Implant resonant frequency measurement analysis method and system based on deep learning

Through deep learning technology, the noise in the resonance frequency of the implant is removed by using a convolutional neural network and analyzing bone density and implantation torque, solving the problem of comprehensive analysis of noise interference and parameters in the existing technology, and improving the accuracy of the resonance frequency analysis of the implant Sex and robustness.

CN119089135APending Publication Date: 2024-12-06LIYANG PEOPLES HOSPITAL
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411113020.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-14
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The existing implant resonance frequency analysis technology is difficult to effectively remove multiple noise interferences, and it is impossible to comprehensively consider the impact of bone density and implantation torque on resonance frequency, resulting in large errors in the analysis results.

Method used

Using a deep learning-based method, a denoising neural network is constructed and analytical neural network is analyzed through convolutional neural networks, a resonance analysis model is established, and analytical neural network is combined with analytical neural network for series connection, removing noise and analyzing the resonance frequency of the implant.

Benefits of technology

It realizes the accurate analysis of the resonance frequency of the implant while removing multiple noises, improves the accuracy and robustness of the analysis results, and can comprehensively consider the impact of bone density and implant torque on the resonance frequency of the implant.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119089135A_ABST
    Figure CN119089135A_ABST
Patent Text Reader

Abstract

The invention discloses an implant resonance frequency measurement and analysis method and system based on deep learning, and relates to the technical field of implant resonance frequency analysis. Comprising the following steps: collecting resonant frequency data of an implant and associated data of the implant for preprocessing, and constructing a first data set; constructing an initial denoising network and an initial analysis network based on the convolutional neural network, performing optimization training on the initial denoising network to obtain a denoising neural network, and constructing a second data set; performing optimization training on the initial analysis network to obtain an analysis neural network; constructing a resonance analysis model based on the denoising neural network and the analysis neural network, and analyzing the resonance frequency by using the resonance analysis model; the method is used for solving the problem that the resonance frequency of the implant cannot be analyzed by integrating the bone mineral density and the implantation torque of the implantation part while various noises contained in the resonance frequency of the implant cannot be simply and effectively removed in the prior art.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of implant resonance frequency analysis, and specifically to an implant resonance frequency measurement and analysis method and system based on deep learning. Background Art

[0002] In the field of oral implantology, the stability of the implant is one of the key factors for successful treatment. Resonance frequency analysis technology, as an important means to evaluate the stability of implants, has the advantages of simple operation, high sensitivity and good repeatability. It evaluates the integration of the implant and the surrounding bone by measuring the resonance frequency generated by the vibration excited by the magnetic field or electric field. It is non-invasive and highly sensitive, and can continuously monitor the stability of the implant at different time points without affecting the integration process between the implant and the bone. However, it also has certain limitations, which is mainly because the resonance frequency is easily affected by the surrounding environmental factors, which makes the resonance frequency measurement of the implant mixed with interference factors, resulting in errors in the results.

[0003] The existing resonance frequency analysis of implants usually measures the resonance frequency of the implant through a resonance frequency measuring instrument, and then converts the resonance frequency into an analysis result through a fixed algorithm in the instrument. Multiple measurements are performed by changing the measurement position to obtain multiple analysis results, and then the average is taken to reduce the error. However, the resonance frequency of the implant is easily affected by the surrounding environmental factors and is easily mixed with various interference noises, such as environmental noise and oral noise. In the case of noise interference, it is difficult to eliminate or reduce the error by taking the average of multiple measurements, and the ability of traditional algorithms to remove multiple noises at one time is limited. At the same time, the resonance frequency is also related to the bone density of the implant implantation site and the implantation torque of the implant, which should also be taken into account when analyzing the resonance frequency of the implant. Therefore, the existing technology cannot simply and effectively remove the various noises contained in the resonance frequency of the implant, and at the same time, comprehensively analyze the resonance frequency of the implant based on the bone density of the implant implantation site and the implantation torque. Summary of the invention

[0004] The present invention aims to solve at least one of the technical problems in the prior art to a certain extent, and proposes a method and system for measuring and analyzing the resonance frequency of an implant based on deep learning. A denoising neural network and an analysis neural network are established based on a convolutional neural network, and then a resonance analysis model is established, which can achieve accurate analysis of the resonance frequency of the implant; so as to solve the problem that the prior art cannot simply and effectively remove the various noises contained in the resonance frequency of the implant, and at the same time analyze the resonance frequency of the implant by comprehensively considering the bone density of the implant site and the implant torque.

[0005] To achieve the above objectives, in the first aspect, a method for measuring and analyzing implant resonance frequency based on deep learning includes:

[0006] Collecting the resonance frequency data of the implant and the implant-related data, and performing preprocessing to construct a first data set;

[0007] An initial denoising network is constructed based on a convolutional neural network, and the initial denoising network is optimized and trained based on the first data set to obtain a denoising neural network, and a second data set is constructed;

[0008] constructing an initial analysis network based on a convolutional neural network, and optimizing and training the initial analysis network based on a second data set to obtain an analysis neural network;

[0009] A resonance analysis model is constructed based on the denoising neural network and the analytical neural network, and the resonance frequency of the implant is analyzed using the resonance analysis model.

