are not fully used and the total number of components. The waste index characterizes the ratio between the number of components in the medical consumable kit that are discarded without being used and the total number of components. These four indicators quantify and evaluate the rationality of the kit's component configuration from different dimensions, providing multi-dimensional status input for the subsequent AI dynamic optimization module. In one embodiment, the definitions and calculation methods of the aforementioned four indicators are shown in (A) to (D).

[0031] (A) Regarding the component utilization rate: For example, for a kit corresponding to a total knee arthroplasty, the kit contains a total of N components (or component sets, each set being considered a counting unit), including bone cement, pulse irrigation device, sutures, surgical gloves, electrosurgical pen, and suction tube. After surgery, the component usage status sensing module determines that the number of components actually used (i.e., in a fully used or partially used state) is M. Then, the component utilization rate of the kit is M÷N. In this example, if the kit contains 12 components in a certain surgery, of which 10 components are fully used (e.g., bone cement is fully injected and electrosurgical pen is used), 1 component is partially used (e.g., a portion of a pack of gauze is taken out), and 1 component is not used (e.g., a spare type of catheter is not opened), then the component utilization rate of the kit is 11÷12≈91.7%. In another example, for low-value consumable components, considering that they are usually counted in package units (such as 10 pieces / pack of gauze) rather than individually, the "number of components" in "component utilization" can be calculated based on the smallest unit of use. If 4 pieces of a component package containing 10 pieces of gauze are discarded without being used, the utilization rate of the component is 60%, and this calculation result will be included in the calculation of the overall component utilization rate of the package.

[0032] (B) Regarding the surgical procedure matching degree: In one embodiment, the surgical procedure matching degree can be calculated by constructing a standard component requirement list for the target surgical procedure—this list is derived from authoritative clinical guidelines, expert consensus, or historically optimal configuration templates obtained through large-scale statistical analysis for the surgical procedure—and comparing the component configuration of the current kit with this standard list, calculating the weighted conformity between the two in three dimensions: component type, specifications, and quantity configuration. For example, if the standard list for the target surgical procedure requires the configuration of "6 type A screws and 4 type B screws," and the current kit is configured with "6 type A screws and 3 type B screws," then the conformity of the number of type B screws is 75%. This single deviation will affect the overall surgical procedure matching degree score. The higher the surgical procedure matching degree, the closer the kit configuration is to clinical best practices, and the better it can ensure the smooth progress of the surgery.

[0033] (C) Regarding the configuration redundancy: The configuration redundancy is calculated based on the usage status data output by the component usage status awareness module—components in both partially used and unused states are considered as not fully used components. In the example of total knee arthroplasty above, one of the 12 components is in a partially used state and one is in an unused state, so the configuration redundancy of this kit is 2 ÷ 12 ≈ 16.7%. The higher the configuration redundancy, the more components are configured in the kit but not fully utilized during surgery, indicating an over-configuration problem.

[0034] (D) Regarding the waste index: Unlike configuration redundancy, the waste index only includes components that are discarded before being used (i.e., unused, unopened but cannot be returned for reuse due to sterilization deadlines or management process limitations), and the remaining parts discarded while in a partially used state. It does not include parts that are not fully used but can be returned (e.g., unopened spare components). In the example above, if one of the 12 components is in a partially used state and its remaining part is discarded, and one unused component is discarded because it cannot be returned, then the waste index for this kit is 2 ÷ 12 ≈ 16.7%. The waste index directly reflects the degree of ineffective consumption of disposable consumables and is an important quantitative indicator for evaluating the economic and environmental benefits of kit configuration schemes. In one embodiment, the waste index can also be calculated using a cost-weighted method—incorporating the purchase cost of each discarded component into the weight, so that the waste of high-priced consumables is more prominently reflected in the index.

[0035] In a preferred embodiment of the present invention, the component uses an efficiency analysis module that includes, but is not limited to, an efficiency index calculation unit, an association model construction unit, and an efficiency index correction unit; these three units work together to realize the quantitative calculation of the initial efficiency index and the correction and optimization based on historical data. In detail, the efficiency index calculation unit is used to calculate the initial component-level usage efficiency index of the medical consumable kit based on the usage status data of each component; the association model construction unit is used to construct a procedure-doctor-component consumption association model based on the medical consumable usage records of different surgical procedures and different doctors in historical surgical data, wherein the procedure-doctor-component consumption association model is used to record the usage probability, usage quantity and model preference of each doctor for various components under different surgical procedures, so as to identify the differences between the medical consumable usage preferences of different doctors under the same surgical procedure and the best practices; the efficiency index correction unit is communicatively connected to the efficiency index calculation unit and the association model construction unit, and is used to perform doctor preference correction and procedure difference correction on the initial component-level usage efficiency index based on the procedure-doctor-component consumption association model to obtain the corrected component-level usage efficiency index, wherein the doctor preference correction is used to distinguish the difference between doctors' personal habitual use and clinically necessary use, and the procedure difference correction is used to distinguish the differences between different surgical complexities and different surgical procedure subtypes.

[0036] The initial component-level usage efficiency index mentioned above is calculated directly based on the component usage status data of the current single surgery, according to the aforementioned definition and calculation formula. This initial index reflects the actual usage efficiency of the kit configuration in this surgery, but it does not exclude the influence of interference factors such as individual doctor habits or differences in surgical complexity. For example, under the same surgical procedure, one doctor may be accustomed to using a certain type of screw, while another doctor may be accustomed to using a different type; some surgeries may have significantly higher surgical complexity than the average level of the procedure due to individual patient differences. These factors may all cause the initial component-level usage efficiency index to deviate in reflecting the rationality of the kit configuration itself. To solve the aforementioned problems, it is necessary to use the aforementioned correlation model construction unit and the efficiency index correction unit, etc. For example, for the target procedure of anterior cervical spine fixation and fusion, the procedure-doctor-component consumption association model can record that Doctor A has a 90% probability of choosing a type A fixation plate and uses an average of 6.2 screws, while Doctor B has a 75% probability of choosing a type B fixation plate and uses an average of 5.8 screws. However, the clinical best practice for this procedure recommends using 5 to 6 screws. Therefore, Doctor A's screw usage exceeds the recommended range on average, potentially indicating overuse. This difference can be systematically identified by the association model and used for subsequent efficiency index correction. The doctor preference correction is used to distinguish between doctors' personal habitual use and clinically necessary use. If a doctor habitually uses more consumables than the market standard, not due to clinical necessity but out of personal operating habits, this excess consumable should not be considered "reasonable use" and should be identified and marked when calculating component utilization and waste index. Conversely, if a doctor consistently uses a specific type of consumable and achieves excellent clinical results, this preference may constitute a new best practice, and the probability values ​​and preference records in the procedure-doctor-component consumption association model should be updated promptly. The aforementioned surgical procedure difference correction is used to distinguish between differences in surgical complexity and different surgical procedure subtypes. When a surgery uses more consumables than average due to individual patient differences (such as severe osteoporosis requiring more fixation devices), this difference should not be simply judged as "waste" or "redundant configuration," but should be identified and marked during the correction process. Therefore, the corrected component-level usage efficiency index can more objectively reflect the rationality of the kit configuration itself, providing more accurate driving signals for the subsequent AI dynamic optimization module.

[0037] The component configuration AI optimization module is communicatively connected to the component usage efficiency analysis module. It takes component-level usage efficiency indicators as state input, component configuration schemes as action outputs, and a comprehensive score based on usage efficiency feedback results or usage efficiency and low-carbon feedback results as a reward signal. Through continuous iterative learning based on reinforcement learning algorithms, it generates and dynamically optimizes the component configuration scheme of the medical consumable kit corresponding to the target surgical procedure. In one embodiment, the core working principle of the component configuration AI optimization module includes the following steps.

