Medical instrument management method for dental treatment

Abstract

The invention discloses a medical instrument management method for dental treatment, belongs to the technical field of medical instrument management, and solves the problems that the priority is calculated only based on the vacancy rate and the abnormal reminding frequency of the instruments and the specific requirements and treatment scenes of patients are neglected when the use priority of the same kind of medical instruments is sequenced by the existing method. The method comprises the steps of performing normalization processing in combination with management associated data, performing feature extraction and fusion processing on a self-encoding set, calculating an instrument risk value of a dental instrument to be selected, and calculating an instrument recommendation evaluation value in combination with the instrument risk value and the Bayesian theorem; according to the method, when the use priority in the use demand information is determined, the vacancy rate and the use frequency of the medical instrument are considered, multi-dimensional information such as the treatment time of the patient and the disease priority is introduced, the use priority of the medical instrument is dynamically updated, and the use priority of the medical instrument is determined through the Gumbel-Copula function and the particle swarm optimization algorithm. And the relevance between the medical instrument management data is deeply mined.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical equipment management, and in particular relates to a medical equipment management method for dental treatment. Background Art

[0002] Dental medical devices play an important role in modern oral care. Dental treatment involves a variety of medical devices, such as dental mobile phones, implants, electric motors, treatment handpieces, etc. There are many types of dental medical devices, ranging from basic inspection tools to complex treatment equipment. As people pay more attention to oral health, the market size of dental medical devices continues to grow. With the development of dental treatment technology, the digitalization and automation of device management has become an inevitable trend in the development of the industry. In this context, how to improve the efficiency of device use and reduce risks through efficient management methods has become an urgent problem to be solved in the industry.

[0003] Chinese patent CN118197588B discloses an automated medical device operation management system and method, the management method comprising the following steps: setting a time window, calculating the high-frequency usage rate of medical devices in a unit cycle, and determining several usage peaks; obtaining the abnormal interval time and the estimated unit usage time of the medical device according to the time interval between two adjacent abnormal reminders sent by the medical device and the unit usage time; obtaining the patient's disease information, selecting the same type of target medical devices required by the patient, and sorting the use priority of the same type of medical devices; setting an equipment maintenance cycle, calculating the estimated number of uses of the medical device in the remaining time, and judging whether to adjust the equipment maintenance time; obtaining the status of each medical device in real time, predicting the idle number of medical devices after the next usage peak period, and judging whether to send a replenishment reminder; however, when the existing method sorts the use priority of the same type of medical devices, the priority is calculated only based on the idle rate of the device and the number of abnormal reminders, ignoring the specific needs of the patient and the treatment scenario, and the priority is determined statically, and the priority of the medical device cannot be dynamically adjusted according to real-time data. In view of the above problems, we propose a medical device management method for dental treatment. Summary of the invention

[0004] The purpose of the present invention is to address the shortcomings of the prior art and provide a medical device management method for dental treatment, which solves the problem that the existing method only calculates the priority based on the idle rate and the number of abnormal reminders of the equipment when sorting the use priority of similar medical equipment, ignoring the specific needs of the patient and the treatment scenario, and the priority determination is static, and the priority of the medical equipment cannot be dynamically adjusted according to real-time data.

[0005] The present invention is implemented as follows: a medical device management method for dental treatment, the medical device management method for dental treatment comprising: Obtaining dental equipment usage demand information, identifying the equipment name and management identification code based on the usage demand information, and using the management identification code to trigger equipment usage instructions; In response to the device use instruction, indexing the management-related data of the dental device based on the device use instruction, loading the use demand information, determining the use priority in the use demand information based on the Bayesian theorem, using the use priority as a weight factor, and performing normalization processing in combination with the management-related data to obtain a normalized data set; Load the normalized data set, encode and reconstruct the normalized data set based on the pre-trained device evaluation model and decode it to obtain an auto-encoded set, extract and fuse the features of the auto-encoded set, calculate the device risk value of the dental device to be selected, and the device evaluation model combines the device risk value and Bayesian theorem to calculate the device recommendation evaluation value; An instrument usage queue is generated based on the instrument recommendation evaluation value, and dental instruments in the instrument usage queue that exceed the recommendation threshold are pushed.

[0006] Preferably, the method for determining the use priority in the use demand information based on the Bayesian theorem includes: Load the usage demand information, identify the patient treatment time, patient disease priority, device name, and management identification code in the usage demand information, index the device idle rate and usage frequency corresponding to the device name based on the management identification code, and use the patient treatment time as the initial probability of use; Load the initial probability of use and the patient's disease priority, and correct the initial probability of use based on the particle swarm optimization algorithm combined with the patient's disease priority to obtain the corrected probability of use; (1) (2) (3) in, Respectively After iterations, use the corrected probability and the initial probability. are the current iteration number and the total iteration number, respectively. is the adaptive acceleration factor based on the particle swarm optimization algorithm, Indicates the patient's disease priority, Particle swarm optimization algorithm The optimization iteration value, The optimization iteration value, represents the search for optimal uniform random numbers, Respectively The individual optimal extreme position of each particle and the global optimal extreme position; Obtain the device idle rate and usage frequency corresponding to the device name, and calculate the likelihood probability based on the patient's disease priority and the device idle rate and usage frequency; (4) (5) in, represents the likelihood probability, For the The weight coefficient of each dental instrument, For the The initial weight coefficient of each dental instrument, They are the equipment idle rate and usage frequency. is the number of dental instruments, for The average value of the initial weight coefficients of the dental instruments; Load the idle rate and usage frequency of the instruments, and take into account the Gumbel-Copula function to calculate the total probability of dental instruments; The corrected probability, likelihood probability, and total probability of use of dental instruments are obtained, and the use priority in the use demand information is determined based on the Bayesian theorem.

[0007] Preferably, the total probability of the dental instrument is calculated by the following formula: (6) in, represents the total probability of dental instruments, is the correlation parameter of the Gumbel-Copula function, They are the equipment idle rate and usage frequency; (7) in, Indicates the usage priority in the usage requirement information.

