Implant tooth central screw torque classification method and device and storage medium

Abstract

The invention discloses a dental implant central screw torque classification method and device and a storage medium. The method comprises the following steps: outputting a central screw torque classification result corresponding to first implant tooth root tip data according to the first implant tooth root tip data and an implant tooth central screw torque evaluation model; in the training process of the dental implant central screw torque evaluation model, performing transfer learning training on a pre-trained basic model by establishing first in-vitro torque stable data and first in-vitro torque loss data to obtain a first screw torque evaluation model; and performing fine adjustment on the model through the first in-vivo torque stability data and the first in-vivo torque loss data to generate a final implant tooth central screw torque evaluation model. By adopting the technical scheme of the invention, the model application cost can be reduced while the image classification precision is ensured through the two-stage training framework.

Description

Technical Field

[0001] This application relates to the field of medical image processing, and particularly to a method, device, and storage medium for classifying the torque of dental implant central screws. Background Art

[0002] With the progress of oral imaging technology and the breakthrough of deep learning algorithms, it has become possible to non-invasively evaluate the torque of the central screw of the dental implant crown based on the image features of periapical films. Currently, the traditional imaging-based manual diagnosis method mainly faces two key technical bottlenecks: 1. The subjective judgment method that only relies on doctors' naked-eye observation of the looseness of the crown leads to insufficient sensitivity in the early identification of torque attenuation; 2. The existing invasive torque detection method requires destructive removal of the sealing resin and mechanical testing, and its repeated force application operation is likely to cause secondary damage to the central screw, resulting in irreversible damage to the implant system.

[0003] In addition, the model methods based on deep learning in the existing technology often rely on a large amount of training data and require a high-performance GPU cluster for long-term training, resulting in a sharp increase in the clinical implementation cost and severely restricting its promotion and applicability in primary medical institutions.

[0004] Application Content

[0005] This application provides a method, device, and storage medium for classifying the torque of dental implant central screws, which can reduce the model training cost while ensuring the image classification accuracy.

[0006] In a first aspect, an embodiment of this application provides a method for classifying the torque of dental implant central screws, including:

[0007] Obtaining first periapical film data of a dental implant to be classified;

[0008] According to the first periapical film data of the dental implant and a dental implant central screw torque evaluation model, outputting a classification result corresponding to the first periapical film data of the dental implant; wherein, the dental implant central screw torque evaluation model is obtained by fine-tuning a first screw torque evaluation model according to in-vivo torque data; the first screw torque evaluation model is obtained by training a basic model according to in-vitro torque data and a preset transfer learning method.

[0009] In the embodiment of the present application, by obtaining the first dental implant apical film data to be classified, data support is provided for classification. In the construction of the dental implant central screw torque evaluation model, transfer learning technology is used to train the first screw torque evaluation model. In transfer learning technology, through feature layer reuse, the training cost of the first screw torque evaluation model is reduced, and in the first stage of training, the model learns the physical deformation characteristics under different torques. Then, fine-tuning is performed based on the first screw torque evaluation model, and in the second stage of training, clinical image-specific features are injected through the first in-vivo torque stability data and the first in-vivo torque loss data. In this way, the dental implant central screw torque evaluation model adopted in the dental implant central screw torque classification method is constructed through a two-stage training framework of in-vitro first-stage training and in-vivo second-stage fine-tuning, reducing the model training scale while ensuring the accuracy of image classification, thereby reducing the usage cost of the classification method.

[0010] As a preferred example of the first aspect, the first screw torque evaluation model is obtained by training a basic model according to in-vitro torque data and a preset transfer learning method, including:

[0011] Obtain the original data, use the original data as input to train a preset convolutional neural network to obtain the basic model, and process the in-vitro torque data into the first in-vitro torque stability data and the first in-vitro torque loss data through the first torque division rule and the first annotation method;

[0012] Use the first in-vitro torque stability data and the first in-vitro torque loss data as input, and train the basic model using the transfer learning method until the current iteration meets the first preset condition to obtain the first screw torque evaluation model.

