AUTOMATIC DETECTION OF DENTAL INDICATIONS

Description

[0001] The present invention relates to a recognition method for a dental object, a manufacturing apparatus for a dental object and a computer program for recognizing a dental object.

[0002] In common digital workflows, dental restorations are created in CAD software and saved in a 3D CAD file format. Depending on the CAD software used and the subsequent workflow, so-called metadata containing additional information can be generated and saved along with the 3D CAD file.

[0003] Metadata can be helpful for controlling or optimizing ablation manufacturing processes, such as assigning milling strategies for a specific restoration type. In three-dimensional printing processes that require support structures, such as stereolithography, VAT polymerization, 3D DLP printing, or selective laser sintering of metals, certain surfaces of dental restorations should be free of support structures to positively influence precision and surface quality.

[0004] However, metadata has the disadvantage that it is already generated in the CAD software and is a proprietary data format. In this case, the CAM software used must be able to import and interpret the associated proprietary metadata. Since these metadata formats are not universal, a separate interface must be programmed for each proprietary metadata format. If no metadata is available for an object, the CAM software cannot determine which indication it is during import.

[0005] However, the indication is important for determining which manufacturing process should be used to produce a dental object. The classification can also be used for further data processing and to determine further process steps for manufacturing or material processing in the digital workflow.

[0006] The document US 10,856,957 B2 relates to a method for producing a three-dimensional digital model of a prosthesis base for manufacture using a light-based three-dimensional printing device.

[0007] The documents US 2021 / 255,600 A1 and EP 3 868 330 A1 relate to a method for producing a dental restoration, a computer program for producing a dental restoration, and a milling device for producing a dental restoration. A machine learning algorithm is used to automatically integrate spatial data for retaining bars into a three-dimensional data set of a blank. This algorithm learns from previous examples in the form of three-dimensional data sets and, after the learning phase, can generalize these to automatically calculate retaining bars. For this purpose, the algorithm can build a statistical model based on training data during machine learning. Patterns and regularities in the training data for the positioning of retaining bars are recognized. The machine learning algorithm can, for example, comprise a trained neural network.

[0008] EP 4 113 373 A1, published on April 1, 2023, claims a priority date of July 1, 2021, is prior art under Art. 54(3) EPC, and relates to computer-implemented methods and products for assisting dental procedures. It discloses a method for classifying a tooth structure using trained data-driven logic, the method comprising providing a 2D image as input to the trained data-driven logic and determining a classification value of the 2D image using the data-driven logic.

[0009] The technical task of the invention is to recognize the indication of digital dental objects based on the geometric shape so that a suitable manufacturing process can be selected.

[0010] This technical problem is solved by the subject matter according to the independent claims. Technically advantageous embodiments are the subject matter of the dependent claims, the description, and the drawings.

[0011] According to a first aspect, the technical problem is solved by a recognition method for a dental object, comprising the steps of providing a digital dental object in a coordinate system that describes a shape of the dental object to be manufactured; automatically assigning the digital dental object to a predetermined class based on the shape using a self-learning algorithm; assigning a manufacturing method to the digital dental object based on the assigned class; and adding further spatial structures to the digital dental object based on the assigned manufacturing method. The class can be assigned a specific manufacturing method with which the real dental object can be manufactured based on the digital dental object. Depending on the class of the dental object, adapted manufacturing methods with optimal parameters can be used.The spatial structures can be support structures for a build platform or holding structures for a blank. This provides the technical advantage of further improving the production of the dental object. This provides the technical advantage of allowing the dental object to be manufactured efficiently and the production of different dental objects to be automated.

[0012] In a technically advantageous embodiment of the recognition method, a number of points on the surface of the digital dental object are detected. This achieves the technical advantage, for example, that the digital dental object can be quickly classified using a small amount of classification data.

[0013] In another technically advantageous embodiment of the recognition method, the points on the surface of the digital dental object are randomly selected. This achieves the technical advantage, for example, that the population of classification data can be obtained without complex calculations.