[0008] Preferably, the method for normalizing the management-related data includes: Load management-related data and usage priority, where management-related data includes device idle rate, usage frequency, device operating parameters, number of adverse events, type of adverse events, injury level, risk level, usage period, and maintenance records. Use the maximum-minimum method to standardize and preprocess the management-related data; Obtain the management-related data after standardization preprocessing, read and identify the management-related data based on the Pandas library, calculate the parameter correlation coefficient of the characteristic parameters in the management-related data using the Minkowski distance, and determine the average correlation coefficient of the characteristic parameters in the management-related data based on the parameter correlation coefficient; The parameter correlation coefficient of the characteristic parameters is calculated by the following formula: (8) in, Indicates the current feature parameters Associated feature parameters The parameter correlation coefficient, Indicates the associated feature parameters The mean of is the adjustment parameter of the Minkowski distance algorithm; Based on the preset dynamic threshold of Davidson-Botting Index, the dynamic threshold of association is used to filter the characteristic parameters in the management association data that meet the requirements, and obtain the parameter screening set; Load the parameter filter set and usage priority, use the usage priority as the weight factor, and calculate the correlation matrix of the parameter filter set and usage priority based on the Pandas algorithm combined with the Pearson coefficient; The parameter filter set and the correlation matrix of the usage priority are loaded, the parameter filter set and the correlation matrix of the usage priority are mapped to the parameter filter set based on the dynamic time warping algorithm, and the parameter filter set is set as a normalized data set.

[0009] Preferably, the associated dynamic threshold is calculated by the following formula: (9) (10) in, represents the associated dynamic threshold, The Davidson-Botting index for managing feature parameters in linked data, is the number of characteristic parameters, are the current feature parameters Associated feature parameters The parameter correlation coefficient and the average correlation coefficient of the characteristic parameters are They are the current feature parameter cluster center and the associated feature parameter cluster center respectively. Indicates the Euclidean distance between the current feature parameter cluster center and the associated feature parameter cluster center.

[0010] Preferably, the device evaluation model training method comprises: The modeling data set is crawled based on crawler technology, and the crawled modeling data set is normalized. The normalized modeling data set is divided into a training set and a test set in a ratio of 3:1. Load the pre-built device evaluation model, initialize the initial weights and thresholds of the device evaluation model, and set the activation function, loss function, and training rounds of the device evaluation model; The equipment evaluation model is iteratively trained using the training set. During the training, the optimal values ​​of the initial weights and thresholds of the equipment evaluation model are searched in combination with the sparrow algorithm. The initial weights and thresholds are adjusted based on the convergence and alienation operations of the sparrow population, and the converged equipment evaluation model is output; Load the test set, use the test set as input, solve the device risk value and recommended assessment value, and calculate the comprehensive error of the device risk value and recommended assessment value; Determine whether the comprehensive error meets the preset error threshold. If the comprehensive error meets the preset error threshold, output the trained device evaluation model.

[0011] Preferably, the device evaluation model uses a BP neural network model as an initial model, the initial model includes an input layer, a hidden layer and an output layer, an autoencoder module is introduced between the input layer and the hidden layer, and the hidden layer is frozen, an improved CNN architecture is used to replace the hidden layer, and an evaluation classifier is introduced between the improved CNN architecture and the output layer; Among them, the input layer is provided with three layers, the autoencoder module is composed of an encoder and a decoder, the encoder and the decoder are connected through a convolution layer, the autoencoder includes three groups of 3×3 convolution blocks, a maximum pooling layer, an average pooling layer and a CBAM layer, and the CBAM layer introduces a CBAM attention mechanism. The autoencoder is used to encode and reconstruct the normalized data set, and extract the core features of the feature parameters in the normalized data set, and reduce the core features of the feature parameters to low-dimensional feature codes. The decoder includes two groups of 3×3 convolutional layers, two groups of 3×3 deconvolutional layers, a FreeLB layer and a fully connected layer. The decoder is used to obtain low-dimensional feature codes, and the low-dimensional feature codes are perturbed and decoded based on the adversarial perturbation algorithm, and the decoding process obtains the autoencoder set; The improved CNN architecture consists of a feature extraction block, a fully connected layer, and a feature fusion block connected in sequence. A cross-attention mechanism is introduced in the feature extraction block, and an activation function Sigmoid is defined. The feature extraction block is used to identify the autoencoder set, extract the local features of the autoencoder set based on the cross-attention mechanism and the batch normalization layer stack, and output a multi-level feature set. The feature fusion block consists of an upsampling layer, a non-local attention layer, an SPP layer, a batch normalization layer, and an SC block. The feature fusion block is based on a late fusion strategy, uses a non-local attention layer to enhance the long-distance dependency of the multi-level feature set, and uses the SPP layer to extract multi-scale features. The multi-scale features of different scales are fused through the SC block to generate a final fused feature set. The evaluation classifier includes an LSTM layer and a Bi-FPM block, and introduces a multi-scale constraint algorithm. The evaluation classifier calculates the instrument risk value of the dental instrument to be selected based on the multi-scale constraint algorithm, and calculates the instrument recommendation evaluation value through the Bi-FPM block in combination with the instrument risk value and Bayesian theorem.

[0012] Preferably, the device evaluation model combines the device risk value and the Bayesian theorem to calculate the device recommended evaluation value, including: Load the normalized data set, reconstruct the normalized data set encoding by the autoencoder, extract the core features of the feature parameters in the normalized data set, and reduce the core features of the feature parameters to low-dimensional feature encoding; The decoder obtains low-dimensional feature codes, perturbs and decodes the low-dimensional feature codes based on an adversarial perturbation algorithm, and obtains an autoencoder set through decoding. The feature extraction block identifies the autoencoder set, extracts local features of the autoencoder set based on the cross-attention mechanism and batch normalization layer stacking, and outputs a multi-level feature set; The feature fusion block is based on the late fusion strategy, uses the non-local attention layer to enhance the long-distance dependencies of the multi-level feature set, and uses the SPP layer to extract multi-scale features. The multi-scale features of different scales are fused through the SC block to generate the final fused feature set; The evaluation classifier calculates the device risk value of the dental device to be selected based on a multi-scale constraint algorithm, and calculates the device recommendation evaluation value by combining the device risk value and Bayesian theorem through the Bi-FPM block.

[0013] Preferably, the device recommendation evaluation value is calculated by the following formula: (11) (12) in, Respectively represent the recommended assessment value of the device and the risk value of the device. are the number of characteristic parameters and the number of dental instruments to be selected, are the usage priority and likelihood probability respectively, is the fused feature set and the variance of the fused feature set.

[0014] Compared with the prior art, the embodiments of the present application have the following beneficial effects: In the embodiment of the present invention, when determining the use priority in the use demand information based on the Bayesian theorem, not only the idle rate and use frequency of medical devices are considered, but also multi-dimensional information such as patient treatment time and disease priority is introduced. Through the Bayesian theorem, this information is integrated to dynamically update the use priority of medical devices, and through the Gumbel-Copula function and particle swarm optimization algorithm, the correlation between medical device management data is deeply mined. It can make more comprehensive use of data information and provide more accurate support for management decisions. It can further more comprehensively reflect the actual needs and use scenarios of medical devices, ensure the accuracy of the recommended evaluation value of the device, and overcome the existing method of sorting the use priority of similar medical devices, only based on the idle rate of the device and the number of abnormal reminders to calculate the priority, ignoring the specific needs and treatment scenarios of patients, and the priority is determined statically, and the priority of medical devices cannot be dynamically adjusted according to real-time data.