[0010] Furthermore, collecting the resonance frequency data of the implant and the implant-related data includes:

[0011] Collect the implant's resonance frequency: Use a resonance frequency measurement device to obtain the implant's resonance frequency data, marked as F i ; i represents the i-th implant;

[0012] Collect implant-related data: Obtain bone density data at the implant site, marked as BMD i ; Obtain implantation torque data, marked as IT i , using implant stability measuring instrument to measure the ISQ data at the initial stage of implant placement, marked as ISQ i ;

[0013] All data are stored according to the implants they belong to, including resonance frequency data, bone density data of the implant placement site, implant implantation torque data and ISQ data at the initial stage of implantation for each implant.

[0014] Further, preprocessing the resonance frequency data of the implant and the implant-related data to construct the first data set includes:

[0015] Preprocess the resonance frequency data of the implant: normalize the resonance frequency data of the implant, and perform time domain segmentation on the vibration signal in the normalized resonance frequency data of the implant. The time domain segmentation includes: calculating the description length L based on the sampling frequency of the resonance frequency measuring instrument and the resonance period of the implant. min , based on the description length L min Set the sliding step size, collect data in the form of a sliding window, complete time domain segmentation, and obtain N segmentation samples;

[0016] Preprocess the implant-related data: normalize the bone density data of the implant placement site and the implant insertion torque data using the min-max normalization method according to the data category;

[0017] The preprocessed resonance frequency data of the implant is stored and marked as a first data set, and the first data set is divided into a first training set and a first test set in a ratio of 8:2.

[0018] Furthermore, constructing an initial denoising network based on a convolutional neural network includes:

[0019] Based on the convolutional neural network, improvements are made by adding a batch normalization layer BN between the convolutional layer and the ReLU activation function in the convolutional neural network structure; after the improvement is completed, an initial denoising network is obtained.

[0020] Further, the initial denoising network is optimized and trained based on the first data set to obtain a denoising neural network, and the second data set is constructed, including:

[0021] Set the training hyperparameters of the initial denoising network, set the learning rate to E1, the batch size to B1, and the number of model training rounds to P1;

[0022] According to the set initial denoising network training hyperparameters, the first data set is used to train and optimize the initial denoising network, the WO optimization algorithm is introduced during the training process to optimize the initial denoising network training process, the weight parameters in the initial denoising network are encoded as the position parameters in the WO optimization algorithm, the loss function of the initial denoising network training is used as the objective function of the WO algorithm, and the optimal weight parameters of the initial denoising network are obtained by finding the position parameters with the minimum loss function;

[0023] The optimal weight parameters are updated to the initial denoising network, and the initial denoising network with the updated optimal weight parameters is trained with the first training set to obtain a denoising neural network;

[0024] The first data set was denoised using a denoising neural network, and the denoised first data set and implant-related data were merged and classified and stored according to the implants to which they belonged. Each implant included resonance frequency data, bone density data at the implant implantation site, implant implantation torque data, and ISQ data at the initial stage of implantation. After merging, it was marked as the second data set, and the second data set was divided into a second training set and a second test set in a ratio of 8:2.

[0025] Furthermore, constructing an initial analysis network based on a convolutional neural network includes:

[0026] Improvements are made based on convolutional neural networks: the convolution kernel and pooling kernel in the convolutional neural network are changed to a one-dimensional structure; the size of the convolution kernel of the first convolution layer in the convolutional neural network is increased to x*y; a batch normalization layer is added between the convolution layer and the ReLU activation function in the convolutional neural network, and a Dropout operation is added after the convolution layer in the convolutional neural network. The improvement is completed to obtain the initial analysis network.

[0027] Further, the initial analysis network is optimized and trained based on the second data set, and the analysis neural network obtained includes:

[0028] Initialize the network structure parameters of the initial analysis network; set the initial analysis network training hyperparameters, set the learning rate to E2; set the batch size Batch_size to B2; set the number of model training rounds to P2 rounds;

[0029] According to the set initial analysis network training hyperparameters, the initial analysis network is trained using the second training set. After the training is completed, the initial analysis network is tested using the second test set to obtain the test results, and the loss function is used to determine whether the model loss value is qualified. If the model loss value is qualified, the training is completed to obtain the analysis neural network. If the model loss value is unqualified, the network structure parameters of the initial analysis network are updated through back propagation, and the initial analysis network is trained again using the second training set. After the training is completed, the initial analysis network is tested using the second test set to obtain the test results, and the loss function is used to determine whether the model loss value is qualified. The training is stopped until the model loss value is qualified to obtain the analysis neural network.