[0013] In this preferred example, the in-vitro torque data is processed into the first in-vitro torque stability data and the first in-vitro torque loss data through the first torque division rule and the first annotation method, which ensures the clinical relevance of torque state classification. Then, the basic model retains the general image recognition features, improving the accuracy of image recognition.

[0014] As a preferred example of the first aspect, the obtaining of the original data includes:

[0015] Obtain a torque set, and use a preset device to obtain the first original picture corresponding to each torque in the torque set, and form all the first original pictures into the original data.

[0016] In this preferred example, the in-vitro training stage reduces the dependence on the amount of dental professional data through the original data.

[0017] As a preferred example of the first aspect, the implant central screw torque evaluation model is obtained by fine-tuning the first screw torque evaluation model according to in-vivo torque data, including:

[0018] Processing the in-vivo torque data into first in-vivo torque stable data and first in-vivo torque loss data through a second torque division rule and a second annotation method;

[0019] Adjusting the weights of the fully connected layer in the first screw torque evaluation model to obtain a second screw torque evaluation model;

[0020] Using the first in-vivo torque stable data and the first in-vivo torque loss data as inputs, training the second screw torque evaluation model using transfer learning until the current iteration meets the second preset condition to obtain the implant central screw torque evaluation model.

[0021] In this preferred example, by reconstructing the weight distribution of the fully connected layer, the model can better adapt to the data characteristics in vivo, and on the basis of retaining the in-vitro pre-training recognition ability, significantly enhance the model's classification ability for clinical real images.

[0022] As a preferred example of the first aspect, outputting a classification result corresponding to the first implant apical film data according to the first implant apical film data and the implant central screw torque evaluation model, including:

[0023] Annotating the screw area of the first implant apical film data to obtain second implant apical film data;

[0024] Performing convolution and pooling operations on the second implant apical film data to obtain a first feature vector, and inputting the first feature vector into a fully connected network to obtain a probability value;

[0025] If the probability value is greater than or equal to a preset threshold, it is determined as torque stable, otherwise it is determined as torque loss.

[0026] In this preferred example, by accurately annotating the screw area of the implant apical film and deploying the trained implant central screw torque evaluation model to complete the classification, a high-precision image classification result is provided for the user in a low-risk situation.

[0027] In a second aspect, an embodiment of the present application further provides an implant central screw torque classification device, including: a data acquisition module and a result output module;

[0028] The data acquisition module is used to acquire first implant apical film data to be classified;

[0029] The result output module is configured to output a classification result corresponding to the first dental implant apical film data according to the first dental implant apical film data and the dental implant central screw torque evaluation model; wherein, the dental implant central screw torque evaluation model is obtained by fine-tuning the first screw torque evaluation model according to in-vivo torque data; the first screw torque evaluation model is obtained by training a basic model according to in-vitro torque data and a preset transfer learning method.

[0030] As a preferred example of the second aspect, the result output module includes: a data acquisition unit and a first training unit;

[0031] The data acquisition unit is configured to acquire original data, use the original data as input to train a preset convolutional neural network to obtain the basic model, and process the in-vitro torque data into first in-vitro torque stable data and first in-vitro torque loss data through a first torque division rule and a first annotation method;

[0032] The first training unit is configured to use the first in-vitro torque stable data and the first in-vitro torque loss data as input to train the basic model using a transfer learning method until the current iteration meets a first preset condition to obtain the first screw torque evaluation model.

[0033] As a preferred example of the second aspect, the data acquisition unit includes an original data acquisition subunit;

[0034] The original data acquisition subunit is configured to acquire a torque set and use a preset device to acquire a first original image corresponding to each torque in the torque set, and form all the first original images into the original data.

[0035] As a preferred example of the second aspect, the result output module further includes: a data processing unit, a weight adjustment unit, and a second training unit;

[0036] The data processing unit is configured to process the in-vivo torque data into first in-vivo torque stable data and first in-vivo torque loss data through a second torque division rule and a second annotation method;

[0037] The weight adjustment unit is configured to adjust the weights of the fully connected layer in the first screw torque evaluation model to obtain a second screw torque evaluation model;

[0038] The second training unit is configured to use the first in-vivo torque stable data and the first in-vivo torque loss data as input to train the second screw torque evaluation model using a transfer learning method until the current iteration meets a second preset condition to obtain the dental implant central screw torque evaluation model.