[0015] In the embodiment of the present invention, when determining the usage priority in the usage demand information based on the Bayesian theorem, the correlation between the management data is deeply mined through the Gumbel-Copula function and the PSO algorithm, providing more comprehensive data support for management decisions. In addition, advanced methods such as the Bayesian theorem, the particle swarm optimization algorithm, the Gumbel-Copula function, etc. are introduced to overcome the limitations of the existing technology and realize multi-dimensional comprehensive evaluation, dynamic adjustment and optimization of the usage priority.

[0016] In the embodiment of the present invention, the dynamic threshold of correlation is preset based on the Davidson-Borgin Index. The Davidson-Borgin Index is a commonly used statistic for measuring the degree of correlation between variables, which can more accurately reflect the linear and nonlinear relationships between variables. By presetting the threshold of the index, the setting of the threshold can be made more scientific and reasonable, avoiding the deviation and inaccuracy that may be caused by subjective setting of the threshold. At the same time, the dynamic threshold of correlation is used to filter the characteristic parameters in the management-related data that meet the requirements to obtain a parameter screening set. This step can remove those characteristic parameters that are less relevant to the research objectives and retain the characteristic parameters that are of great significance to the problem analysis, thereby simplifying the data structure, reducing the data dimension, and improving the efficiency and accuracy of subsequent analysis.

[0017] In an embodiment of the present invention, an equipment evaluation model and a training method thereof are provided. The equipment evaluation model combines an autoencoder module, an improved CNN architecture and an evaluation classifier, can process various types of data, and fuse multimodal information, improve the feature extraction capability of the model, and enhance the model's ability to focus on key features through a cross-attention mechanism and a non-local attention layer, thereby improving the model's processing accuracy for complex data. The DTW algorithm can effectively process nonlinear changes in time series data, thereby improving the model's alignment and matching capabilities for time series data. The evaluation classifier includes an LSTM layer and a Bi-FPM block, and introduces a multi-scale constraint algorithm and Bayesian theorem, which can comprehensively consider the calculation of equipment risk values ​​and recommended evaluation values, accurately evaluate the risk value through a multi-scale constraint algorithm, and then use the Bi-FPM block in combination with the risk value and Bayesian theorem to calculate the recommended evaluation value, thereby improving the accuracy and reliability of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the implementation flow of the medical device management method for dental treatment provided by the present invention.

[0019] Figure 2 A schematic diagram of the implementation process of a method for determining the use priority in use demand information based on the Bayesian theorem is shown.

[0020] Figure 3 A schematic diagram of the implementation process of a normalization processing method combined with management-related data is shown.

[0021] Figure 4 A schematic diagram of the implementation process of the equipment evaluation model training method is shown.

[0022] Figure 5 The present invention shows a schematic diagram of the implementation process of the device evaluation model combining the device risk value and the Bayesian theorem to calculate the recommended evaluation value of the device.

[0023] Figure 6 A model block diagram of the device assessment model is shown. DETAILED DESCRIPTION

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of this application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0025] When the existing methods rank the usage priorities of similar medical devices, the priority is calculated only based on the idle rate of the device and the number of abnormal reminders, ignoring the specific needs of the patient and the treatment scenario. In addition, the priority is determined statically and the priority of the medical device cannot be dynamically adjusted according to real-time data. To address the above problems, we propose a medical device management method for dental treatment. In short, when the method is implemented, the usage demand information of the dental device is first obtained, the device name and management identification code are identified based on the usage demand information, and then the management-related data of the dental device are indexed based on the device usage instructions. The usage priority in the usage demand information is determined based on the Bayesian theorem, and the usage priority is used as a weight factor. The management-related data is normalized, and the normalized data set is encoded, reconstructed and decoded based on the pre-trained device evaluation model to obtain an autoencoder set. The features of the autoencoder set are extracted and fused, and the device risk value of the dental device to be selected is calculated. The device evaluation model combines the device risk value and the Bayesian theorem to calculate the device recommendation evaluation value. Finally, the device usage queue is generated based on the device recommendation evaluation value, and the dental devices in the device usage queue that exceed the recommendation threshold are pushed. In the embodiment of the present invention, when determining the use priority in the use demand information based on the Bayesian theorem, not only the idle rate and use frequency of medical devices are considered, but also multi-dimensional information such as patient treatment time and disease priority is introduced. Through the Bayesian theorem, this information is integrated to dynamically update the use priority of medical devices, and through the Gumbel-Copula function and particle swarm optimization algorithm, the correlation between medical device management data is deeply mined. It can make more comprehensive use of data information and provide more accurate support for management decisions. It can further more comprehensively reflect the actual needs and use scenarios of medical devices, ensure the accuracy of the recommended evaluation value of the device, and overcome the existing method of sorting the use priority of similar medical devices, only based on the idle rate of the device and the number of abnormal reminders to calculate the priority, ignoring the specific needs and treatment scenarios of patients, and the priority is determined statically, and the priority of medical devices cannot be dynamically adjusted according to real-time data.

[0026] The embodiment of the present invention provides a medical device management method for dental treatment. Figure 1 The figure shows a schematic diagram of the implementation process of the medical device management method for dental treatment, wherein the medical device management method for dental treatment specifically includes: Step S10, obtaining dental instrument use demand information, identifying the instrument name and management identification code based on the use demand information, and using the management identification code to trigger the instrument use instruction; It should be noted that the usage demand information includes but is not limited to the name of the instrument, management identification code, basic patient information, patient treatment time, patient disease priority, treatment time window, doctor's advice, and historical usage data, and the management identification code is a code or identifier used to uniquely identify and manage dental instruments. It is a key information field used to quickly locate and trigger the use instructions of the instrument in the system. An exemplary management identification code may be "DentalDrill-130-005" to represent a dental drill of model 130 with a serial number of 005. Dental instruments include but are not limited to dental treatment chairs, dental comprehensive treatment machines and accessories (side boxes, dental lights, instrument trays, mouthwash water supply devices), dental chairs, electric motors / pneumatic motors for dental handpieces, dental electric suction machines / air compressors, curing equipment, dental implant machines, root canal treatment equipment, tooth bleaching equipment, dental implant equipment, implant positioners, dental abutment placers, implant stability testers, dental finishing needles, and dental brushes.