[0030] Furthermore, constructing a resonance analysis model based on the denoising neural network and the analysis neural network includes:

[0031] The denoising neural network and the analysis neural network are connected in series in sequence. The collected resonance frequency data is first processed by the denoising neural network and then output to the analysis neural network for processing. The denoising neural network and the analysis neural network connected in series are marked as a resonance analysis model.

[0032] Furthermore, the resonance frequency of the implant is analyzed using the resonance analysis model including:

[0033] Use a resonance frequency measurement device to obtain the resonance frequency data of the implant, obtain the implant-related data of the implant, including: the bone density data of the implant implantation site and the implant implantation torque data, normalize the resonance frequency data, the bone density data of the implant implantation site and the implant implantation torque data, then input the resonance frequency data, the bone density data of the implant implantation site and the implant implantation torque data into the resonance analysis model, and output the analysis results.

[0034] In a second aspect, the present invention provides an implant resonance frequency measurement and analysis system based on deep learning, comprising a data acquisition module, a denoising network module, an analysis network module and a resonance analysis module;

[0035] The data acquisition module includes a collection unit and a processing unit, the collection unit is used to collect the resonance frequency data of the implant and the implant-related data; the processing unit is used to pre-process the resonance frequency data of the implant and the implant-related data, and construct a first data set;

[0036] The denoising network module includes a denoising construction unit and a denoising training unit. The denoising construction unit constructs an initial denoising network based on a convolutional neural network. The denoising training unit performs optimization training based on the initial denoising network of the first data set to obtain a denoising neural network, and uses the denoising neural network to construct a second data set.

[0037] The analysis network module includes an analysis construction unit and an analysis training unit. The analysis construction unit constructs an initial analysis network based on a convolutional neural network. The analysis training unit optimizes and trains the initial analysis network based on a second data set to obtain an analysis neural network.

[0038] The resonance analysis module includes a model construction unit and a frequency analysis unit. The model construction unit constructs a resonance analysis model based on a denoising neural network and an analysis neural network. The frequency analysis unit includes a resonance analysis model. The resonance analysis model is used to analyze the resonance frequency of the implant.

[0039] The beneficial effects of the present invention are as follows: the present invention first collects the resonance frequency data of the implant and the implant-related data, and performs preprocessing to construct a first data set; constructs an initial denoising network based on a convolutional neural network, optimizes and trains the initial denoising network based on the first data set to obtain a denoising neural network, and constructs a second data set; constructs an initial analysis network based on a convolutional neural network, optimizes and trains the initial analysis network based on the second data set to obtain an analysis neural network; constructs a resonance analysis model based on the denoising neural network and the analysis neural network, and uses the resonance analysis model to analyze the resonance frequency of the implant; the present invention establishes a resonance analysis model based on deep learning technology, which can simply and effectively remove various noises contained in the resonance frequency of the implant, and can comprehensively analyze the resonance frequency of the implant by considering the bone density of the implant site and the implant torque, and output the analysis results.

[0040] The present invention establishes an initial denoising network based on a convolutional neural network, can use the convolutional neural network to automatically learn noise features, and effectively remove noise of various types and intensities, thereby reducing analysis errors. Through training, the convolutional neural network can quickly and effectively remove noise while maintaining detailed information of the resonance frequency, thereby ensuring analysis efficiency. Moreover, the WO optimization algorithm trains and optimizes the initial denoising network, thereby improving model training efficiency. Based on the convolutional neural network, an analysis neural network is established, and then a resonance analysis model is established through the analysis neural network and the denoising neural network. The convolutional neural network can be used to learn and extract useful features from the resonance frequency, and good analysis performance can be maintained even when data quality is not high, thereby improving robustness. Furthermore, the resonance analysis model can analyze the resonance frequency of the implant by comprehensively considering the bone density of the implant site and the implant torque, thereby further reducing analysis errors and improving accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a functional block diagram of the system of the present invention;

[0042] Figure 2 is a flow chart of the steps of the method of the present invention;

[0043] Figure 3 It is a schematic diagram of the denoising neural network structure of the present invention;

[0044] Figure 4 It is a schematic diagram of the Dropout operation of the present invention;

[0045] Figure 5 It is a schematic diagram of the structure of the resonance analysis model of the present invention. DETAILED DESCRIPTION

[0046] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0047] Example 1, please refer to Figure 1 As shown, the deep learning-based implant resonance frequency measurement and analysis system includes a data acquisition module, a denoising network module, an analysis network module, and a resonance analysis module;

[0048] The data acquisition module includes a collection unit and a processing unit, the collection unit is used to collect the resonance frequency data of the implant and the implant-related data; the processing unit is used to pre-process the resonance frequency data of the implant and the implant-related data, and construct a first data set;

[0049] The acquisition unit is configured with an acquisition strategy, which includes: collecting the resonance frequency of the implant: using a resonance frequency measurement device to obtain the resonance frequency data of the implant, marked as F i ; i represents the i-th implant;

[0050] Collect implant-related data: Obtain bone density data at the implant site, marked as BMD i ; Obtain implantation torque data, marked as IT i , using implant stability measuring instrument to measure the ISQ data at the initial stage of implant placement, marked as ISQ i ;ISQ data refers to the "implant stability quotient" obtained using an implant stability measuring instrument, such as the Osstell instrument. ISQ is a unit of measurement used to evaluate the stability of dental implants. This measurement technology is based on resonant frequency analysis and predicts the possibility of long-term success by measuring the stability of the implant's integration with the bone.