[0038] First, the component-level utilization efficiency metrics are used as the state input of the reinforcement learning model. The state input is a multi-dimensional vector, whose dimensions include at least efficiency metrics such as component utilization, surgical procedure matching degree, configuration redundancy, and / or waste index. In an extended embodiment, the state input also includes auxiliary features such as surgical complexity scores from historical surgical data, individual patient characteristics (e.g., age, body mass index, and underlying diseases), and seasonal or regional disease trends, to enhance the model's adaptability to changes in clinical scenarios. The aforementioned state input vector comprehensively describes the utilization efficiency of the current package configuration scheme under a specific surgical procedure, providing the AI ​​model (i.e., the reinforcement learning model) with an informational foundation for configuration optimization.

[0039] Secondly, the component configuration scheme is used as the action output of the reinforcement learning model. This action output is the kit component configuration scheme generated by the model for the target surgical procedure—specifically defining the types of components that should be included in the kit, the specifications of each component, and the configuration quantity. In a concrete implementation, the action output is a configuration vector, where each element corresponds to an optional component type, and the element value is the recommended configuration quantity of that component (a non-negative integer). The model searches for the optimal configuration scheme in the combination space of all optional components—for example, for total knee arthroplasty, optional components include bone cement, pulse irrigation device, sutures, surgical gloves, electrosurgical pen, suction tube, sterile dressing, incision protective sleeve, and drainage tube, among dozens of others. The AI ​​model needs to search for the optimal configuration scheme that meets clinical needs and maximizes efficiency within a high-dimensional discrete combination space composed of the optional quantities and specifications of each component.

[0040] Furthermore, the comprehensive score is used as a reward signal for the reinforcement learning model. This comprehensive score is derived from usage efficiency feedback results or a combination of usage efficiency and low-carbon feedback results, and is used to evaluate the merits of the current configuration scheme and drive the model to iterate towards a better direction. In one embodiment, the comprehensive score is calculated based on a reward function that includes a usage efficiency reward item, a low-carbon reward item, and a completeness reward item. The value of the usage efficiency reward item is positively correlated with the component-level usage efficiency index from the efficiency feedback results (higher efficiency, larger reward); the value of the low-carbon reward item is negatively correlated with the estimated carbon emissions from the low-carbon feedback results and the component configuration scheme (lower carbon emissions, larger reward); and the value of the completeness reward item is positively correlated with the clinical completeness verification results of the component configuration scheme meeting surgical requirements (higher completeness, larger reward). In a preferred embodiment, the three reward items can be combined into a comprehensive score through a weighted summation, with each weight coefficient configured according to the actual management priorities of the medical institution—for example, if a hospital currently focuses on cost control and environmental protection, the weight of the low-carbon reward item can be appropriately increased; if a hospital currently focuses on clinical quality assurance, the weight of the completeness reward item can be appropriately increased. Furthermore, in another embodiment, the reward function may further include a cost reward item (negatively correlated with the total cost of consumable procurement) and a waste reward item (negatively correlated with the amount of waste generated) to support more comprehensive multi-objective optimization.

[0041] In one embodiment, preferably, the estimated carbon emissions are calculated as follows: First, the type, quantity, and material of each component in the component configuration scheme of the medical consumable kit are obtained, as well as at least one of the following for each component: supplier, transportation distance, and sterilization method; then, based on the obtained results, the full-cycle carbon emissions corresponding to the component configuration scheme are estimated based on a preset life cycle assessment model, and this is used as the estimated carbon emissions. In detail, the parameters used in the aforementioned estimation process have clear carbon emission implications: the type and material of the component determine its "implicit carbon emissions"—that is, the carbon emissions generated during the production process from raw material mining and processing to finished product delivery; for example, stainless steel orthopedic implants and their counterparts made of titanium alloy have significantly different carbon emission intensities during production—titanium alloys have higher melting temperatures and are more difficult to process, resulting in a higher carbon emission factor per unit weight compared to stainless steel; furthermore, disposable surgical drapes and reusable surgical drapes have completely different carbon emission distributions throughout their entire lifecycle—the former is mainly concentrated in upstream production and end-of-life disposal stages, while the latter is mainly concentrated in repeated sterilization stages; the number of components directly affects the total amount of the aforementioned carbon emissions—before the carbon emission factor of a single component is known... It should be noted that the more components are configured, the higher the total carbon emissions of the package. Supplier information is used to differentiate the carbon emission differences of similar components produced by different suppliers in terms of manufacturing processes, energy structure, and logistics routes. Different suppliers may use different production processes (such as using clean energy or traditional energy) and / or be located in different geographical locations (affecting transportation distance and mode of transportation), all of which affect the carbon emission factor of the component. Transportation distance is used to quantify the logistics carbon emissions generated during the transportation of components from the supplier's warehouse to the hospital's warehouse. The longer the transportation distance, the higher the transportation carbon emissions per unit component. Disinfection method is used to quantify the operational carbon emissions generated during the disinfection and sterilization process of components in the hospital. Different disinfection methods (such as high-temperature steam sterilization, ethylene oxide sterilization, or low-temperature plasma sterilization with hydrogen peroxide, etc.) have different energy consumption intensities and carbon emission characteristics. In one embodiment, the aforementioned parameters can be routinely obtained from the consumable master data of the hospital's SPD system, the environmental product declaration or carbon footprint report provided by the supplier, the transportation records of the hospital's logistics management system, and the disinfection records of the disinfection supply center.

[0042] In one specific implementation, the life cycle assessment model adopts a process-based LCA (Life Cycle Assessment) methodology, dividing the total life cycle carbon emissions corresponding to the component configuration scheme into four stages: production, transportation, use (mainly in-hospital disinfection and sterilization), and disposal. The carbon emissions of each stage are calculated separately and summed to obtain the total life cycle carbon emissions. Taking the kit corresponding to total knee replacement surgery as an example, the kit configuration includes multiple components such as bone cement (1 piece, PMMA material), pulse irrigation device (1 piece, PVC material), sutures (2 pieces, silk material), and surgical gloves (4 pairs, natural rubber material). For bone cement, its carbon emissions are mainly concentrated in the upstream production stage (the polymerization process of PMMA has high energy consumption). For surgical gloves, its carbon emissions are distributed in the natural rubber planting and processing stage (related to the carbon emission absorption of the plantation and processing energy consumption), the transportation stage (related to the distance between the supplier's production site and the hospital), and the disposal stage (the biodegradation process of natural rubber produces methane gas). After obtaining the carbon emission factor corresponding to each component, the system multiplies it by the configuration quantity of each component in the package, and sums them to obtain the total carbon emissions during the production stage; it calculates the total carbon emissions during the transportation stage based on the supplier location and transportation method of each component; it calculates the total carbon emissions during the usage stage based on the disinfection method of each component; and it calculates the total carbon emissions during the waste disposal stage based on the material of each component. The sum of these four stages is the estimated total carbon emissions for the entire lifecycle corresponding to the component configuration scheme. This estimated value serves as the input for calculating the low-carbon reward item—the lower the carbon emissions, the higher the value of the low-carbon reward item, and the greater the incentive for the AI ​​model to generate low-carbon configuration schemes.