[0027] Step S20, in response to the instrument use instruction, indexing the management-related data of the dental instrument based on the instrument use instruction, loading the use requirement information, determining the use priority in the use requirement information based on the Bayesian theorem, using the use priority as a weight factor, and performing normalization processing in combination with the management-related data to obtain a normalized data set; Step S30, loading a normalized data set, encoding and reconstructing the normalized data set based on a pre-trained device evaluation model, and decoding the normalized data set to obtain an auto-encoded set, extracting and fusing features of the auto-encoded set, and calculating the device risk value of the dental device to be selected. The device evaluation model combines the device risk value and Bayesian theorem to calculate the device recommendation evaluation value; Step S40: generating an instrument use queue based on the instrument recommendation evaluation value, and pushing dental instruments in the instrument use queue that exceed the recommendation threshold.

[0028] In this embodiment, by generating an instrument usage queue, dentists can quickly obtain the required instruments, reduce the time for instrument preparation and search, and thus significantly improve treatment efficiency. This not only reduces the patient's waiting time, but also improves the overall management and operation efficiency.

[0029] In the embodiment of the present invention, when determining the use priority in the use demand information based on the Bayesian theorem, not only the idle rate and use frequency of medical devices are considered, but also multi-dimensional information such as patient treatment time and disease priority is introduced. Through the Bayesian theorem, this information is integrated to dynamically update the use priority of medical devices, and through the Gumbel-Copula function and particle swarm optimization algorithm, the correlation between medical device management data is deeply mined. It can make more comprehensive use of data information and provide more accurate support for management decisions. It can further more comprehensively reflect the actual needs and use scenarios of medical devices, ensure the accuracy of the recommended evaluation value of the device, and overcome the existing method of sorting the use priority of similar medical devices, only based on the idle rate of the device and the number of abnormal reminders to calculate the priority, ignoring the specific needs and treatment scenarios of patients, and the priority is determined statically, and the priority of medical devices cannot be dynamically adjusted according to real-time data.

[0030] The embodiment of the present invention provides a method for determining the usage priority in usage demand information based on Bayesian theorem. Figure 2 The following is a schematic diagram of the implementation process of a method for determining the use priority in the use demand information based on the Bayesian theorem. The method for determining the use priority in the use demand information based on the Bayesian theorem specifically includes: Step S101, loading usage demand information, identifying the patient treatment time, patient disease priority, device name, and management identification code in the usage demand information, indexing the device idle rate and usage frequency corresponding to the device name based on the management identification code, and taking the patient treatment time as the initial usage probability; In an embodiment of the present invention, the loaded usage demand information covers multi-dimensional data such as patient treatment time, disease priority, instrument name and management identification code, providing a comprehensive basis for subsequent priority evaluation. By taking the patient treatment time as the initial probability of use, it can reflect the needs of the current treatment scenario in real time and ensure the timeliness of the priority evaluation.

[0031] Step S102, loading the initial probability of use and the patient's disease priority, and correcting the initial probability of use based on a particle swarm optimization algorithm combined with the patient's disease priority to obtain a corrected probability of use; (1) (2) (3) in, Respectively After iterations, use the corrected probability and the initial probability. are the current iteration number and the total iteration number, respectively. In this embodiment, the total iteration number may be 4-15. is the adaptive acceleration factor based on the particle swarm optimization algorithm, Indicates the patient's disease priority. In this embodiment, the patient's disease priority can be determined based on the severity of the disease, the urgency of treatment, the patient's age, the patient's health, and the opinions of dentists, nurses, and other medical professionals. For example, in this embodiment, the patient's disease priority can be 1-12. Particle swarm optimization algorithm The optimization iteration value, The optimization iteration value, It indicates the search for optimal uniform random number, which can be 0.5-1.5. Respectively The individual optimal extreme position of each particle and the global optimal extreme position.

[0032] In the embodiment of the present invention, the PSO algorithm optimizes the initial probability through swarm intelligence, avoids the problem of local optimal solution, and ensures the global optimality of priority evaluation. The adaptive acceleration factor of the PSO algorithm can automatically adjust the optimization speed according to the iterative process, thereby improving the convergence efficiency of the algorithm.

[0033] Step S103, obtaining the idle rate and usage frequency of the equipment corresponding to the equipment name, taking the patient's disease priority as a priori condition, and calculating the likelihood probability in combination with the idle rate and usage frequency of the equipment; (4) (5) in, represents the likelihood probability, For the The weight coefficient of each dental instrument, For the The initial weight coefficient of each dental instrument, They are the equipment idle rate and usage frequency. is the number of dental instruments, for The average value of the initial weight coefficients of the dental instruments; In the embodiment of the present invention, the calculation of likelihood probability is based on real-time data, which can dynamically reflect the usage status of the equipment and ensure the accuracy of the priority evaluation. A weight coefficient is introduced to comprehensively consider the idle rate and usage frequency of the equipment to avoid the influence of a single factor on the priority evaluation.

[0034] Step S104, loading the idle rate and usage frequency of the instrument, taking into account the Gumbel-Copula function to calculate the total probability of the dental instrument; Among them, the total probability of dental instruments is calculated by the following formula: (6) in, represents the total probability of dental instruments, is the correlation parameter of the Gumbel-Copula function, which can be 1-5. They are equipment idle rate and usage frequency respectively.

[0035] It should be noted that the Gumbel-Copula function can simultaneously consider the correlation between multiple variables (such as idle rate and frequency of use), thereby improving the comprehensiveness of priority assessment. The calculation of total probability provides a basis for the risk quantification of the device and can more accurately assess the risk of device use.

[0036] Step S105, obtaining the usage correction probability, likelihood probability, and total probability of the dental instrument, and determining the usage priority in the usage demand information based on the Bayesian theorem.

[0037] (7) in, Indicates the usage priority in the usage requirement information.

[0038] In the embodiment of the present invention, when determining the usage priority in the usage demand information based on the Bayesian theorem, the correlation between the management data is deeply mined through the Gumbel-Copula function and the PSO algorithm, providing more comprehensive data support for management decisions. In addition, advanced methods such as the Bayesian theorem, the particle swarm optimization algorithm, the Gumbel-Copula function, etc. are introduced to overcome the limitations of the existing technology and realize multi-dimensional comprehensive evaluation, dynamic adjustment and optimization of the usage priority.

[0039] The embodiment of the present invention provides a method for normalizing management-related data. Figure 3 A schematic diagram of the implementation process of a method for normalizing the management-related data is shown. The method for normalizing the management-related data specifically includes: Step S201, loading management-related data and usage priority, wherein the management-related data includes device idle rate, usage frequency, device operating parameters, number of adverse events, adverse event type, injury level, risk level, usage period, and maintenance record, and using the maximum-minimum method to standardize and preprocess the management-related data; In the embodiment of the present invention, the maximum-minimization method scales the data to the interval [0, 1], eliminating the differences in different feature dimensions and orders of magnitude. For example, the operating parameters of the device may have large values, while the number of adverse events is relatively small. After standardization, they can participate in subsequent analysis on an equal basis to avoid excessive amplification or reduction of certain indicators in the analysis due to excessive numerical differences, making the data comparable. At the same time, in order to ensure data quality, outliers and missing values ​​can also be deleted. At the same time, the idle rate of the device can reflect the saturation of the use of the device, and the number and type of adverse events can reflect the risk status during the use of the device.