[0051] The processing unit is configured with a processing strategy, which includes: normalizing the resonance frequency data of the implant, and performing time domain segmentation on the vibration signal in the resonance frequency data of the implant after normalization. The time domain segmentation includes: calculating the description length L based on the sampling frequency of the resonance frequency measuring instrument and the resonance period of the implant. min , describing the length L min is the minimum length that can describe the implant status. In this embodiment, L min =T0*Fs, where Fs is the sampling frequency of the resonance frequency measuring instrument, T0 is the implant resonance period; based on the description length L min Set the sliding step size. In this embodiment, the sliding step size W, L min and resonance frequency data length N min The three must satisfy W≤L min ≤N min ≤; W≤Lmin means that the window is overlapped and moved, which can avoid sample waste and ensure the generation of a sufficient number of samples; Lmin≤Nmin means the number of points in at least one vibration cycle of the experimental sample, ensuring that the segmented samples contain sufficient vibration information; data collection is performed in the form of a sliding window, time domain segmentation is completed, and N segmented samples are obtained;

[0052] The bone density data of the implant implantation site and the implant implantation torque data were normalized using the min-max normalization method according to the data category; the size of the data of the same type was scaled to [0, 1];

[0053] The min-max normalization formula is y i =x i -x minWhere x max is the maximum value in the data, x min for

[0054] x max -x min

[0055] The minimum value in the data, x i is the original value of the data, y i is the normalized value;

[0056] In this embodiment, the resonance frequency data of the implant is normalized, mainly the amplitude in the resonance frequency data is normalized, and the min-max normalization method is also used;

[0057] The preprocessed implant resonance frequency data set is stored and marked as a first data set, and the first data set is divided into a first training set and a first test set in a ratio of 8:2;

[0058] In the specific implementation process, the resonance frequency of the implant reflects the stability of the implant, that is, the ISQ value. The bone density data of the implant implantation site and the implant implantation torque data have a certain relationship with the resonance frequency of the implant. Comprehensively considering the bone density data of the implant implantation site and the implant implantation torque data to analyze the resonance frequency of the implant can improve the accuracy of the analysis. In oral implantology, the ISQ value is usually between 1 and 100, and the higher the value, the more stable the implant. This data is widely used in clinical practice to guide decision-making and monitor the healing process of implants.

[0059] The denoising network module includes a denoising construction unit and a denoising training unit. The denoising construction unit constructs an initial denoising network based on a convolutional neural network.

[0060] See also Figure 3 As shown, the denoising construction unit is configured with a denoising construction strategy, and the denoising construction strategy includes: improving based on the convolutional neural network, adding a batch normalization layer BN between the convolution layer and the ReLU activation function in the convolutional neural network structure; after the improvement is completed, an initial denoising network is obtained;

[0061] The denoising training unit optimizes and trains the initial denoising network of the first data set to obtain a denoising neural network, and constructs a second data set using the denoising neural network, including: setting the training hyperparameters of the initial denoising network, setting the learning rate to E1; setting the batch size to B1; setting the number of model training rounds to P1 rounds; in this embodiment, E1=0.001, B1=16, P1=100;

[0062] According to the set initial denoising network training hyperparameters, the initial denoising network is trained and optimized using the first data set. The WO optimization algorithm is introduced during the training process to optimize the initial denoising network training process. The weight parameters in the initial denoising network are encoded as position parameters in the WO optimization algorithm. In this embodiment, encoding means encoding each weight parameter vector in the neural network as an individual position in the optimization algorithm, and the specific representation is U ij =[ω 14 ,ω 15 ,ω 16 ,ω ij ,……,ω 24 ,ω 25 ,ω 26 ,……], where ω ij Represents the weight parameters in the initial denoising network;

[0063] The loss function of the initial denoising network training is used as the objective function of the WO algorithm, and the optimal weight parameters of the initial denoising network are obtained by finding the position parameters with the minimum loss function; selecting a suitable objective function can improve the convergence speed of the algorithm and expand the search range to obtain the global optimal solution; in this embodiment, the objective function is: K loss is the objective function, P is the number of samples at the training time; Q is the number of neurons in the output layer; is the predicted ideal value of the jth output node in the i-th sample, O j,i is the actual value of the jth output node in the ith sample; the optimal weight parameters are updated to the initial denoising network, and the initial denoising network with updated optimal weight parameters is trained with the first training set to obtain a denoising neural network;