[0043] In a further preferred embodiment, to achieve continuous improvement in carbon emission management, the component configuration AI optimization module is further used to calculate the actual carbon emissions based on the actually used components and unused returned components after each actual surgery of the target procedure, and to feed back the deviation between the actual carbon emissions and the estimated carbon emissions to the life cycle assessment model and the reward function, so as to continuously improve the accuracy of carbon emission estimation and the accuracy of low-carbon decision-making. Specifically, the workflow of the aforementioned carbon trace feedback mechanism is as follows: After each actual surgery, the system obtains the actual component consumption data for that surgery from the component usage status perception module—including which components were fully used, which components were partially used (the actual consumption ratio needs to be recorded), and which components were not used and were returned (the number of returned components needs to be recorded). Based on the aforementioned actual consumption data and the carbon emission factors of each component, the actual carbon emissions of this surgery are calculated. For example, if the preoperative estimated total carbon emissions of the kit are 25.6 kg CO2 equivalent (based on the assumption that all components in the configuration are used), and the actual components consumed postoperatively are only 80% of the configuration (the remaining 20% ​​of components were not used and were returned), then the actual carbon emissions are approximately 20.5 kg CO2 equivalent (considering that returned components can be used in other surgeries, their carbon emissions are not entirely included in this surgery). The aforementioned actual carbon emissions are compared with the carbon emissions estimated preoperatively based on the LCA model to obtain the deviation value—if the preoperative estimate is 25.6 kg CO2 equivalent and the postoperative actual value is 20.5 kg CO2 equivalent, then the deviation value is -5.1 kg CO2 equivalent (negative deviation, actual is lower than estimated), indicating that the carbon emissions of this surgery are lower than expected; conversely, if the actual value is higher than the estimated value, it indicates that the preoperative estimate was conservative and the estimation coefficients of relevant factors need to be increased. The aforementioned deviation values ​​are fed back through a dual-channel feedback mechanism, affecting two stages simultaneously: First, they are fed back to the life cycle assessment model to correct the carbon emission factor estimation parameters of each component, making the estimation model more closely reflect the hospital's actual operating conditions. For example, if a hospital's remote location results in a significantly longer transportation distance than the industry average, leading to consistently higher actual transportation carbon emissions than estimated values, the system will automatically increase the corresponding transportation carbon emission factor parameters after accumulating multiple rounds of surgeries. Second, they are fed back to the reinforcement learning reward function, using the deviation values ​​to adjust the calculation benchmark for low-carbon reward items, making the reward signal more accurately reflect the true carbon emission reduction effect. If the actual carbon emission factor of a component is significantly higher than the preset value of the LCA model, the reinforcement learning model will be more inclined to reduce the number of configurations of that component or recommend alternative components in subsequent optimizations. Through the continuous iteration of multiple rounds of "preoperative estimation → postoperative calculation → deviation feedback → model update," the accuracy of the system's carbon emission estimation gradually improves, and the accuracy and effectiveness of low-carbon decisions are also enhanced.For example, after 50 similar surgeries, the estimation bias of the LCA model converged from ±15% initially to within ±3%, and the accuracy of the reinforcement learning model in recommending low-carbon components also significantly improved. This carbon footprint feedback mechanism enables the system not only to predict and optimize carbon emissions before surgery, but also to continuously improve its carbon emission management capabilities through feedback from actual data after surgery, truly achieving green and sustainable management of medical consumables.

[0044] In a specific training process, the AI ​​model undergoes offline pre-training based on multiple rounds of surgical records accumulated from historical surgical data (each round's record includes actual usage data, calculated efficiency indicators, and corresponding configuration schemes) to obtain an initial strategy. During the pre-training phase, the AI ​​model continuously attempts to generate different configuration schemes and judges the merits of the schemes based on reward signals, gradually learning which configuration scheme can achieve the optimal balance between efficiency, low carbon footprint, and completeness. Subsequently, the AI ​​model enters the online learning phase—after each actual surgery is completed, the system automatically acquires the actual usage data of that surgery, calculates the actual efficiency indicators, generates actual reward signals, and stores the experience of that round (state, action, reward, next state) in an experience replay buffer for the AI ​​model to continuously perform incremental learning and strategy improvement. As training data accumulates, the configuration schemes generated by the AI ​​model gradually converge to the optimal scheme and can adaptively adjust according to changes in the hospital's actual operating environment (such as the implementation of new surgical procedures, supplier changes, or updates to consumable products).

[0045] Through continuous iterative learning based on reinforcement learning algorithms, the component configuration AI optimization module can generate and dynamically optimize the component configuration scheme of the medical consumable kit corresponding to the target surgical procedure. The resulting optimal component configuration scheme is not only a one-time static optimal solution, but also a dynamic scheme that can continuously evolve with the accumulation of clinical practice data. For example, for a certain target surgical procedure, the system may initially generate a relatively conservative configuration scheme (with a large number and variety of components to ensure clinical completeness). With the accumulation of multiple rounds of surgical data and the iterative optimization of the AI ​​model, the system gradually identifies which components are frequently used "essential components," which components are only needed in specific situations as "backup components," and which components are long-term unused "redundant components." Then, "backup components" are removed from the essential kit and assigned to the backup kit, and "redundant components" are completely removed from the configuration scheme. Finally, an optimized configuration scheme is generated that has a simplified range of component types, accurate configuration quantities, and can both ensure clinical needs and maximize efficiency and minimize waste. In medical institutions with a large number of surgeries, this optimization process can be carried out continuously, constantly approaching the global optimal solution. When surgical guidelines are updated, new consumable products are launched, or the doctor team changes, the system can automatically capture changes in usage patterns and adjust the configuration accordingly, maintaining the long-term sustainability of the optimization effect.

[0046] In a preferred embodiment, the component configuration AI optimization module employs a multi-objective reinforcement learning framework to simultaneously optimize at least two objectives. It dynamically balances these objectives using a Pareto front approach or a weighted scalarization approach. These objectives include, but are not limited to, maximizing component utilization efficiency, minimizing carbon emissions, maximizing clinical completeness, minimizing consumable costs, and minimizing waste generation. When conflicts exist between objectives (e.g., further reducing carbon emissions might require selecting more expensive, environmentally friendly components, conflicting with the cost minimization objective), the Pareto front approach provides a set of non-dominated solutions in the objective space, allowing managers to choose based on the hospital's current needs and policy direction. When the hospital faces pressure to control medical insurance costs, it can prioritize Pareto solutions with better cost indicators; when facing carbon emission reduction assessment requirements, it can prioritize Pareto solutions with better carbon emission indicators. The weighted scalarization approach, through preset weight coefficients, comprehensively weighs the relative importance of each objective and automatically selects the optimal compromise. In one specific embodiment, when there is a conflict between minimizing carbon emissions and maximizing clinical suitability, the system prioritizes satisfying the clinical suitability constraint—that is, suitability must reach a preset minimum acceptable threshold (e.g., 85% clinical need coverage). Within this feasible domain, the system seeks the optimal solution for minimizing carbon emissions. This priority constraint mechanism ensures that patient safety and surgical quality are not affected by low-carbon optimization.