[0040] Step S202, obtaining the management-related data after standardization preprocessing, reading and identifying the management-related data based on the Pandas library, calculating the parameter correlation coefficient of the characteristic parameters in the management-related data using the Minkowski distance, and determining the average correlation coefficient of the characteristic parameters in the management-related data based on the parameter correlation coefficient; In this embodiment, the associated data is read, identified and managed based on the Pandas library, which is a powerful data processing and analysis tool in Python with efficient data reading, cleaning and conversion functions. It can quickly process large-scale data sets, improve the efficiency and accuracy of data processing, and save time and labor costs.

[0041] The parameter correlation coefficient of the characteristic parameters is calculated by the following formula: (8) in, Indicates the current feature parameters Associated feature parameters The parameter correlation coefficient, Indicates the associated feature parameters The mean of is an adjustment parameter of the Minkowski distance algorithm. In this embodiment, the adjustment parameter can be set to 4-9; Step S203, presetting a correlation dynamic threshold based on the Davidson-Botting index, and using the correlation dynamic threshold to filter characteristic parameters in the management correlation data that meet the requirements to obtain a parameter screening set; The associated dynamic threshold is calculated by the following formula: (9) (10) in, represents the associated dynamic threshold, The Davidson-Botting index for managing feature parameters in linked data, is the number of characteristic parameters, are the current feature parameters Associated feature parameters The parameter correlation coefficient and the average correlation coefficient of the characteristic parameters are They are the current feature parameter cluster center and the associated feature parameter cluster center respectively. Indicates the Euclidean distance between the current feature parameter cluster center and the associated feature parameter cluster center.

[0042] In the embodiment of the present invention, the dynamic threshold of correlation is preset based on the Davidson-Borgin Index. The Davidson-Borgin Index is a commonly used statistic for measuring the degree of correlation between variables, which can more accurately reflect the linear and nonlinear relationships between variables. By presetting the threshold of the index, the setting of the threshold can be made more scientific and reasonable, avoiding the deviation and inaccuracy that may be caused by subjective setting of the threshold. At the same time, the dynamic threshold of correlation is used to filter the characteristic parameters in the management-related data that meet the requirements to obtain a parameter screening set. This step can remove those characteristic parameters that are less relevant to the research objectives and retain the characteristic parameters that are of great significance to the problem analysis, thereby simplifying the data structure, reducing the data dimension, and improving the efficiency and accuracy of subsequent analysis.

[0043] Step S204, load the parameter screening set and the usage priority, use the priority as a weight factor, and calculate the correlation matrix of the parameter screening set and the usage priority based on the Pandas algorithm combined with the Pearson coefficient; the Pandas algorithm is efficient and flexible, and can quickly process large-scale data sets; the Pearson coefficient is a commonly used correlation measurement method that can accurately measure the linear correlation between two variables. The combination of the two can give full play to their respective advantages, more comprehensively and accurately calculate the correlation between the parameter screening set and the usage priority, and provide a reliable basis for subsequent analysis.

[0044] Step S205, load the correlation matrix of the parameter filter set and the usage priority, map the correlation matrix of the parameter filter set and the usage priority to the parameter filter set based on the dynamic time warping algorithm, and set the parameter filter set as a normalized data set.

[0045] In this embodiment, the parameter screening set and the correlation matrix of the use priority are mapped to the parameter screening set based on the dynamic time warping algorithm. The algorithm can effectively process time series data or data with dynamic change characteristics, and has good adaptability to complex management related data. Through this mapping method, the potential laws and relationships in the data can be further explored, and the data analysis and processing capabilities can be improved.

[0046] The embodiment of the present invention provides a device evaluation model training method, Figure 4 The following is a schematic diagram of the implementation process of the device evaluation model training method, which specifically includes: Step S301, crawling a modeling data set based on a crawler technology, normalizing the crawled modeling data set, and dividing the normalized modeling data set into a training set and a test set at a ratio of 3:1; In the embodiment of the present invention, when crawling the modeling data set based on the crawler technology, data can be crawled from public data set platforms (such as UCI Machine Learning Repository), dental clinic websites, medical equipment manufacturer websites, etc. The crawler technology can be scripted based on Python's requests and BeautifulSoup libraries. Acquiring the modeling data set through the crawler technology can collect rich data from multiple sources to ensure the diversity and comprehensiveness of the data.

[0047] Step S302, loading a pre-built device evaluation model, initializing the initial weights and thresholds of the device evaluation model, and setting the activation function, loss function, and training rounds of the device evaluation model; In the embodiment of the present invention, the initial weights and thresholds of the device evaluation model are initialized, and key parameters such as activation function, loss function and training rounds are reasonably set, providing good initial conditions for model training. Reasonable parameter settings can accelerate the convergence speed of the model, avoid falling into a local optimal solution, and improve the effect of model training. The training rounds can be 50-100 times.

[0048] Step S303, iteratively train the equipment evaluation model using the training set. During the training, the initial weights and thresholds of the equipment evaluation model are searched for in combination with the sparrow algorithm. The initial weights and thresholds are adjusted based on the convergence and alienation operations of the sparrow population, and the converged equipment evaluation model is output. The sparrow algorithm is used to search for the initial weights and thresholds of the equipment evaluation model. The algorithm has strong global search capabilities and optimization efficiency, and can find a better combination of parameters in a complex parameter space to improve the performance of the model. At the same time, adjusting the initial weights and thresholds based on the convergence and alienation operations of the sparrow population can increase the diversity and adaptability of the model and avoid the occurrence of overfitting.

[0049] Step S304, loading the test set, taking the test set as input, solving the device risk value and the recommended evaluation value, calculating the comprehensive error of the device risk value and the recommended evaluation value, calculating the comprehensive error of the device risk value and the recommended evaluation value, comprehensively considering the error of multiple indicators, and being able to more comprehensively and objectively reflect the performance of the model, providing a reliable basis for subsequent judgment and optimization; Step S305, determining whether the comprehensive error meets a preset error threshold. In this embodiment, the error threshold is set to 0.105-0.152; Step S306: if the comprehensive error meets the preset error threshold, the trained instrument evaluation model is output.