[0014] In another technically advantageous embodiment of the recognition method, the coordinates of the detected points form input for an artificial neural network. This achieves the technical advantage, for example, of efficiently assigning the digital dental object to a class.

[0015] In another technically advantageous embodiment of the recognition method, the artificial neural network has been trained using a plurality of training data sets. This also achieves the technical advantage, for example, of efficiently assigning the digital dental object to a class.

[0016] In another technically advantageous embodiment of the recognition method, the class of the digital dental object is output by the artificial neural network. This also achieves the technical advantage, for example, of efficiently assigning the digital dental object to a class.

[0017] In another technically advantageous embodiment of the recognition method, a digital reference object is assigned to the digital dental object based on the assigned class. This provides the technical advantage, for example, of further optimizing a workflow.

[0018] In a further technically advantageous embodiment of the recognition method, the digital dental object is transformed based on the assigned class and / or the reference object in the coordinate system. The digital dental object can, for example, be aligned with the reference object or shifted relative to the reference object. This achieves the technical advantage, for example, that the manufacturing process can be carried out with an adapted orientation and / or position of the digital dental object.

[0019] In a further technically advantageous embodiment of the detection method, the dental object is manufactured using the manufacturing method. This achieves, for example, the technical advantage that the dental object is manufactured using a suitable manufacturing method.

[0020] In a further technically advantageous embodiment of the detection method, the manufacturing process is an additive or subtractive manufacturing process. This achieves the technical advantage, for example, of using particularly suitable manufacturing processes for dental objects.

[0021] In another technically advantageous embodiment of the recognition method, the correctness of the assignment is checked using geometric features of the digital dental object. This achieves the technical advantage of improving the accuracy of the method, for example.

[0022] According to a second aspect, the technical problem is solved by a manufacturing device for a dental object, which is configured to perform the detection method according to the first aspect. The manufacturing device achieves the same technical advantages as the detection method.

[0023] According to a third aspect, the technical problem is solved by a computer program with instructions that, when executed by a computer, cause the computer to carry out the method according to the first aspect. The computer program achieves the same technical advantages as the recognition method.

[0024] Embodiments of the invention are illustrated in the drawings and are described in more detail below.

[0025] They show: Fig. 1 shows a schematic representation of classification methods; Fig. 2 shows a plurality of points from a classification data set; Fig. 3 shows a schematic view of an artificial neural network; Fig. 4 shows digital dental objects of different classes; Fig. 5 shows a display of a recognized class; and Fig. 6 shows a block diagram of a recognition method for a dental object.

[0026] Fig. 1 shows a schematic representation of classification methods. A real dental object 100-2 is, for example, a crown, a bridge, a veneer, an abutment, an inlay, an onlay, a splint, or a partial or full denture. In general, the dental object 100-2 can be any object in the dental field that is to be manufactured additively or subtractively as part of a three-dimensional manufacturing process.

[0027] For each of these real dental objects 100-2, a digital dental object 100-1 can exist, in which the three-dimensional shape and color values are specified. This information is stored, for example, in a data record for the digital dental object 100-1.

[0028] By classifying the digital dental object 100-1, it can be determined which type it belongs to. The type of the digital dental object 100-1 can be, for example, "crown," "bridge," "veneer," "abutment," "inlay," "onlay," "splint," or "partial denture," or "full denture." In general, the type of the digital dental object 100-1 can be any type that can be used to distinguish real dental objects 100-2 from one another.

[0029] A distinction between the digital dental objects 100-1 can be made, for example, based on their respective shape. The classification can therefore be performed based on classification data sets 101, which include a large number of points on the surface of the digital dental object 100-1. If numerous points 115 on the surface of the digital dental object 100-1 are known, it can be automatically determined, for example, whether it is an abutment or a full denture.

[0030] Each point in the diagrams represents a single classification data set 101 for a digital dental object 100-1 as input data. Nearest-neighbor classification performs a nonparametric method for estimating probability density functions for the classification data sets 101. The resulting k-nearest-neighbor algorithm is a classification method in which a class assignment is performed taking into account its k nearest neighbors.