[0047] In a specific implementation, the reinforcement learning algorithm can be any of the mainstream deep reinforcement learning algorithms, such as Proximal Policy Optimization (PPO) or Deep Q-Network (DQN). The PPO algorithm introduces a trust region constraint for policy updates, limiting the magnitude of each policy update and ensuring that the generated configuration scheme does not fluctuate drastically due to excessively large single update steps—for example, it prevents excessive reduction of the configuration quantity of a component due to an abnormally high waste rate in a particular surgery, thus avoiding the clinical risk of component shortage in subsequent surgeries. The DQN algorithm, on the other hand, breaks the temporal correlation between consecutive surgical data through an experience replay mechanism, improving the utilization efficiency of training data and the stability of model training. Furthermore, in an extended embodiment, the component configuration AI optimization module can also employ a model-based reinforcement learning method—by constructing a component consumption prediction model for the target surgical procedure (used to predict expected efficiency indicators and carbon emissions under a given configuration scheme), the AI ​​model can perform a large number of "hypothesis" deductions in a virtual environment (i.e., explore the effects of massive configuration schemes without waiting for real surgery), significantly accelerating the convergence speed of the strategy and reducing dependence on real surgical data. The above algorithm selection and implementation methods are conventional techniques for those skilled in the art and will not be elaborated further here.

[0048] In one specific implementation, the system further includes an intraoperative real-time monitoring module that is communicatively connected to the component usage status perception module. The intraoperative real-time monitoring module is used to identify intraoperative component usage behavior in real time through computer vision technology during the operation. When at least one wasteful behavior of consumable components is detected in real time based on the identification results, the module proactively sends a reminder message to the operating room terminal or medical staff mobile terminal. The at least one wasteful behavior of consumable components includes abnormal selection of consumable component specifications, improper use, and / or unnecessary unpacking. The reminder message includes a description of the identified wasteful behavior, component information of the wasted component, and / or suggested alternative operating procedures. Unlike the component usage status perception module, which collects data twice—once before and once after surgery—the intraoperative real-time monitoring module runs continuously throughout the operation. Using computer vision technology, it analyzes the usage scenarios in real time, extending the window for identifying and intervening in component waste from "pre- and post-operative" to "intraoperative real-time." This overcomes the limitation of post-operative analysis, which can only summarize patterns but cannot immediately prevent waste. Post-operative efficiency analysis tells managers "which components are consistently wasteful and how to adjust their configuration next time," while intraoperative real-time monitoring immediately alerts medical staff to "ongoing wasteful behavior" during the current surgery. Together, they constitute a waste identification and intervention system covering the entire surgical process.

[0049] To achieve the above functions, in one embodiment, the intraoperative real-time monitoring module may include at least one set of high-definition camera devices deployed in the operating room—such as industrial cameras or operating room-specific cameras mounted on the surgical lamp arm, pendant, or operating room ceiling—for real-time acquisition of video stream data from areas such as the surgical back desk area, the instrument nurse's workstation area, and / or the consumables storage area. The selection of camera deployment locations should ensure clear capture of the retrieval, unpacking, and use of consumable components, while avoiding obstruction or interference with the surgical operation area and the normal activities of medical staff. The acquired video stream data is transmitted in real-time via a local area network or 5G network to edge computing devices (such as GPU servers or embedded AI computing platforms) deployed locally in the operating room or in the hospital data center. Real-time inference is performed by a pre-trained computer vision model deployed on the edge computing device—the edge computing architecture ensures that all video data is processed locally, meeting operating room data security compliance requirements while controlling inference latency to the millisecond level, achieving true "real-time" recognition.

[0050] In one embodiment, the computer vision model can be built based on mainstream deep learning object detection and scene understanding architectures such as YOLO (short for "You Only Look Once," a real-time object detection system based on deep learning that transforms object detection into a single regression problem, directly predicting bounding boxes and class probabilities through a single neural network) or Vision Transformer. It is pre-trained using a large amount of labeled operating room scene video data to form an operating room scene understanding model. By analyzing real-time video streams, the aforementioned model can automatically detect and classify intraoperative behaviors.

[0051] The specific identification methods for the above three wasteful behaviors are shown in (a) to (c).

[0052] (a) Abnormal specification selection refers to a mismatch between the specifications of consumable components selected by medical staff and the patient's individual characteristics or surgical plan, which may lead to the need for subsequent replacement and additional consumption of the same type of component, resulting in avoidable waste. In one embodiment, the system compares in real-time the specification information of the components being used (such as the diameter and length of catheters, the diameter and length of screws, the diameter and needle type of sutures, etc.) with a recommended specification list automatically generated preoperatively based on the patient's imaging data and surgical plan. When a mismatch is detected between the used specification and the recommended specification, it is determined to be an abnormal specification selection. For example, for a total knee replacement surgery, the system predicts that a type A femoral prosthesis should be used preoperatively based on the patient's CT imaging data. If the system detects during the operation that the scrub nurse has removed a type B femoral prosthesis and is preparing to unpack it, the system determines that the specification selection is abnormal and proactively issues a reminder. In addition, in an extended embodiment, the real-time acquisition of specification information can also be assisted by extracting the specification identification characters on the outer packaging of the consumables for OCR text recognition.

[0053] (b) Improper use refers to the use of consumable components in a manner that does not conform to the device operating procedures or recognized clinical best practices, which may lead to damage, failure, or failure to achieve the intended treatment purpose, thus requiring additional component replacement and resulting in unintended waste. In one embodiment, improper use is identified by comparing the sequence of device operation actions detected in real-time video with predefined standard operating procedures—if it is detected that a healthcare worker operates a component in a manner exceeding the device's design limits (e.g., screwing in a screw beyond the maximum torque value specified in the product manual, which may cause the screw to break, or using a disposable component in a non-designed scenario, resulting in its damage), it is determined to be improper use. For example, when the system detects that a physician uses a disposable electrosurgical pen to perform a prolonged high-power operation on an unintended tissue type (which may cause the electrosurgical pen to overheat and fail), and this operation deviates from the recommended usage parameters range stated in the electrosurgical pen's product manual, the system determines it to be improper use; similarly, when the system detects that an instrument nurse passes sterile consumables in a manner that may cause component contamination (e.g., sterile gloves touching non-sterile areas), causing the component to need to be discarded and replaced, this also falls under the category of improper use.

[0054] (c) Unnecessary unpacking refers to the premature unpacking of consumable components before the actual clinical need has been confirmed. This results in the components being unable to be returned or re-sterilized once the sterilization barrier has been breached, leading to avoidable waste. In one embodiment, the system identifies unnecessary unpacking by tracking the surgical progress in real time to determine the type of components required for the current surgical stage. This is achieved by combining surgical step recognition technology with temporal matching of unpacking behavior detected in the video with the current surgical stage. If a component is not typically used in the current surgical stage (e.g., during the initial anatomical exposure stage of the surgery, the scrub nurse unpacks an implant component that is usually only needed in the fixation and fusion stage), and there is no indication that the surgery is about to enter a subsequent stage that requires the use of the component, it is determined to be unnecessary unpacking. Regarding surgical step recognition, in one embodiment, the intraoperative real-time video can be temporally correlated with the surgical steps based on a surgical workflow recognition model (using a spatiotemporal graph convolutional network or Transformer architecture to identify the specific stage and steps of the surgery from the video stream) to obtain information about the current surgical stage. For example, during a laparoscopic surgery, the nurse removed the titanium clip assembly during the pneumoperitoneum establishment phase. However, according to the standard surgical procedure, the titanium clip is only needed during the tissue mobilization phase, which involves multiple steps such as trocar placement and abdominal exploration. Removing the clip too far in advance may result in contamination of the titanium clip during its placement on the operating table or it may no longer be needed due to changes in the intraoperative situation. The system judges this behavior as unnecessary removal of the clip.

[0055] In one embodiment, the judgment of the above three wasteful behaviors can be performed in combination: for a single use of consumables, the system can simultaneously detect whether the specifications are correct, whether the operation method is standardized, and whether the timing of opening the package is reasonable; if any one or more of these triggers an anomaly judgment, the system will generate a reminder message. This triple detection mechanism not only ensures comprehensive coverage of wasteful behaviors but also provides medical staff with real-time guidance from different dimensions.