[0039] As a preferred example of the second aspect, the result output module includes a data annotation unit, a probability value acquisition unit, and a determination unit;

[0040] The data annotation unit is configured to annotate the screw area of the first dental implant apical film data to obtain the second dental implant apical film data;

[0041] The probability value acquisition unit is configured to perform convolution and pooling operations on the second dental implant apical film data to obtain a first feature vector, and input the first feature vector into a fully connected network to obtain a probability value;

[0042] The determination unit is configured to determine that the torque is stable if the probability value is greater than or equal to a preset threshold, otherwise determine that the torque is lost.

[0043] In a third aspect, the present application further provides a computer storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, it implements a method for classifying the torque of a central screw of a dental implant as described in the present application. Description of the Drawings

[0044] Figure 1 is a schematic flowchart of a method for classifying the torque of a central screw of a dental implant provided by some embodiments of the present application;

[0045] Figure 2 is a schematic structural diagram of a device for classifying the torque of a central screw of a dental implant provided by some embodiments of the present application. Detailed Embodiments

[0046] In the prior art, the implant restoration is mechanically connected to the implant through the abutment screw, and the clinical standard recommends an initial retention torque of 35 N·cm; however, under the long-term occlusal load, the abutment screw is prone to mechanical fatigue, resulting in attenuation of the retention torque. Torque attenuation will cause two clinical risks: First, the micro-movement of the restoration forms a micro-gap at the interface, which promotes the colonization of plaque and induces peri-implant mucositis or even peri-implantitis; Second, the micro-movement causes stress concentration, significantly increasing the risk of screw fracture, which may cause complications such as restoration detachment, aspiration, and swallowing, and requires secondary surgery when the fractured stump is difficult to remove, seriously damaging the health of the patient.

[0047] The current clinical monitoring scheme has significant deficiencies: conventional follow-up only assesses the looseness of the prosthesis through instrument palpation, but cannot quantitatively detect the actual torque state of the screw. Only when the torque decays to the critical threshold (15 N·cm), palpation can identify loosening, while subcritical decay is difficult to detect, resulting in a risk of missed diagnosis. Accurate detection requires destroying the sealing resin and then reversely loosening the screw with a torque wrench and recording the residual torque value. However, this invasive operation is not only time-consuming and cumbersome, but repeated application of force will accelerate the degradation of the screw's mechanical properties, inducing thread stripping or fracture. In addition, the model methods based on deep learning in the existing technology often rely on a large amount of training data and require a high-performance GPU cluster for long-term training, resulting in a sharp increase in the clinical implementation cost, seriously restricting its promotion and applicability in primary medical institutions.

[0048] To solve the above technical problems, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0049] Embodiment 1

[0050] Please refer to Figure 1 , a method for classifying the torque of the central screw of an implant provided in the embodiment of the present application, including S11 to S12, specifically:

[0051] S11: Obtain the first implant apical radiograph data to be classified;

[0052] S12: According to the first implant apical radiograph data and the implant central screw torque evaluation model, output the classification result corresponding to the first implant apical radiograph data; wherein, the implant central screw torque evaluation model is obtained by fine-tuning the first screw torque evaluation model according to in-vivo torque data; the first screw torque evaluation model is obtained by training a basic model according to in-vitro torque data and a preset transfer learning method.

[0053] Further, in some embodiments of the present application, the first screw torque evaluation model is obtained by training a basic model according to in-vitro torque data and a preset transfer learning method, including steps S121 to S123, and the specific steps are as follows:

[0054] S121: Obtain the original data, use the original data as the input, train a preset convolutional neural network to obtain the basic model, and process the in-vitro torque data into first in-vitro torque stable data and first in-vitro torque loss data through the first torque division rule and the first annotation method;

[0055] S122: Use the first in-vitro torque stability data and the first in-vitro torque loss data as inputs, and train the basic model using transfer learning until the current iteration meets the first preset condition to obtain the first screw torque evaluation model.