[0064] The first data set is denoised using a denoising neural network, and the denoised first data set and implant-related data are merged and stored according to the implants to which they belong, including resonance frequency data, bone density data of the implant implantation site, implant implantation torque data, and ISQ data at the initial stage of implantation for each implant; the merged data set is marked as the second data set, and the second data set is divided into a second training set and a second test set in a ratio of 8:2;

[0065] In the specific implementation process, the denoising neural network introduced the residual learning method to learn the noise characteristics in the noisy signal; the amplitude spectrum of the noisy vibration signal and the time domain waveform of the noise signal were used as training features and training targets for training and testing respectively, and the denoised vibration signal was obtained by difference; the denoising neural network first used the convolution layer and the ReLU activation function to extract the features of the input signal; secondly, it was input into a parallel module composed of a convolution layer, a BN layer and a ReLU activation function, and the features were extracted from large to small through convolution kernels of different sizes. Finally, a fully connected layer was used for feature fusion, and the noise signal was output, and the denoised vibration signal was obtained by difference.

[0066] The analysis network module includes an analysis construction unit and an analysis training unit, and the analysis construction unit constructs an initial analysis network based on a convolutional neural network;

[0067] The analysis construction unit is configured with an analysis construction strategy, which includes: improvement based on the convolutional neural network: changing the convolution kernel and pooling kernel in the convolutional neural network to a one-dimensional structure; because the input layer is one-dimensional original data segmented in the time domain, in order to adapt to the characteristics of the one-dimensional vibration time series, the size of the convolution kernel of the first convolution layer in the convolutional neural network is increased to x*y, and a large convolution kernel is more conducive to suppressing high-frequency noise and further improving the analysis accuracy. In this embodiment, the convolution kernel size of the first convolution layer is 32*1, and the step size is 8*1. A batch normalization layer is added between the convolution layer and the ReLU activation function in the convolutional neural network to enhance the generalization ability; a Dropout operation is added after the convolution layer in the convolutional neural network, please refer to Figure 4 As shown in the figure, the Dropout operation will randomly drop a part of neurons according to the set probability during the training process. Only the weight information of these neurons is retained. The robustness of the network is improved through multiple batches of iterations. The initial analysis network is obtained after the improvement.

[0068] The analysis training unit optimizes and trains the initial analysis network based on the second data set to obtain the analysis neural network, including: initializing the network structure parameters of the initial analysis network; setting the initial analysis network training hyperparameters, setting the learning rate to E2; setting the batch size Batch_size to B2; setting the number of model training rounds to P2 rounds; in this embodiment, E2=0.001, B2=16, P2=100;

[0069] According to the set initial analysis network training hyperparameters, the initial analysis network is trained using the second training set. After the training is completed, the initial analysis network is tested using the second test set to obtain the test results, and the loss function is used to determine whether the model loss value is qualified. If the model loss value is qualified, the training is completed to obtain the analysis neural network. If the model loss value is unqualified, the network structure parameters of the initial analysis network are updated through back propagation, and the initial analysis network is trained again using the second training set. After the training is completed, the initial analysis network is tested using the second test set to obtain the test results, and the loss function is used to determine whether the model loss value is qualified, until the model loss value is qualified;

[0070] In the specific implementation process, the vibration signal is one-dimensional time series data, and the one-dimensional convolutional neural network has a strong ability to process one-dimensional sequences. The convolutional layer of the convolutional neural network can use different convolution kernels to perform convolution operations on the input along the direction of the sequence according to the step size, and then through an activation function, the corresponding feature map can be obtained; the role of the activation function is to introduce nonlinear factors into the neural network to enhance the nonlinear expression ability of the network; the role of the pooling layer is to compress the feature dimension to reduce the number of parameters of the network operation and reduce the possibility of overfitting; the role of the fully connected layer is to expand the features extracted by the previous layers into a one-dimensional vector.

[0071] The resonance analysis module includes a model building unit and a frequency analysis unit, and the model building unit builds a resonance analysis model based on a denoising neural network and an analysis neural network;

[0072] The analysis build unit is configured with an analysis build strategy, which includes: Figure 5 As shown, the denoising neural network and the analysis neural network are connected in series in sequence, the collected resonance frequency data is first processed by the denoising neural network and then output to the analysis neural network for processing, and the denoising neural network and the analysis neural network connected in series are marked as a resonance analysis model;

[0073] The frequency analysis unit includes a resonance analysis model, and the resonance analysis model is used to analyze the resonance frequency of the implant, including: using a resonance frequency measurement device to obtain the resonance frequency data of the implant, obtaining the implant-related data of the implant, including: the bone density data of the implant implantation site and the implant implantation torque data, normalizing the resonance frequency data, the bone density data of the implant implantation site and the implant implantation torque data, and then inputting the resonance frequency data, the bone density data of the implant implantation site and the implant implantation torque data into the resonance analysis model, and outputting the analysis result; the analysis result is ISQ data, which is used to evaluate the stability of the dental implant;

[0074] In the specific implementation process, the resonance frequency input into the resonance analysis model is first processed by a denoising neural network to remove the noise to obtain the denoised resonance frequency. The denoised resonance frequency is then input into the analysis neural network together with the implant-related data. After being processed and analyzed by the analysis neural network, the analysis results are output.