[0056] In one embodiment, the reminder information is generated as follows: When wasteful behavior is detected, the system first extracts the type tag of the behavior (abnormal specification selection, improper use, or unnecessary unpacking), and automatically generates corresponding wasteful behavior description text based on the type tag; secondly, it retrieves detailed component information of the identified component from the consumable master database or SPD system, including component name, specification model, batch number, and storage location, to help medical staff quickly locate and confirm the component; thirdly, it generates suggested alternative operation schemes based on a preset recommendation rule base—for example, if abnormal specification selection is detected, the alternative scheme suggestion may include recommending the selection of the correct specification component and its storage location in the consumable cabinet; if improper use is detected, the alternative scheme suggestion may include the correct operating steps or parameter setting guidance for the device, such as "It is recommended to reduce the power of the electrosurgical pen to the recommended power range stated in the product manual to avoid overheating damage"; if unnecessary unpacking is detected, the alternative scheme suggestion may include the feasibility and operation procedure for emergency re-sterilization of the component, or remind the user to mark the component as unpacked and include it in postoperative special traceability management for subsequent loss cause analysis and supplier settlement adjustment. The reminder information is pushed through devices such as wall-mounted displays, anesthesia information terminals, or mobile PDAs (Personal Digital Assistants, also known as handheld computers) used by medical staff in the operating room. The push methods may include screen pop-ups, voice broadcasts, or vibration reminders to adapt to the information reception needs of different operating room scenarios.

[0057] Furthermore, in a preferred embodiment, the identification results of the intraoperative real-time monitoring module are also fed back to the component usage efficiency analysis module for synchronous storage—the waste behavior data identified during the operation serves as a supplementary dimension to the usage status data and can indirectly affect the calculation of component-level usage efficiency indicators. For example, if an unnecessary unpacking occurs during a surgery, resulting in a component being judged as "discarded without use," then that component will be included in the numerator of the waste index calculation. In addition, the doctor's personal operating habits identified during the operation (such as a doctor frequently unpacking specific components prematurely during a particular surgical phase) can be transmitted to the efficiency indicator correction unit to update the preference records in the procedure-doctor-component consumption association model. Through the above data synchronization mechanism, the intraoperative real-time monitoring module not only immediately prevents waste behavior but also provides richer intraoperative process data for postoperative efficiency analysis and continuous AI optimization, forming a dual closed-loop management system of "real-time feedback + offline optimization." The closed-loop design of the above data link further strengthens the system's intelligent management capabilities throughout the entire process, from perception to decision-making, and from intraoperative to postoperative. This data loop design is not taught in existing technologies—existing intraoperative monitoring solutions (such as operating room black box systems or surgical instrument handover behavior detection) focus on recording surgical safety events and checking surgical procedure compliance. Their detection results are only used for postoperative quality review and performance evaluation, and there is no data feedback link between them and the consumable configuration optimization system.

[0058] To further enhance the system's adaptability and long-term stability, in one embodiment, the computer vision model used by the intraoperative real-time monitoring module supports an online learning mechanism—using recognition results confirmed by manual feedback during actual surgery as incremental training samples to fine-tune and update the model. Specifically, when medical staff receive an alert, they can provide one-click feedback on the accuracy of the alert via an operating room terminal or mobile device (e.g., "confirm acceptance," "false alarm," or "ignore"). The system automatically associates and stores this feedback information with the corresponding video clip for periodic iterative training of the model. In this way, the recognition accuracy of the computer vision model can continuously improve with the accumulation of surgical data, while also adapting to differences in instrument brands, consumable packaging types, and surgical procedures among different hospitals, continuously improving monitoring accuracy. At the clinical environment deployment level, considering the differences in operating room space layout, equipment configuration, and informatization levels among different medical institutions, the intraoperative real-time monitoring module supports integration with the hospital's existing video surveillance system. It can fully utilize existing camera hardware and achieve functional upgrades through software-level algorithm deployment, reducing construction costs and deployment barriers. The above-mentioned scenario-adaptive design makes the intraoperative real-time monitoring module highly scalable and meets the actual needs of medical institutions at all levels for the construction of smart operating rooms.

[0059] In one specific implementation, a configuration report automatic generation module is also included, which is communicatively connected to both the component configuration AI optimization module and the component usage efficiency analysis module. This module invokes a large language model, combining the optimization results of the component configuration scheme, component-level usage efficiency indicators, surgical procedure clinical guidelines, and physician operational preference data. Using a multi-round inference strategy based on a preset Prompt template, it sequentially generates efficiency comparison analysis paragraphs, optimization suggestion analysis paragraphs, and risk warning paragraphs, ultimately synthesizing a structured medical consumable package configuration optimization suggestion report for the target surgical procedure. In clinical practice, both the optimal component configuration scheme generated by the component configuration AI optimization module and the component-level usage efficiency indicators output by the usage efficiency analysis module are structured data. While accurate, their readability is limited. Hospital management and clinical department directors often require a natural language report that comprehensively presents the efficiency analysis results, explains the AI ​​optimization logic, and highlights potential risks to quickly understand the current consumable management efficiency and make appropriate decisions. The configuration report automatic generation module is designed to address this need—it establishes an automated conversion bridge between the two types of structured data and the natural language decision support required by managers.

[0060] In one embodiment, the large language model can be any of the mainstream open-source or commercial large language models such as the GPT series, Llama series, DeepSeek series, or ChatGLM series, and integrated into the system through local deployment or API (Application Programming Interface) calls. For medical institutions with high data security requirements, a local deployment solution is preferred, deploying the model on the hospital's GPU (Graphics Processing Unit) server or private cloud environment. All input data (including component configuration schemes, efficiency indicators, doctor operation preferences, etc.) is inferred within the hospital network and is not transmitted to external public networks, meeting medical data privacy protection and compliance requirements. For scenarios using commercial APIs, the input data can be anonymized before the call (e.g., removing doctor names and patient personal identification information), retaining only the anonymized data required for statistical analysis. At the model selection level, considering that the generation of package configuration optimization suggestion reports involves professional terminology in the field of medical consumables management (such as component names, surgical procedure names, and specifications) and specific reasoning logic (such as correlation analysis between efficiency indicators and priority ranking of optimization suggestions), in a preferred embodiment, a large language model fine-tuned with medical domain corpus can be selected (such as using SPD management specification text, consumable catalog data, clinical pathway guidelines, and other medical consumables management professional corpus for low-rank adaptation fine-tuning based on publicly available general language models) to improve the model's accuracy in understanding professional terminology in the field of consumables management—for example, enabling the model to correctly distinguish whether the "screw" component under different surgical procedures specifically refers to orthopedic screws or maxillofacial surgery screws, avoiding deviation of the analysis conclusions in the report from the actual clinical scenario due to ambiguity in terminology.