[0050] If the comprehensive error does not meet the preset error threshold, the initial weights and thresholds of the device evaluation model are tracked and optimized using adaptive moment estimation, and the process returns to step S303 to continue iterative training of the device evaluation model.

[0051] In this embodiment, if the comprehensive error does not meet the preset error threshold, adaptive moment estimation is used to track and optimize the initial weights and thresholds of the device evaluation model, which can automatically adjust the model parameters according to the error situation, so that the model can adaptively learn and improve, and continuously improve the performance of the model until the preset error threshold is reached.

[0052] In this embodiment, if Figure 6 As shown, a model block diagram of the device evaluation model is shown, wherein the device evaluation model uses a BP neural network model as an initial model, the initial model includes an input layer, a hidden layer and an output layer, an autoencoder module is introduced between the input layer and the hidden layer, and the hidden layer is frozen, an improved CNN architecture is used to replace the hidden layer, and an evaluation classifier is introduced between the improved CNN architecture and the output layer; Among them, the input layer is provided with three layers, the autoencoder module is composed of an encoder and a decoder, the encoder and the decoder are connected through a convolution layer, the autoencoder includes three groups of 3×3 convolution blocks, a maximum pooling layer, an average pooling layer and a CBAM layer, and the CBAM layer introduces a CBAM attention mechanism. The autoencoder is used to encode and reconstruct the normalized data set, and extract the core features of the feature parameters in the normalized data set, and reduce the core features of the feature parameters to low-dimensional feature codes. The decoder includes two groups of 3×3 convolutional layers, two groups of 3×3 deconvolutional layers, a FreeLB layer and a fully connected layer. The decoder is used to obtain low-dimensional feature codes, and the low-dimensional feature codes are perturbed and decoded based on the adversarial perturbation algorithm, and the decoding process obtains the autoencoder set; The improved CNN architecture consists of a feature extraction block, a fully connected layer, and a feature fusion block connected in sequence. A cross-attention mechanism is introduced in the feature extraction block, and an activation function Sigmoid is defined. The feature extraction block is used to identify an autoencoder set, extract local features of the autoencoder set based on a cross-attention mechanism and a batch normalization layer stack, and output a multi-level feature set. The feature fusion block consists of an upsampling layer, a non-local attention layer, an SPP layer, a batch normalization layer, and an SC block. The feature fusion block is based on a late fusion strategy, uses a non-local attention layer to enhance the long-distance dependency of a multi-level feature set, and uses an SPP layer to extract multi-scale features. Multi-scale features of different scales are fused through the SC block to generate a final fused feature set. The activation function of the feature fusion block may be a Tanh activation function, and the loss function may be a mean square error loss function. The evaluation classifier includes an LSTM layer and a Bi-FPM block, and introduces a multi-scale constraint algorithm. The evaluation classifier calculates the instrument risk value of the dental instrument to be selected based on the multi-scale constraint algorithm, and calculates the instrument recommendation evaluation value through the Bi-FPM block in combination with the instrument risk value and Bayesian theorem. The activation function of the evaluation classifier can be a Relu activation function, and the loss function can be a cross entropy loss function.

[0053] It should be noted that the evaluation classifier combines LSTM and Bi-FPM blocks. The LSTM layer can process time series data and is suitable for dynamic risk assessment of medical devices. Through the multi-scale constraint algorithm, it can comprehensively consider the characteristics of different scales to improve the accuracy of the evaluation. Combined with the Bayesian theorem, it can dynamically update the evaluation results and improve the adaptability and reliability of the model. In an embodiment of the present invention, an equipment evaluation model and a training method thereof are provided. The equipment evaluation model combines an autoencoder module, an improved CNN architecture and an evaluation classifier, can process various types of data, and fuse multimodal information, improve the feature extraction capability of the model, and enhance the model's ability to focus on key features through a cross-attention mechanism and a non-local attention layer, thereby improving the model's processing accuracy for complex data. The DTW algorithm can effectively process nonlinear changes in time series data, thereby improving the model's alignment and matching capabilities for time series data. The evaluation classifier includes an LSTM layer and a Bi-FPM block, and introduces a multi-scale constraint algorithm and Bayesian theorem, which can comprehensively consider the calculation of equipment risk values ​​and recommended evaluation values, accurately evaluate the risk value through a multi-scale constraint algorithm, and then use the Bi-FPM block in combination with the risk value and Bayesian theorem to calculate the recommended evaluation value, thereby improving the accuracy and reliability of the evaluation results.

[0054] The embodiment of the present invention provides a method for calculating the recommended evaluation value of an instrument by combining an instrument evaluation model with an instrument risk value and Bayesian theorem. Figure 5 The schematic diagram of the implementation flow of the method for calculating the recommended evaluation value of the device by combining the device evaluation model with the device risk value and the Bayesian theorem is shown. The method for calculating the recommended evaluation value of the device by combining the device evaluation model with the device risk value and the Bayesian theorem specifically includes: Step S401, loading a normalized data set, reconstructing the normalized data set by an autoencoder, extracting the core features of feature parameters in the normalized data set, and reducing the dimensions of the core features of the feature parameters into low-dimensional feature codes; In an embodiment of the present invention, when the core features of the feature parameters are reduced to low-dimensional feature codes, a convolutional layer, a maximum pooling layer, and an average pooling layer can be used to construct an encoder to extract local features of the data. The CBAM attention mechanism is introduced into the encoder to enhance the model's attention to key features. A fully connected layer is used to reduce the core features of the feature parameters to low-dimensional feature codes.

[0055] Step S402, the decoder obtains low-dimensional feature codes, perturbs and decodes the low-dimensional feature codes based on an adversarial perturbation algorithm, and obtains an autoencoder set through decoding; It should be noted that the decoder is constructed using deconvolution layers and fully connected layers. In the decoding process, an adversarial perturbation algorithm (such as FreeLB) is introduced to perturb the low-dimensional feature encoding. The perturbation parameters are adjusted using the adaptive moment estimation (Adam) optimizer. The low-dimensional feature encoding after adversarial perturbation contains richer information and stronger anti-interference ability. The decoder decodes it and obtains an autoencoder set. This set not only contains the main features of the original data, but also has certain robustness and generalization ability, providing a better foundation for subsequent feature extraction. In the training process, in addition to the normal loss function, an adversarial loss function is added. The goal of the adversarial loss function is to make the generated data have a certain difference from the original data, but not completely deviate from the true distribution of the original data. By optimizing these two loss functions, the model can learn how to accurately reconstruct the data in the presence of adversarial perturbations. For example, the Flipout method can be used, which is an adversarial training method based on stochastic gradient descent, which introduces perturbations by randomly flipping the output of some neurons during the encoding process.