[0056] Specifically, to fully explain step S121, the following solution is used as an example for illustration:

[0057] To obtain in-vitro torque data, implants are implanted on in-vitro pig ribs, and then abutments are connected to the implants. Different torques are designed, namely 35, 30, 25, 20, and 15 N·cm. 35, 30, and 25 N·cm are defined as torque stability, and 20 and 15 N·cm are defined as torque loss. Then, periapical radiographs are taken of the samples, and two projection angles, parallel and inclined at 15°, are used for shooting. Next, a specific image processing software, Label Me, is used to mark the central screw of the implant in the periapical radiograph (i.e., the rectangular range in the periapical radiograph that contains the full length of the central screw) to enhance the subsequent feature extraction ability of the model and reduce the training interference of irrelevant information. After marking, it is saved in a picture format that can be recognized by the target detection convolutional neural network. Finally, the first in-vitro torque stability data and the first in-vitro torque loss data are constructed.

[0058] Specifically, to fully explain step S122, the following solution is used as an example for illustration:

[0059] First, use the first in-vitro torque stability data and the first in-vitro torque loss data marked by Label Me as the training set and validation set to train and optimize the convolutional neural network, and use the test set to test the convolutional neural network. Finally, a detection and classification network for the central screw torque of the implant based on the in-vitro model is obtained. In terms of data division, the training set, validation set, and test set are divided according to a ratio of 7:2:1, where the training set and validation set are used for the training and validation processes of the model. This application uses the deep convolutional neural network transfer learning technology to extract features and train the model. The specific steps are as follows:

[0060] (1) Select ResNet as the basic model framework and freeze the weight parameters pre-trained for the first time on the ImageNet data to utilize its existing rich feature extraction ability;

[0061] (2) Transfer the binary classification (in vitro model - torque stable / torque loss) task of this study to the fully connected layers at the head of the basic model framework. There are 4 layers in this fully connected layer, and the specific structure is as follows: the number of nodes in the first layer is 1024, the number of nodes in the second layer is 512, the number of nodes in the third layer is 256, the number of nodes in the fourth layer is 64, and finally, it is connected to an activation layer with 2 nodes, and the activation function is Sigmoid. Assign random weights to these 4 fully connected layers so that they can learn and adjust according to the tasks of this study;

[0062] (3) Use the Adam optimizer to adjust the parameter gradients, and specify the learning rate as 0.0001. Use 32 images for each gradient update, and train the entire data 100 times. During the model training process, to enhance the generalization ability of the model, image enhancement processes such as flipping and rotation are involved. At the same time, an early stopping strategy (Early-stop) and a random dropout strategy (Drop rate set to 0.2) are adopted to improve the robustness and stability of the model.

[0063] Compared with the prior art, the above embodiments have the following beneficial effects: By using the first torque division rule and the first annotation method, the in vitro torque data is processed into the first in vitro torque stable data and the first in vitro torque loss data, which ensures the clinical relevance of torque state classification. Then, by using the basic model, the general image recognition features are retained, improving the accuracy of image recognition.

[0064] Further, in some embodiments of the present application, the obtaining of the basic model includes:

[0065] Obtain a torque set, and obtain the first original image corresponding to each torque in the torque set through a preset device, and form the original data by combining all the first original images.

[0066] Compared with the prior art, the above embodiments have the following beneficial effects: By training the basic model with the preset original data, the visual understanding ability of millions of natural images can be directly inherited in the in vitro training stage, reducing the dependence on the amount of dental professional data.

[0067] Further, in some embodiments of the present application, the implant central screw torque evaluation model is obtained by fine-tuning the first screw torque evaluation model according to the in vivo torque data, including:

[0068] Process the in vivo torque data into the first in vivo torque stable data and the first in vivo torque loss data through the second torque division rule and the second annotation method;

[0069] Adjust the weights of the fully connected layers in the first screw torque evaluation model to obtain the second screw torque evaluation model;

[0070] Using the first in-vivo torque stability data and the first in-vivo torque loss data as inputs, the transfer learning method is used to train the second screw torque evaluation model until the current iteration meets the second preset condition, and the implant central screw torque evaluation model is obtained.