[0075] Example 2, please refer to Figure 2 As shown, the implant resonance frequency measurement and analysis method based on deep learning includes the following steps:

[0076] Step S1, collecting the resonance frequency data of the implant and the implant-related data, and performing preprocessing to construct a first data set; Step S1 includes the following sub-steps:

[0077] Step S101, collecting the resonance frequency of the implant: using a resonance frequency measurement device to obtain the resonance frequency data of the implant, marked as F i ; i represents the i-th implant;

[0078] Step S102, collect implant-related data: obtain the bone density data of the implant implantation site, marked as BMD i ; Obtain implantation torque data, marked as IT i , using implant stability measuring instrument to measure the ISQ data at the initial stage of implant placement, marked as ISQ i ;

[0079] Step S103, storing all data according to the implants to which they belong, including resonance frequency data, bone density data of the implant implantation site, implant implantation torque data, and ISQ data at the initial stage of implantation for each implant;

[0080] Step S104, normalizing the resonance frequency data of the implant;

[0081] Step S105, performing time domain segmentation on the vibration signal in the normalized resonance frequency data of the implant, the time domain segmentation includes: calculating the description length L based on the sampling frequency of the resonance frequency measuring instrument and the resonance period of the implant min , based on the description length L min Set the sliding step size, collect data in the form of a sliding window, complete time domain segmentation, and obtain N segmentation samples;

[0082] Step S106, normalizing the bone density data of the implant implantation site and the implant implantation torque data using the min-max normalization method according to the data categories;

[0083] Step S107 , storing the preprocessed resonance frequency data of the implant, marking it as a first data set, and dividing the first data set into a first training set and a first test set in a ratio of 8:2.

[0084] Step S2, constructing an initial denoising network based on a convolutional neural network, optimizing and training the initial denoising network based on the first data set to obtain a denoising neural network, and constructing a second data set; Step S2 also includes the following sub-steps:

[0085] Step S201, based on the convolutional neural network, an improvement is performed by adding a batch normalization layer BN between the convolutional layer and the ReLU activation function in the convolutional neural network structure; after the improvement is completed, an initial denoising network is obtained;

[0086] Step S202, setting the training hyperparameters of the initial denoising network, setting the learning rate to E1; setting the batch size to B1; and setting the number of model training rounds to P1 rounds;

[0087] Step S203, according to the set initial denoising network training hyperparameters, the first data set is used to train and optimize the initial denoising network, the WO optimization algorithm is introduced in the training process to optimize the initial denoising network training process, the weight parameters in the initial denoising network are encoded as the position parameters in the WO optimization algorithm, the loss function of the initial denoising network training is used as the objective function of the WO algorithm, and the optimal weight parameters of the initial denoising network are obtained by finding the position parameters with the minimum loss function;

[0088] Step S204, updating the optimal weight parameters to the initial denoising network, and using the first training set to train the initial denoising network with the updated optimal weight parameters to obtain a denoising neural network;

[0089] Step S205, denoising the first data set using a denoising neural network, merging the denoised first data set and implant-related data, and storing them by category according to the implants they belong to, wherein each implant includes resonance frequency data, bone density data of the implant implantation site, implant implantation torque data, and ISQ data at the initial stage of implantation;

[0090] Step S206, marking the merged data as a second data set, and dividing the second data set into a second training set and a second test set in a ratio of 8:2.

[0091] Step S3, constructing an initial analysis network based on a convolutional neural network, and optimizing and training the initial analysis network based on the second data set to obtain an analysis neural network; Step S3 also includes the following sub-steps:

[0092] Step S301, based on the convolutional neural network, improvements are made: the convolution kernel and the pooling kernel in the convolutional neural network are changed to a one-dimensional structure; the size of the convolution kernel of the first convolutional layer in the convolutional neural network is increased to x*y; a batch normalization layer is added between the convolutional layer and the ReLU activation function in the convolutional neural network, and a Dropout operation is added after the convolutional layer in the convolutional neural network, and the improvement is completed to obtain an initial analysis network;

[0093] Step S302, initializing the network structure parameters of the initial analysis network; setting the initial analysis network training hyperparameters, setting the learning rate to E2; setting the batch size Batch_size to B2; and setting the number of model training rounds to P2 rounds;

[0094] Step S303, according to the set initial analysis network training hyperparameters, the initial analysis network is trained using the second training set; after the training is completed, the initial analysis network is tested using the second test set to obtain the test results, and the loss function is used to determine whether the model loss value is qualified. If the model loss value is qualified, the training is completed to obtain the analysis neural network.