[0056] Step S403, the feature extraction block identifies the autoencoder set, extracts local features of the autoencoder set based on the cross attention mechanism and batch normalization layer stacking, and outputs a multi-level feature set; In this embodiment, the batch normalization layer can accelerate the training process of the model and improve the stability and generalization ability of the model. By stacking multiple batch normalization layers, the model can normalize the data at different levels to further optimize the effect of feature extraction. For example, when processing a large amount of dental instrument data, the batch normalization layer can help the model converge faster and reduce the risk of overfitting. In a neural network, the batch normalization layer is usually placed after a convolutional layer or a fully connected layer. During the training process, for each small batch of data, the batch normalization layer calculates the mean and variance of the batch data and normalizes the input data. By stacking multiple batch normalization layers, the data can be normalized at different levels to further improve the performance of the model. For example, a smaller batch normalization layer can be used in a shallow network to process local features, and a larger batch normalization layer can be used in a deep network to process global features.

[0057] Step S404: The feature fusion block uses a non-local attention layer to enhance the long-distance dependency of the multi-level feature set based on a late fusion strategy, and uses an SPP layer to extract multi-scale features. The multi-scale features of different scales are fused through the SC block to generate a final fused feature set. In this embodiment, the SPP layer (spatial pyramid pooling layer) can process features of different scales and extract multi-scale features. This can make the model better adapt to input data of different sizes or resolutions and improve the generalization ability of the model. For example, for dental instruments of different models or specifications, their features may differ in spatial scale, and the SPP layer can effectively handle such differences. SC block fuses multi-scale features: The SC block (channel shuffle block) can fuse multi-scale features of different scales to form a unified feature representation. In this way, the model can integrate information from different scales to obtain a more comprehensive and accurate feature description. For example, fusing large-scale overall features and small-scale local detail features can provide a richer basis for subsequent evaluation.

[0058] Step S405, the evaluation classifier calculates the device risk value of the dental device to be selected based on the multi-scale constraint algorithm, and calculates the device recommendation evaluation value by combining the device risk value and Bayesian theorem through the Bi-FPM block.

[0059] In this embodiment, the device recommendation evaluation value is calculated by the following formula: (11) (12) in, Respectively represent the recommended assessment value of the device and the risk value of the device. are the number of characteristic parameters and the number of dental instruments to be selected, are the usage priority and likelihood probability respectively, is the fused feature set and the variance of the fused feature set.

[0060] In the embodiment of the present invention, in the evaluation of dental instruments, it is not enough to consider only a single factor, and it is necessary to comprehensively consider risk factors in multiple aspects. By calculating various risk values ​​of the selected dental instruments, such as selection risk, safety risk, quality risk, life risk, economic risk and applicability risk, the advantages and disadvantages of the instruments can be comprehensively evaluated, providing medical workers with a more accurate basis for decision-making. For example, when selecting dental instruments, it is necessary to consider not only their therapeutic effects (selection risks), but also their safety during use (safety risks), whether the product quality is reliable (quality risks), whether the service life meets the needs (life risk), whether the purchase and use costs are reasonable (economic risks), and whether they are suitable for the specific circumstances of the patient (applicability risks).

[0061] In summary, the present invention provides a medical device management method for dental treatment. In the embodiment of the present invention, when determining the use priority in the use demand information based on the Bayesian theorem, not only the idle rate and use frequency of the medical device are considered, but also multi-dimensional information such as the patient's treatment time and disease priority is introduced. Through the Bayesian theorem, this information is integrated to dynamically update the use priority of the medical device, and through the Gumbel-Copula function and the particle swarm optimization algorithm, the correlation between the medical device management data is deeply mined. It can make more comprehensive use of data information and provide more accurate support for management decisions. It can further more comprehensively reflect the actual needs and use scenarios of medical devices, ensure the accuracy of the recommended evaluation value of the device, and overcome the existing method of sorting the use priority of similar medical devices, only based on the idle rate of the device and the number of abnormal reminders to calculate the priority, ignoring the specific needs and treatment scenarios of the patient, and the priority is determined statically, and the priority of the medical device cannot be dynamically adjusted according to real-time data.

[0062] It should be noted that, for the above-mentioned embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, some steps may be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0063] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also belong to the scope of protection of the present invention.

Claims

1. A method for managing medical equipment for dental treatment, characterized in that: include: Obtaining dental equipment usage demand information, identifying the equipment name and management identification code based on the usage demand information, and using the management identification code to trigger equipment usage instructions; In response to the device use instruction, indexing the management-related data of the dental device based on the device use instruction, loading the use demand information, determining the use priority in the use demand information based on the Bayesian theorem, using the use priority as a weight factor, and performing normalization processing in combination with the management-related data to obtain a normalized data set; Load the normalized data set, encode and reconstruct the normalized data set based on the pre-trained device evaluation model and decode it to obtain an auto-encoded set, extract and fuse the features of the auto-encoded set, calculate the device risk value of the dental device to be selected, and the device evaluation model combines the device risk value and Bayesian theorem to calculate the device recommendation evaluation value; An instrument usage queue is generated based on the instrument recommendation evaluation value, and dental instruments in the instrument usage queue that exceed the recommendation threshold are pushed.

2. The medical device management method for dental treatment according to claim 1, characterized in that: The method for determining the use priority in the use demand information based on the Bayesian theorem includes: Load the usage demand information, identify the patient treatment time, patient disease priority, device name, and management identification code in the usage demand information, index the device idle rate and usage frequency corresponding to the device name based on the management identification code, and use the patient treatment time as the initial probability of use; Load the initial probability of use and the patient's disease priority, and correct the initial probability of use based on the particle swarm optimization algorithm combined with the patient's disease priority to obtain the corrected probability of use; （1） （2） （3） in, Respectively After iterations, use the corrected probability and the initial probability. are the current iteration number and the total iteration number, respectively. is the adaptive acceleration factor based on the particle swarm optimization algorithm, Indicates the patient's disease priority, Particle swarm optimization algorithm The optimization iteration value, The optimization iteration value, represents the search for optimal uniform random numbers, Respectively The individual optimal extreme position of each particle and the global optimal extreme position; Obtain the device idle rate and usage frequency corresponding to the device name, and calculate the likelihood probability based on the patient's disease priority and the device idle rate and usage frequency; （4） （5） in, represents the likelihood probability, For the The weight coefficient of each dental instrument, For the The initial weight coefficient of each dental instrument, They are the equipment idle rate and usage frequency. is the number of dental instruments, for The average value of the initial weight coefficients of the dental instruments; Load the idle rate and usage frequency of the instruments, and take into account the Gumbel-Copula function to calculate the total probability of dental instruments; The corrected probability, likelihood probability, and total probability of use of dental instruments are obtained, and the use priority in the use demand information is determined based on the Bayesian theorem.