[0071] Compared with the prior art, the above embodiments have the following beneficial effects: By reconstructing the weight distribution of the fully connected layer, the model can better adapt to the data characteristics in vivo, and on the basis of retaining the pre-training recognition ability in vitro, significantly enhance the model's classification ability for clinical real images.

[0072] Further, in some embodiments of the present application, outputting the classification result corresponding to the first implant periapical radiograph data according to the first implant periapical radiograph data and the implant central screw torque evaluation model includes:

[0073] Label the screw area of the first implant periapical radiograph data to obtain the second implant periapical radiograph data;

[0074] Perform convolution and pooling operations on the second implant periapical radiograph data to obtain a first feature vector, and input the first feature vector into a fully connected network to obtain a probability value;

[0075] If the probability value is greater than or equal to a preset threshold, it is determined as torque stable, otherwise it is determined as torque loss.

[0076] Compared with the prior art, the above embodiments have the following beneficial effects: By accurately labeling the screw area of the implant periapical radiograph and deploying the trained implant central screw torque evaluation model to complete the classification, high-precision image classification results are provided for users in low-risk situations.

[0077] In summary, it can be seen that the embodiments of the present application provide data support for classification by obtaining the first implant periapical radiograph data to be classified. In the construction of the implant central screw torque evaluation model, the transfer learning technology is used to train the first screw torque evaluation model. In the transfer learning technology, the training cost of the first screw torque evaluation model is reduced through feature layer reuse, and in the first stage of training, the model learns the physical deformation characteristics under different torques. Then, based on the first screw torque evaluation model, fine-tuning is performed, and in the second stage of training, clinical image-specific features are injected through the first in-vivo torque stability data and the first in-vivo torque loss data. In this way, the implant central screw torque evaluation model used in the implant central screw torque classification method is constructed through a two-stage training framework of in-vitro first-stage training and in-vivo second-stage fine-tuning, reducing the model training scale while ensuring the image classification accuracy, thereby reducing the usage cost of the classification method.

[0078] Embodiment 2

[0079] Please refer to Figure 2 , a central screw torque classification device provided by an embodiment of the present application, including: a data acquisition module 11 and a result output module 12.

[0080] Further, in some embodiments of the present application, the data acquisition module 11 is configured to acquire first implant apical film data to be classified; the result output module 12 is configured to output a classification result corresponding to the first implant apical film data according to the first implant apical film data and an implant central screw torque evaluation model; wherein, the implant central screw torque evaluation model is obtained by fine-tuning a first screw torque evaluation model according to in-vivo torque data; the first screw torque evaluation model is obtained by training a basic model according to in-vitro torque data and a preset transfer learning method.

[0081] Further, in some embodiments of the present application, the result output module 12 includes: a data acquisition unit and a first training unit; the data acquisition unit is configured to acquire original data, use the original data as input to train a preset convolutional neural network to obtain the basic model, and process the in-vitro torque data into first in-vitro torque stable data and first in-vitro torque loss data through a first torque division rule and a first annotation method; the first training unit is configured to use the first in-vitro torque stable data and the first in-vitro torque loss data as input to train the basic model using a transfer learning method until the current iteration meets a first preset condition, and obtain the first screw torque evaluation model.

[0082] Further, in some embodiments of the present application, the data acquisition unit includes an original data acquisition subunit; the original data acquisition subunit is configured to acquire a torque set, and use a preset device to acquire a first original image corresponding to each torque in the torque set, and form all the first original images into the original data.