[0095] Step S304, if the model loss value is unqualified, the network structure parameters of the initial analysis network are updated through back propagation, and the initial analysis network is trained again using the second training set. After the training is completed, the initial analysis network is tested using the second test set to obtain the test results, and the loss function is used to determine whether the model loss value is qualified. The training is stopped until the model loss value is qualified to obtain the analysis neural network.

[0096] Step S4, constructing a resonance analysis model based on the denoising neural network and the analysis neural network, and analyzing the resonance frequency of the implant using the resonance analysis model; Step S4 also includes the following sub-steps:

[0097] Step S401, connecting the denoising neural network and the analysis neural network in series in sequence, the collected resonance frequency data is first processed by the denoising neural network and then output to the analysis neural network for processing, and the denoising neural network and the analysis neural network connected in series are marked as a resonance analysis model;

[0098] Step S402, using a resonance frequency measurement device to obtain the resonance frequency data of the implant, obtain the implant-related data of the implant, including: the bone density data of the implant implantation site and the implant implantation torque data, normalize the resonance frequency data, the bone density data of the implant implantation site and the implant implantation torque data, then input the resonance frequency data, the bone density data of the implant implantation site and the implant implantation torque data into a resonance analysis model, and output the analysis results.

[0099] Embodiment 3, third aspect, the present application provides a storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps in the above method are executed. Through the above technical scheme, when the computer program is executed by the processor, the method in any optional implementation mode of the above embodiment is executed to achieve the following functions: collecting the resonance frequency data of the implant and the implant-related data, and performing preprocessing to construct a first data set; constructing an initial denoising network based on a convolutional neural network, optimizing and training the initial denoising network based on the first data set to obtain a denoising neural network, and constructing a second data set; constructing an initial analysis network based on a convolutional neural network, optimizing and training the initial analysis network based on the second data set to obtain an analysis neural network; constructing a resonance analysis model based on the denoising neural network and the analysis neural network, and using the resonance analysis model to analyze the resonance frequency of the implant.

[0100] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0101] It should be understood by those skilled in the art that the embodiments of the present invention can be provided as methods, systems or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program codes. Among them, the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic memory, flash memory, disk or optical disk. These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0102] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

Claims

1. A method for measuring and analyzing implant resonance frequency based on deep learning, characterized in that: The steps include: Collecting the resonance frequency data of the implant and the implant-related data, and performing preprocessing to construct a first data set; An initial denoising network is constructed based on a convolutional neural network, and the initial denoising network is optimized and trained based on the first data set to obtain a denoising neural network, and a second data set is constructed; constructing an initial analysis network based on a convolutional neural network, and optimizing and training the initial analysis network based on a second data set to obtain an analysis neural network; A resonance analysis model is constructed based on the denoising neural network and the analytical neural network, and the resonance frequency of the implant is analyzed using the resonance analysis model.

2. The implant resonance frequency measurement and analysis method based on deep learning according to claim 1, characterized in that: Collecting implant resonance frequency data and implant-related data includes: Collect the implant's resonance frequency: Use a resonance frequency measurement device to obtain the implant's resonance frequency data, marked as F i ; i represents the i-th implant; Collect implant-related data: Obtain bone density data at the implant site, marked as BMD i ; Obtain implantation torque data, marked as IT i , using implant stability measuring instrument to measure the ISQ data at the initial stage of implant placement, marked as ISQ i ; All data are stored according to the implants they belong to, including resonance frequency data, bone density data of the implant placement site, implant implantation torque data and ISQ data at the initial stage of implantation for each implant.

3. The implant resonance frequency measurement and analysis method based on deep learning according to claim 2, characterized in that: Preprocessing the resonance frequency data of the implant and the implant-related data to construct the first data set includes: Preprocess the resonance frequency data of the implant: normalize the resonance frequency data of the implant, and perform time domain segmentation on the vibration signal in the normalized resonance frequency data of the implant. The time domain segmentation includes: calculating the description length L based on the sampling frequency of the resonance frequency measuring instrument and the resonance period of the implant. min , based on the description length L min Set the sliding step size, collect data in the form of a sliding window, complete time domain segmentation, and obtain N segmentation samples; Preprocess the implant-related data: normalize the bone density data of the implant placement site and the implant insertion torque data using the min-max normalization method according to the data category; The preprocessed resonance frequency data of the implant is stored and marked as a first data set, and the first data set is divided into a first training set and a first test set in a ratio of 8:

2.

4. The implant resonance frequency measurement and analysis method based on deep learning according to claim 3, characterized in that: Building an initial denoising network based on a convolutional neural network includes: Based on the convolutional neural network, improvements are made by adding a batch normalization layer BN between the convolutional layer and the ReLU activation function in the convolutional neural network structure; after the improvement is completed, an initial denoising network is obtained.