[0061] The Prompt template is a key technical means to guide the large language model to generate reports on demand. In one embodiment, the Prompt template adopts a hierarchical structure design, including three levels: system-level instructions, task-level instructions, and constraint-level instructions. The system-level instructions are used to set the role and behavioral boundaries of the large language model—for example, setting the model role as a "medical consumables management and analysis expert," specifying that its output language is Chinese, the output format is structured Markdown text, and prohibiting the fabrication of statistical figures or clinical conclusions not appearing in the input data, thus constraining the compliance and professional direction of the model's generation behavior from the top level. The task-level instructions are used to describe the specific tasks that need to be completed—that is, generating efficiency based on the optimization results of the input component configuration scheme, component-level usage efficiency indicators, surgical procedure clinical guidelines, and physician operation preference data. The report is divided into three sections: comparative analysis, optimization suggestion analysis, and risk warning, specifying the core content and output format of each section. The constraint layer instructions are used to embed specific domain knowledge rules, including a procedure-component adaptation rule base (which specifies the minimum configuration requirements, acceptable specification range, and mandatory compatibility relationships between components under a specific procedure), infection control compliance inspection rules (such as sterilization packaging integrity standards, a list of consumables that cannot be returned after opening, and mandatory disposal requirements for single-use components), and report quality standard inspection rules (such as the word count range of each section, the format of index data citation, and the feasibility verification standards for optimization suggestions). This ensures that the large language model automatically follows the above domain rules when generating the report, guaranteeing the professionalism and compliance of the report content.

[0062] The multi-round reasoning is the core technical mechanism for successfully generating high-quality reports in segments. Its principle is to break down a complex report generation task into multiple steps, with each round focusing on a specific sub-task and using the output of the previous round as the context for the next, progressively deepening the analysis logic. In one embodiment, the multi-round reasoning strategy specifically includes three rounds of reasoning, corresponding to the generation of the following three paragraphs.

[0063] The first round of inference generates an efficiency comparison analysis section: This section uses component-level utilization efficiency indicators as core data to compare and present the quantitative differences between the current package configuration and historical configurations (or historical best practice benchmarks based on large-scale statistical analysis). In one embodiment, this round of Prompt-guided model compares the following dimensions: the difference between the component utilization rate of the current configuration and the historical average (such as the average utilization rate of similar surgeries in the previous three months), the trend of configuration redundancy (such as whether it shows a monthly downward trend, indicating that previous optimization measures have been effective), and the gap between the waste index and the industry benchmark (such as the waste level of this procedure in similar hospitals). This section is mainly declarative analysis, aiming to answer "At what level is the current package utilization efficiency, and how does it perform compared to the past and compared to industry benchmarks?"

[0064] The second round of reasoning generates optimization suggestions analysis paragraph: Based on the results of efficiency comparison analysis, combined with the optimal component configuration scheme output by the component configuration AI optimization module and the clinical guidelines for surgical procedures, this paragraph proposes specific and actionable package component configuration optimization suggestions. In one embodiment, this round of Prompt analysis unfolds from the following dimensions: Recommending which components to add (components currently missing from the kit but recommended for addition based on the optimal configuration, with explanations of the reasons for addition – whether based on deficiencies identified through efficiency analysis, requirements of updated surgical guidelines, or to replace existing poorly performing components); Recommending which components to remove (redundant components present in the kit but consistently unused in multiple surgeries, with usage data for the component over the past three months provided as the basis for removal; or removing corresponding components based on deletions in the latest version of the surgical clinical guidelines); Recommending which components to replace (components present in the kit but whose model, specifications, or supplier need adjustment – ​​for example, replacing a component from supplier A with the same type from supplier B, with explanations of the reasons for replacement, such as better cost-effectiveness, lower carbon emission factors, or more reliable supply stability); Recommending which component configuration quantities to adjust (components present in the kit but with excessive or insufficient configuration quantities, providing suggested adjustment quantities, and offering the basis for adjustment based on correlation model analysis). The above optimization suggestions are sorted in order of priority from high to low. The priority sorting takes into account the potential for efficiency improvement (the expected increase in component utilization or reduction in waste ratio), implementation difficulty (whether the adjustment involves supplier changes or only requires adjustment of internal packaging processes) and risk level (whether the adjustment may affect clinical kit availability). This allows managers to select the optimization measures to be implemented in the current stage based on priority.

[0065] The third round of reasoning generates a risk warning paragraph: This paragraph identifies and warns of potential risks that may be faced when implementing the above optimization suggestions, ensuring that management decisions are made on the premise of fully understanding the risks, and avoiding the neglect of clinical safety and management compliance due to blindly pursuing efficiency improvement or cost reduction. In one embodiment, this round of Prompt-guided model identifies risks from the following dimensions: Completeness risk – the probability of a component shortage in specific patient groups or surgical complication scenarios after removing or reducing the configuration quantity of a component, and contingency plans (e.g., keeping the removed component as spare inventory for use in a few cases); Supply chain risk – the risk of supply chain disruption such as extended delivery cycles, unstable supply, or changes in after-sales service levels if supplier A is changed to B, and provides risk mitigation suggestions (e.g., maintaining a dual-supplier strategy to ensure supply flexibility, or setting a transition period and inventory buffer before switching suppliers); Clinical acceptance risk – the potential for physician resistance and implementation obstacles if the adjusted configuration conflict with some doctors' long-term operating habits, and provides suggestions for gradual promotion and targeted communication; Compliance risk – whether the adjusted configuration is inconsistent with the latest version of surgical clinical guidelines, medical insurance payment policies, UDI (Unique Device Identification) management regulations, or sterilization standards, and provides a compliance checklist. Each risk item includes a risk description (clearly defining the specific content and causes of the risk), a risk level (assessed as high / medium / low), triggering conditions (clearly defining under what circumstances the risk will actually occur), and recommended countermeasures (specific mitigation plans or alternative paths for the risk), forming a complete risk management closed loop, enabling managers to monitor and control various risks in a targeted manner while promoting optimization.

[0066] The three rounds of reasoning described above do not operate independently, but rather form a progressive analytical chain through contextual connection. The efficiency comparison analysis in the first round provides data support and a basis for prioritizing the optimization suggestions in the second round. The specific content of the suggestions in the second round provides the scope of identification and assessment objects for the risk analysis in the third round. For example, if the second round suggests removing a component, the third round will specifically conduct a completeness risk assessment for the removal of that component; if the second round suggests changing suppliers, the third round will specifically examine the supply chain stability of that supplier.

[0067] After the above three sections are generated, the automatic configuration report generation module merges the generated results of the three sections in a fixed order: efficiency comparison analysis section, optimization suggestion analysis section, and risk warning section. It also adds a report title (including the target procedure name, report generation date, and report number), a report summary (summarizing the core findings and main recommendations of this analysis, within approximately 200 words), and an appendix (including a description of the data sources used in this analysis, the efficiency index calculation method, and the version information of the life cycle assessment model used). Finally, it synthesizes and generates a structured medical consumable package configuration optimization suggestion report under the target procedure.

[0068] In one embodiment, the automatic report generation module also supports regular automatic report generation and proactive push functions. The report generation frequency can be set according to hospital management needs (e.g., weekly, monthly, or quarterly). It automatically collects the latest optimization results from the component configuration AI optimization module and the latest efficiency indicators from the component usage efficiency analysis module, triggering the report generation process. The generated report is then pushed to designated managers and medical staff—such as members of the consumables management committee, the operating room head nurse, and the orthopedic department director—via email or the hospital's internal communication platform. For situations requiring urgent attention (e.g., the waste index of a certain surgical procedure exceeds a preset threshold for three consecutive weeks), the system can automatically generate a special analysis report and proactively push early warning notifications, upgrading the management model from "passively reviewing reports" to "reports proactively reaching out to people."