[0083] Further, in some embodiments of the present application, the result output module 12 further includes a data processing unit, a weight adjustment unit, and a second training unit; the data processing unit is configured to process the in-vivo torque data into first in-vivo torque stable data and first in-vivo torque loss data through a second torque division rule and a second annotation method; the weight adjustment unit is configured to adjust the weights of the fully connected layer in the first screw torque evaluation model to obtain a second screw torque evaluation model; the second training unit is configured to use the first in-vivo torque stable data and the first in-vivo torque loss data as input to train the second screw torque evaluation model using a transfer learning method until the current iteration meets a second preset condition, and obtain the implant central screw torque evaluation model.

[0084] Further, in some embodiments of the present application, the result output module 12 includes a data annotation unit, a probability value acquisition unit, and a determination unit; the data annotation unit is configured to annotate the screw area of the first dental implant apical film data to obtain second dental implant apical film data; the probability value acquisition unit is configured to perform convolution and pooling operations on the second dental implant apical film data to obtain a first feature vector, and input the first feature vector into a fully connected network to obtain a probability value; the determination unit is configured to determine that the torque is stable if the probability value is greater than or equal to a preset threshold, otherwise determine that the torque is lost.

[0085] For a more detailed step flow and working principle of this embodiment, reference may be made to, but not limited to, the relevant records of Embodiment 1.

[0086] In summary, it can be seen that in the embodiments of the present application, by obtaining the first dental implant apical film data to be classified, data support is provided for classification. In the construction of the dental implant central screw torque evaluation model, a transfer learning technique is used to train the first screw torque evaluation model. In the transfer learning technique, through feature layer reuse, the training cost of the first screw torque evaluation model is reduced, and in the first stage of training, the model learns the physical deformation characteristics under different torques. Then, based on the first screw torque evaluation model, fine-tuning is performed, and in the second stage of training, clinical image specific features are injected through the first in-vivo torque stable data and the first in-vivo torque loss data. In this way, the dental implant central screw torque evaluation model adopted in the dental implant central screw torque classification method is constructed through a two-stage training framework of in-vitro first-stage training and in-vivo second-stage fine-tuning, while ensuring the image classification accuracy, reducing the model training scale, and thus reducing the usage cost of the classification method.

[0087] Embodiment 3

[0088] Based on the above embodiments of the dental implant central screw torque classification method, another embodiment of the present application provides a storage medium, which includes a stored computer program. When the computer program runs, it controls the device where the storage medium is located to execute the dental implant central screw torque classification method of any embodiment of the present application.

[0089] In this embodiment, the above storage medium is a computer-readable storage medium, and the computer program includes computer program code, which can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice within the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0090] The specific embodiments described above further elaborate on the purpose, technical solutions, and beneficial effects of the present application. It should be understood that the above description is only for the specific embodiments of the present application and is not used to limit the protection scope of the present application. In particular, for those skilled in the art, any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for classifying the torque of an implant central screw, characterized in that, Including: Obtain the first dental implant apical film data to be classified; According to the first dental implant apical film data and the dental implant central screw torque evaluation model, output the classification result corresponding to the first dental implant apical film data; wherein, the dental implant central screw torque evaluation model is obtained by fine-tuning the first screw torque evaluation model according to in-vivo torque data; the first screw torque evaluation model is obtained by training a basic model according to in-vitro torque data and a preset transfer learning method.

2. The method for classifying the torque of the central screw of an implantable tooth according to claim 1, wherein, The first screw torque evaluation model is obtained by training a basic model according to in-vitro torque data and a preset transfer learning method, including: Obtain original data, use the original data as input to train a preset convolutional neural network to obtain the basic model, and process the in-vitro torque data into first in-vitro torque stable data and first in-vitro torque loss data through a first torque division rule and a first annotation method; Use the first in-vitro torque stable data and the first in-vitro torque loss data as input, and train the basic model using a transfer learning method until the current iteration meets a first preset condition to obtain the first screw torque evaluation model.

3. The method for classifying the torque of an implant central screw according to claim 2, characterized in that The obtaining of the original data includes: Obtain a torque set, and use a preset device to obtain the first original image corresponding to each torque in the torque set, and form all the first original images into the original data.