5. The implant resonance frequency measurement and analysis method based on deep learning according to claim 4, characterized in that: The initial denoising network is optimized and trained based on the first data set to obtain a denoising neural network, and the second data set is constructed including: Set the training hyperparameters of the initial denoising network, set the learning rate to E1, the batch size to B1, and the number of model training rounds to P1; According to the set initial denoising network training hyperparameters, the first data set is used to train and optimize the initial denoising network, the WO optimization algorithm is introduced during the training process to optimize the initial denoising network training process, the weight parameters in the initial denoising network are encoded as the position parameters in the WO optimization algorithm, the loss function of the initial denoising network training is used as the objective function of the WO algorithm, and the optimal weight parameters of the initial denoising network are obtained by finding the position parameters with the minimum loss function; The optimal weight parameters are updated to the initial denoising network, and the initial denoising network with the updated optimal weight parameters is trained with the first training set to obtain a denoising neural network; The first data set was denoised using a denoising neural network, and the denoised first data set and implant-related data were merged and classified and stored according to the implants to which they belonged. Each implant included resonance frequency data, bone density data at the implant implantation site, implant implantation torque data, and ISQ data at the initial stage of implantation. After merging, it was marked as the second data set, and the second data set was divided into a second training set and a second test set in a ratio of 8:

2.

6. The implant resonance frequency measurement and analysis method based on deep learning according to claim 5, characterized in that: Building an initial analysis network based on a convolutional neural network includes: Improvements are made based on convolutional neural networks: the convolution kernel and pooling kernel in the convolutional neural network are changed to a one-dimensional structure; the size of the convolution kernel of the first convolution layer in the convolutional neural network is increased to x*y; a batch normalization layer is added between the convolution layer and the ReLU activation function in the convolutional neural network, and a Dropout operation is added after the convolution layer in the convolutional neural network. The improvement is completed to obtain the initial analysis network.

7. The implant resonance frequency measurement and analysis method based on deep learning according to claim 6, characterized in that: The initial analysis network is optimized and trained based on the second data set, and the analysis neural network obtained includes: Initialize the network structure parameters of the initial analysis network; set the initial analysis network training hyperparameters, set the learning rate to E2; set the batch size Batch_size to B2; set the number of model training rounds to P2 rounds; According to the set initial analysis network training hyperparameters, the initial analysis network is trained using the second training set. After the training is completed, the initial analysis network is tested using the second test set to obtain the test results, and the loss function is used to determine whether the model loss value is qualified. If the model loss value is qualified, the training is completed to obtain the analysis neural network. If the model loss value is unqualified, the network structure parameters of the initial analysis network are updated through back propagation, and the initial analysis network is trained again using the second training set. After the training is completed, the initial analysis network is tested using the second test set to obtain the test results, and the loss function is used to determine whether the model loss value is qualified. The training is stopped until the model loss value is qualified to obtain the analysis neural network.

8. The implant resonance frequency measurement and analysis method based on deep learning according to claim 7, characterized in that: The resonance analysis model based on the denoising neural network and the analytical neural network includes: The denoising neural network and the analysis neural network are connected in series in sequence. The collected resonance frequency data is first processed by the denoising neural network and then output to the analysis neural network for processing. The denoising neural network and the analysis neural network connected in series are marked as a resonance analysis model.

9. The implant resonance frequency measurement and analysis method based on deep learning according to claim 8, characterized in that: The resonance analysis model is used to analyze the resonance frequency of the implant, including: Use a resonance frequency measurement device to obtain the resonance frequency data of the implant, obtain the implant-related data of the implant, including: the bone density data of the implant implantation site and the implant implantation torque data, normalize the resonance frequency data, the bone density data of the implant implantation site and the implant implantation torque data, then input the resonance frequency data, the bone density data of the implant implantation site and the implant implantation torque data into the resonance analysis model, and output the analysis results.

10. A system for measuring and analyzing implant resonance frequency based on deep learning, used to implement the method for measuring and analyzing implant resonance frequency based on deep learning according to any one of claims 1 to 9, characterized in that: It includes data acquisition module, denoising network module, analysis network module and resonance analysis module; The data acquisition module includes a collection unit and a processing unit, the collection unit is used to collect the resonance frequency data of the implant and the implant-related data; the processing unit is used to pre-process the resonance frequency data of the implant and the implant-related data, and construct a first data set; The denoising network module includes a denoising construction unit and a denoising training unit, and the denoising construction unit constructs an initial denoising network based on a convolutional neural network; The denoising training unit performs optimization training based on the initial denoising network of the first data set to obtain a denoising neural network, and constructs a second data set using the denoising neural network; The analysis network module includes an analysis construction unit and an analysis training unit. The analysis construction unit constructs an initial analysis network based on a convolutional neural network. The analysis training unit optimizes and trains the initial analysis network based on a second data set to obtain an analysis neural network. The resonance analysis module includes a model construction unit and a frequency analysis unit. The model construction unit constructs a resonance analysis model based on a denoising neural network and an analysis neural network. The frequency analysis unit includes a resonance analysis model. The resonance analysis model is used to analyze the resonance frequency of the implant.