[0069] In practical applications, the structured reports generated by the automatic configuration report generation module can provide data support for the regular review meetings of the hospital's consumables management committee. Managers can directly present the report's charts and key indicators at these meetings, facilitating data-driven discussions and decision-making, significantly improving the efficiency and scientific rigor of management reviews. Simultaneously, the automatic configuration report generation module can also archive historical reports, constructing a time-series archive of configuration optimization for various surgical procedures over the years. This provides comprehensive data support for internal hospital audits, external certifications (such as ISO 14001 environmental management system certification), and academic research. For example, after introducing this system, a hospital continuously optimized and automatically generated reports for total knee replacement surgery packages for 12 months using AI. Managers can intuitively understand the overall improvement trend in consumable usage efficiency for this procedure by reviewing the efficiency indicator change curves of the 12 monthly reports, providing a quantitative basis for subsequent management decisions.

[0070] Furthermore, in a preferred embodiment, the automatic report generation module also supports interactive report customization—administrators can input additional areas of focus or analytical needs (such as "Please focus on changes in the usage efficiency of high-value consumables" or "Please analyze the differences between various subspecialty groups in joint surgery") through natural language input fields on the report generation interface. The system then injects this instruction as a supplementary Prompt into the existing three-layer Prompt template, re-executes local inference (updating only the affected paragraphs), and generates customized report content. This interactive customization approach allows the same generation framework to flexibly adapt to the needs of different management levels (hospital-level administrators focus on cost control and carbon reduction effectiveness, while clinical department administrators focus on surgical procedure matching and differences in professional subdivisions), achieving a balance between standardization and personalization.

[0071] In another preferred embodiment, the automatic report generation module further includes a report quality audit unit, used to automatically audit the compliance and accuracy of the output content after the report is generated. This audit unit uses a preset rule engine to check whether the statistics in the report are consistent with the input data (such as utilization rate and carbon emission values), whether the cited clinical guidelines are the latest versions, whether the word count of each paragraph is within a reasonable range (e.g., excessively short optimization suggestion paragraphs may indicate that the model has not fully expanded its analysis), and whether there are any unfounded conclusions generated by "illusion" (by automatically comparing key statements in the report with the input data to detect whether the model has fabricated facts that do not exist in the input data). This audit unit, through deep integration of report quality check rules (such as data format validation rules, fact-checking rules, and paragraph integrity check rules) defined in the Prompt constraint layer with the conventional LLM (Large Language Model) generation process, enables the system to automatically screen and mark potential problems in the report without relying on external manual review, for manager review or automatic system correction. This closed-loop mechanism of "verification before generation + auditing after generation" is particularly important in medical scenarios. Medical consumables management involves clinical safety and financial compliance. Incorrect recommendations in the report may lead to patient safety risks or economic losses. Dual control ensures the reliability of the content generated by the large language model in terms of professionalism and accuracy, and fully guarantees the quality and credibility of AI-assisted management decisions.

[0072] In one specific implementation, an intelligent sorting and automatic packaging module is also included, which is communicatively connected to the component configuration AI optimization module. This module, upon receiving the optimal component configuration scheme generated and output by the AI ​​optimization module, controls an intelligent sorting execution structure to sort out the corresponding components from the consumables library according to the optimal configuration scheme. After sorting, the module automatically packages the components into a medical consumable kit corresponding to the target surgical procedure. The intelligent sorting and automatic packaging module converts the digital scheme output by the AI ​​optimization module into a physical kit that can be directly used in the operating room. The intelligent sorting execution structure (such as a multi-axis robotic arm or a Cartesian robot) picks out components one by one from the corresponding storage location in the SPD consumables library according to the type, specifications, and quantity of each component in the scheme. After scanning and verification, the components are transferred to the packaging station, where an automatic packaging device sterilizes and seals the kit and prints a label. Thus, the system realizes a closed-loop management process that is unmanned or with minimal human intervention, from "intraoperative status perception → postoperative efficiency analysis → AI optimization decision-making → intelligent sorting and packaging". It automatically implements the AI ​​optimization results into physical execution, reduces the error rate and labor intensity of manual assembly, and enables the optimization plan to respond to changes in clinical needs in real time.

[0073] In summary, the medical consumable kit component usage efficiency analysis and AI optimization system provided in this embodiment has the following technical effects: (1) This embodiment provides a new closed-loop solution that can perform refined usage efficiency analysis on each component in the package and continuously self-optimize based on the analysis results using AI algorithms. This solution includes a component usage status perception module, a component usage efficiency analysis module, and a component configuration AI optimization module. The component usage status perception module is used to obtain usage status data for each component in the package under the target procedure. The usage status data adopts any one of the following states: fully used, partially used, and unused. The component usage efficiency analysis module is used to calculate component-level usage efficiency indicators based on the usage status data, including component utilization rate, procedure efficiency, and other performance metrics. Matching degree, configuration redundancy, and waste index; the component configuration AI optimization module is built on reinforcement learning algorithm, takes component-level usage efficiency indicators as state input, component configuration schemes as action output, and a comprehensive score based on usage efficiency feedback results or usage efficiency and low-carbon feedback results as reward signal. Through continuous iterative learning, it generates and dynamically optimizes the optimal component configuration scheme. Thus, through the collaborative work of the aforementioned modules, it can realize a refined analysis of the usage efficiency of medical consumable kit components and an AI-driven continuous optimization closed loop, providing medical institutions with an intelligent consumable management solution that combines clinical safety, operational efficiency, and low-carbon benefits. (2) The component usage status awareness module can accurately determine the full usage status, partial usage status and unused status of each component in the kit after a single operation by collecting multimodal information and using a deep learning status classification model, providing a refined data foundation for subsequent efficiency analysis; (3) The component utilization efficiency analysis module calculates component-level utilization efficiency indicators such as component utilization rate, surgical procedure matching degree, configuration redundancy and waste index based on status data. It also performs doctor preference correction and surgical procedure difference correction through the surgical procedure-doctor-component consumption correlation model, which can distinguish the difference between doctors' personal habitual use and clinically necessary use, making the efficiency evaluation results more objective and accurate. (4) The component configuration AI optimization module takes the component-level usage efficiency index as the state input, the component configuration scheme as the action output, and the comprehensive score including usage efficiency feedback results or usage efficiency and low carbon feedback results as the reward signal. Through continuous iterative learning based on reinforcement learning algorithm, it can generate and dynamically optimize the optimal component configuration scheme corresponding to the target technique, overcoming the limitations of the one-time adjustment scheme based on static rules or human experience in the existing technology. (5) By using the full-cycle carbon emissions estimated by the life cycle assessment model as the calculation input for low-carbon reward items, and by feeding back the deviation between the actual carbon emissions after surgery and the estimated carbon emissions before surgery to the life cycle assessment model and reward function through the carbon trace feedback mechanism, carbon emission factors can be incorporated into the closed-loop management of package configuration optimization, which can continuously improve the accuracy of carbon emission estimation and the accuracy of low-carbon decision-making. (6) By using a multi-objective reinforcement learning framework, multiple objectives such as component utilization efficiency, carbon emissions, clinical completeness, consumable costs and waste generation can be optimized simultaneously. This enables a scientific balance when multiple objectives conflict, taking into account the comprehensive needs of clinical safety, operational efficiency and green management. (7) Through the real-time computer vision recognition and active reminder of the intraoperative real-time monitoring module, it is also possible to intervene in the waste of consumables during the operation. Furthermore, through the configuration report automatic generation module, it can call the large language model to automatically generate a structured optimization suggestion report, which can improve the efficiency and scientific nature of management decisions. In addition, through the intelligent sorting and automatic packaging module, the AI ​​optimization scheme can be automatically transformed into a physical package entity, which can realize end-to-end intelligent closed-loop management from perception, analysis, optimization to execution, which is convenient for practical application and promotion.