4. The method for classifying the torque of an implant central screw according to claim 2, characterized in that, The dental implant central screw torque evaluation model is obtained by fine-tuning the first screw torque evaluation model according to in-vivo torque data, including: Process the in-vivo torque data into first in-vivo torque stable data and first in-vivo torque loss data through a second torque division rule and a second annotation method; Adjust the weights of the fully connected layer in the first screw torque evaluation model to obtain a second screw torque evaluation model; Use the first in-vivo torque stable data and the first in-vivo torque loss data as input, and train the second screw torque evaluation model using a transfer learning method until the current iteration meets a second preset condition to obtain the dental implant central screw torque evaluation model.

5. The method for classifying the torque of an implant central screw according to claim 4, characterized in that, The outputting of the classification result corresponding to the first dental implant apical film data according to the first dental implant apical film data and the dental implant central screw torque evaluation model includes: Annotate the screw area of the first dental implant apical film data to obtain second dental implant apical film data; Perform convolution and pooling operations on the second dental implant apical film data to obtain a first feature vector, and input the first feature vector into a fully connected network to obtain a probability value; If the probability value is greater than or equal to a preset threshold, it is determined as torque stable, otherwise it is determined as torque loss.

6. An implant central screw torque classification device, characterized in that, Including: A data acquisition module and a result output module; The data acquisition module is used to obtain the first dental implant apical film data to be classified; The result output module is configured to output a classification result corresponding to the first dental implant apical radiograph data according to the first dental implant apical radiograph data and the dental implant central screw torque evaluation model; wherein, the dental implant central screw torque evaluation model is obtained by fine-tuning the first screw torque evaluation model according to in-vivo torque data; the first screw torque evaluation model is obtained by training a basic model according to in-vitro torque data and a preset transfer learning method.

7. The central screw torque classification device for dental implants according to claim 6, wherein The result output module includes: a data acquisition unit and a first training unit; The data acquisition unit acquires original data, uses the original data as input to train a preset convolutional neural network to obtain the basic model, and processes the in-vitro torque data into first in-vitro torque stable data and first in-vitro torque loss data through a first torque division rule and a first annotation method; The first training unit is configured to use the first in-vitro torque stable data and the first in-vitro torque loss data as input to train the basic model using a transfer learning method until the current iteration meets a first preset condition, and then obtain the first screw torque evaluation model.

8. The implant central screw torque classification device according to claim 7, characterized in that, The data acquisition unit includes an original data acquisition subunit; The original data acquisition subunit is configured to acquire a torque set and obtain a first original image corresponding to each torque in the torque set through a preset device, and form the original data by combining all the first original images.

9. The central screw torque classification device for dental implants according to claim 7, wherein, The result output module further includes: a data processing unit, a weight adjustment unit, and a second training unit; The data processing unit is configured to process the in-vivo torque data into first in-vivo torque stable data and first in-vivo torque loss data through a second torque division rule and a second annotation method; The weight adjustment unit is configured to adjust the weights of the fully connected layer in the first screw torque evaluation model to obtain a second screw torque evaluation model; The second training unit is configured to use the first in-vivo torque stable data and the first in-vivo torque loss data as input to train the second screw torque evaluation model using a transfer learning method until the current iteration meets a second preset condition, and then obtain the dental implant central screw torque evaluation model.

10. The central screw torque classification device for dental implants according to claim 9, characterized in that, The result output module includes a data annotation unit, a probability value acquisition unit, and a determination unit; The data annotation unit is configured to annotate the screw area of the first dental implant apical radiograph data to obtain second dental implant apical radiograph data; The probability value acquisition unit is configured to obtain a first feature vector by performing convolution and pooling operations on the second dental implant apical radiograph data, and input the first feature vector into a fully connected network to obtain a probability value; The determination unit is configured to determine that the torque is stable if the probability value is greater than or equal to a preset threshold, otherwise determine that the torque is lost.

11. A computer storage medium, characterized in that, A computer program is stored on the computer storage medium, and when the computer program is executed by a processor, it implements a method for classifying the central screw torque of a dental implant according to any one of claims 1 to 5.

Images