3. The medical device management method for dental treatment according to claim 2, characterized in that: The total probability of dental instruments is calculated by the following formula: （6） in, represents the total probability of dental instruments, is the correlation parameter of the Gumbel-Copula function, They are the equipment idle rate and usage frequency; （7） in, Indicates the usage priority in the usage requirement information.

4. The medical device management method for dental treatment according to claim 1, characterized in that: The method for normalizing the management-related data includes: Load management-related data and usage priority, where management-related data includes device idle rate, usage frequency, device operating parameters, number of adverse events, type of adverse events, injury level, risk level, usage period, and maintenance records. Use the maximum-minimum method to standardize and preprocess the management-related data; Obtain the management-related data after standardization preprocessing, read and identify the management-related data based on the Pandas library, calculate the parameter correlation coefficient of the characteristic parameters in the management-related data using the Minkowski distance, and determine the average correlation coefficient of the characteristic parameters in the management-related data based on the parameter correlation coefficient; The parameter correlation coefficient of the characteristic parameters is calculated by the following formula: （8） in, Indicates the current feature parameters Associated feature parameters The parameter correlation coefficient, Indicates the associated feature parameters The mean of is the adjustment parameter of the Minkowski distance algorithm; Based on the preset dynamic threshold of Davidson-Botting Index, the dynamic threshold of association is used to filter the characteristic parameters in the management association data that meet the requirements, and obtain the parameter screening set; Load the parameter filter set and usage priority, use the usage priority as the weight factor, and calculate the correlation matrix of the parameter filter set and usage priority based on the Pandas algorithm combined with the Pearson coefficient; The parameter filter set and the correlation matrix of the usage priority are loaded, the parameter filter set and the correlation matrix of the usage priority are mapped to the parameter filter set based on the dynamic time warping algorithm, and the parameter filter set is set as a normalized data set.

5. The medical device management method for dental treatment according to claim 4, characterized in that: The associated dynamic threshold is calculated by the following formula: （9） （10） in, represents the associated dynamic threshold, The Davidson-Botting index for managing feature parameters in linked data, is the number of characteristic parameters, are the current feature parameters Associated feature parameters The parameter correlation coefficient and the average correlation coefficient of the characteristic parameters are They are the current feature parameter cluster center and the associated feature parameter cluster center respectively. Indicates the Euclidean distance between the current feature parameter cluster center and the associated feature parameter cluster center.

6. The medical device management method for dental treatment according to claim 4, characterized in that: The device evaluation model training method comprises: The modeling data set is crawled based on crawler technology, and the crawled modeling data set is normalized. The normalized modeling data set is divided into a training set and a test set in a ratio of 3:

1. Load the pre-built device evaluation model, initialize the initial weights and thresholds of the device evaluation model, and set the activation function, loss function, and training rounds of the device evaluation model; The equipment evaluation model is iteratively trained using the training set. During the training, the optimal values ​​of the initial weights and thresholds of the equipment evaluation model are searched in combination with the sparrow algorithm. The initial weights and thresholds are adjusted based on the convergence and alienation operations of the sparrow population, and the converged equipment evaluation model is output; Load the test set, use the test set as input, solve the device risk value and recommended assessment value, and calculate the comprehensive error of the device risk value and recommended assessment value; Determine whether the comprehensive error meets the preset error threshold. If the comprehensive error meets the preset error threshold, output the trained device evaluation model.

7. The medical device management method for dental treatment according to claim 6, characterized in that: The device evaluation model uses a BP neural network model as an initial model, the initial model includes an input layer, a hidden layer and an output layer, an autoencoder module is introduced between the input layer and the hidden layer, and the hidden layer is frozen, an improved CNN architecture is used to replace the hidden layer, and an evaluation classifier is introduced between the improved CNN architecture and the output layer; The input layer is provided with three layers, the autoencoder module is composed of an encoder and a decoder, the encoder and the decoder are connected through a convolution layer, the autoencoder includes three groups of 3×3 convolution blocks, a maximum pooling layer, an average pooling layer and a CBAM layer, and the CBAM layer introduces a CBAM attention mechanism, and the decoder includes two groups of 3×3 convolution layers, two groups of 3×3 deconvolution layers, a FreeLB layer and a fully connected layer; The improved CNN architecture consists of a feature extraction block, a fully connected layer, and a feature fusion block connected in sequence. A cross attention mechanism is introduced in the feature extraction block, and an activation function Sigmoid is defined. The feature extraction block is used to identify the autoencoder set, extract the local features of the autoencoder set based on the cross attention mechanism and the batch normalization layer stack, and output a multi-level feature set. The feature fusion block consists of an upsampling layer, a non-local attention layer, an SPP layer, a batch normalization layer, and an SC block. The evaluation classifier includes an LSTM layer and a Bi-FPM block, and introduces a multi-scale constraint algorithm. The evaluation classifier calculates the instrument risk value of the dental instrument to be selected based on the multi-scale constraint algorithm, and calculates the instrument recommendation evaluation value through the Bi-FPM block in combination with the instrument risk value and Bayesian theorem.

8. The medical device management method for dental treatment according to claim 7, characterized in that: The device assessment model combines the device risk value and the Bayesian theorem to calculate the recommended device assessment value, including: Load the normalized data set, reconstruct the normalized data set encoding by the autoencoder, extract the core features of the feature parameters in the normalized data set, and reduce the core features of the feature parameters to low-dimensional feature encoding; The decoder obtains low-dimensional feature codes, perturbs and decodes the low-dimensional feature codes based on an adversarial perturbation algorithm, and obtains an autoencoder set through decoding. The feature extraction block identifies the autoencoder set, extracts local features of the autoencoder set based on the cross-attention mechanism and batch normalization layer stacking, and outputs a multi-level feature set; The feature fusion block is based on the late fusion strategy, uses the non-local attention layer to enhance the long-distance dependencies of the multi-level feature set, and uses the SPP layer to extract multi-scale features. The multi-scale features of different scales are fused through the SC block to generate the final fused feature set; The evaluation classifier calculates the device risk value of the dental device to be selected based on a multi-scale constraint algorithm, and calculates the device recommendation evaluation value by combining the device risk value and Bayesian theorem through the Bi-FPM block.

9. The medical device management method for dental treatment according to claim 8, characterized in that: The recommended evaluation value of the device is calculated by the following formula: （11） （12） in, Respectively represent the recommended assessment value of the device and the risk value of the device. are the number of characteristic parameters and the number of dental instruments to be selected, are the usage priority and likelihood probability respectively, is the fused feature set and the variance of the fused feature set.

Images