[0074] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A system for analyzing and optimizing the utilization efficiency of components within a medical consumable kit using AI, characterized in that, include: The component usage status awareness module is used to obtain the usage status data of each component in the medical consumable kit under the target procedure. The medical consumable kit refers to a collection unit that is required for use under the target procedure and is composed of multiple medical consumable components that are pre-combined and packaged. The usage status data adopts any one of the states determined from the fully used state, the partially used state, and the unused state. The component utilization efficiency analysis module, communicatively connected to the component utilization status sensing module, is used to calculate the component-level utilization efficiency index of the medical consumable kit based on the utilization status data of each component. The component-level utilization efficiency index includes component utilization rate, surgical procedure matching degree, configuration redundancy, and / or waste index. The component utilization rate characterizes the ratio between the number of components actually used and the total number of components in the medical consumable kit. The surgical procedure matching degree characterizes the degree of conformity between the component configuration scheme of the medical consumable kit and the clinical guidelines or historical practices of the target surgical procedure. The configuration redundancy characterizes the ratio between the number of components in the medical consumable kit that are not fully used and the total number of components. The waste index characterizes the ratio between the number of components in the medical consumable kit that are discarded without being used and the total number of components. The component configuration AI optimization module is communicatively connected to the component usage efficiency analysis module. It is used to take component-level usage efficiency indicators as state input, component configuration schemes as action output, and a comprehensive score based on usage efficiency feedback results or usage efficiency and low-carbon feedback results as reward signals. Through continuous iterative learning based on reinforcement learning algorithms, it generates and dynamically optimizes the component configuration scheme of the medical consumable kit corresponding to the target procedure.

2. The medical consumable kit component usage efficiency analysis and AI optimization system as described in claim 1, characterized in that, The component uses a state-aware module including: A multimodal information acquisition unit is used to acquire preoperative information sets of each component in the medical consumable kit before surgery, and to acquire postoperative information sets of each component after surgery, wherein the preoperative information sets and the postoperative information sets are acquired by at least two of the following methods: RFID tag reading, image recognition, and weight sensing. The component usage status determination unit is communicatively connected to the multimodal information acquisition unit. For each component, based on multidimensional difference parameters between the corresponding preoperative and postoperative information sets, it outputs a corresponding usage status determination result using a component usage status classification model pre-trained using a deep learning algorithm. It then generates corresponding usage status data based on the usage status determination result. The usage status determination result includes three discrete labels: fully used, partially used, and unused. The determination of the partially used state is based on quantitative analysis of at least one feature among component surface morphology change characteristics, weight change characteristics, and packaging integrity change characteristics.

3. The medical consumable kit component usage efficiency analysis and AI optimization system as described in claim 1, characterized in that, The component utilization efficiency analysis module includes: The efficiency index calculation unit is used to calculate the initial component-level usage efficiency index of the medical consumable kit based on the usage status data of each component. The association model construction unit is used to construct an association model of medical consumables consumption based on the medical consumables usage records of different surgical procedures and different doctors in historical surgical data. The association model of medical consumables consumption is used to record the probability of use, quantity and model preference of each doctor for various components under different surgical procedures, so as to identify the differences between the medical consumables usage preferences of different doctors under the same surgical procedure and the best practices. An efficiency index correction unit is communicatively connected to the efficiency index calculation unit and the association model construction unit, respectively. It is used to correct the initial component-level usage efficiency index based on the procedure-doctor-component consumption association model, and to correct the procedure-level usage efficiency index by doctor preference and procedure difference, so as to obtain the corrected component-level usage efficiency index. The doctor preference correction is used to distinguish the difference between doctor's personal habitual use and clinically necessary use, and the procedure difference correction is used to distinguish the difference between different surgical complexities and different procedure subtypes.

4. The medical consumable kit component usage efficiency analysis and AI optimization system as described in claim 1, characterized in that, The comprehensive score is calculated based on a reward function that includes a usage efficiency reward, a low-carbon reward, and a completeness reward. The value of the usage efficiency reward is positively correlated with the component-level usage efficiency index from the efficiency feedback results. The value of the low-carbon reward is negatively correlated with the estimated carbon emissions from the low-carbon feedback results and the component configuration scheme corresponding to the medical consumable kit. The value of the completeness reward is positively correlated with the clinical completeness verification results of the component configuration scheme in meeting surgical requirements.

5. The medical consumable kit component usage efficiency analysis and AI optimization system as described in claim 4, characterized in that, The estimated carbon emissions were calculated as follows: Obtain the type, quantity, and material of each component in the component configuration scheme of the medical consumable kit, and also obtain at least one of the following for each component: supplier, transportation distance, and sterilization method; Based on the obtained results, the full-cycle carbon emissions corresponding to the component configuration scheme are estimated using a preset life cycle assessment model as the estimated carbon emissions.

6. The medical consumable kit component usage efficiency analysis and AI optimization system as described in claim 5, characterized in that, The component configuration AI optimization module is also used to calculate the actual carbon emissions based on the actual used components and unused returned components after each actual surgery of the target procedure, and to feed back the deviation between the actual carbon emissions and the estimated carbon emissions to the life cycle assessment model and the reward function, so as to continuously improve the accuracy of carbon emission estimation and the accuracy of low-carbon decision-making.

7. The medical consumable kit component usage efficiency analysis and AI optimization system as described in claim 1, characterized in that, The component configuration AI optimization module adopts a multi-objective reinforcement learning framework to simultaneously optimize at least two objectives, and achieves dynamic trade-offs between the at least two objectives through Pareto front solution or weighted scalarization. The at least two objectives include: maximizing component utilization efficiency, minimizing carbon emissions, maximizing clinical completeness, minimizing consumable costs, and minimizing waste generation.

8. The medical consumable kit component usage efficiency analysis and AI optimization system as described in claim 1, characterized in that, It also includes an intraoperative real-time monitoring module that is communicatively connected to the component usage status perception module. The intraoperative real-time monitoring module is used to identify the intraoperative component usage behavior in real time through computer vision technology during the operation. When it detects at least one wasteful behavior of consumable components based on the identification results, it actively sends a reminder message to the operating room terminal or medical staff mobile terminal. The at least one wasteful behavior of consumable components includes abnormal selection of consumable component specifications, improper use and / or unnecessary unpacking. The reminder message includes a description of the identified wasteful behavior, component information of the wasted component and / or suggested alternative operation schemes.

9. The medical consumable kit component usage efficiency analysis and AI optimization system as described in claim 1, characterized in that, It also includes an automatic configuration report generation module that is communicatively connected to the component configuration AI optimization module and the component usage efficiency analysis module. The automatic configuration report generation module is used to call a large language model, combine the optimization results of the component configuration scheme, component-level usage efficiency indicators, surgical procedure clinical guidelines and doctor operation preference data, and adopt a multi-round inference strategy based on a preset Prompt template to sequentially generate an efficiency comparison analysis paragraph, an optimization suggestion analysis paragraph and a risk warning paragraph, and finally synthesize and generate a structured medical consumable package configuration optimization suggestion report under the target surgical procedure.

10. The medical consumable kit component usage efficiency analysis and AI optimization system as described in claim 1, characterized in that, It also includes an intelligent sorting and automatic packaging module that is connected to the component configuration AI optimization module. The intelligent sorting and automatic packaging module is used to control the intelligent sorting execution structure to sort out the corresponding components from the consumables library according to the optimal component configuration scheme generated and output by the component configuration AI optimization module after receiving the optimal component configuration scheme. After sorting, it automatically packages the components to form a medical consumable package corresponding to the target procedure.

Images