[0031] When classifying the classification data sets 101 with a linear support vector machine (SVM), a classifier and a regressor are used for regression analysis. The support vector machine divides the set of digital dental objects into classes so that the widest possible area around the class boundaries remains free of objects. The support vector machine is a so-called large margin classifier. Additionally, a radial basis function (RBF) can be used, the value of which depends on the distance from the origin. When classifying the classification data sets 101 with a Gaussian method, a stochastic process is used in which each finite subset of random variables has a multidimensional normal distribution (Gaussian distribution).

[0032] Each classification data set 101 can be assigned to a class 103-1, ..., 103-n by the classification method, such as whether it is the "abutment" or the "full denture" class. The manufacturing process for a real dental object 100-2 can then be determined based on the class of the digital dental object 100-1.

[0033] Fig. 2 shows a plurality of points 115 on the surface of the digital dental object 100-1, which may be present in a classification data set 101. The points 115 are each defined by the coordinates X, Y, and Z of a coordinate system.

[0034] This classification data set 101 is assigned to the class "full denture" by the classification procedure. To make the classification procedure invariant to an input permutation, symmetric functions are used. A symmetric function is a function in which the variables can be interchanged without changing the function value.

[0035] The classification method can be trained using a plurality of classification data sets 101, each of which has a known class. New classification data sets 101 can then be classified based on their similarity to the learned classification data sets 101.

[0036] During classification, the classification algorithm independently learns a suitable similarity measure during training based on the input data and its properties in order to distinguish between the various classes of dental objects 100-1. Therefore, it is not necessary to define an explicit similarity measure. Like a neural network, the classification algorithm implicitly determines during the training phase, based on the training data, which properties of the point clouds are relevant for depicting a similarity. This is then used to classify the classification dataset 101. Depending on the form of the training dataset, the similarity measure can vary. This is an effect that generally occurs with more complex classification algorithms and is deliberately accepted.

[0037] Fig. 3 shows a schematic view of an artificial neural network 109. The artificial neural network 109 is a network of artificial neurons and can be used to classify the digital dental object 100-1. The artificial neural network 109 comprises an input layer 111-IN with a number of neurons 113 that corresponds, for example, to the number of points 115 on the surface of the digital dental object 100-1 in the classification data set 101. In this case, each point from the classification data set 101 is input into its own neuron 113.

[0038] These are forwarded to the neurons 113 of hidden layers 111-Hidden. Each signal is individually weighted from one neuron 113 to another neuron 113. The result is then output to the output layer 111-OUT. The number of neurons 113 of the output layer 111-OUT corresponds, for example, to the number of classes. This creates a mapping: F(X 1 ,Y 1 ,Z 1 , ..., X n ,Y n ,Z n ) -> Class 1 , ..., Class m

[0039] During training of the neural network 109, a plurality of classification data sets 101 are fed, each of which has a known class. The neural network 109 learns by modifying the weights between the neurons 113. The weights are adjusted until the output class corresponds to the class known for the classification data set 101.

[0040] Generally, a combination of convolutional layers and dense layers can be used. Sigmoid, tanh, or ReLU functions can be used as activation functions. Batch normalization can be performed after each layer.

[0041] The invention can be implemented, for example, using the following source code:

[0042] Fig. 4 shows several digital dental objects 100-1 of different classes, such as a crown (left), an abutment (center), and another crown (right). The surface of these dental objects 100-1 is described by a point cloud 117 of random points.

[0043] Fig. 5 shows a display of a recognized class. The class of the digital dental object 100-1 can be displayed. A unique identifier can then be added to the data set of the digital dental object 100-1. This allows an initially unknown data set to become a data set that is correctly classified and can be used again to train the classification algorithm.

[0044] Fig. 6shows a block diagram of a recognition method for a dental object 100-1. The recognition method comprises step S101 of providing the digital dental object 100-1 in a coordinate system that describes a shape of the dental object 100-2 to be manufactured. For this purpose, for example, the dental 3D data set of the digital dental object 100-1 is imported. Subsequently, a plurality of evenly distributed random points on the surface of the dental object 100-1, such as 4096 random points, can be determined and stored in a classification data set 101.

[0045] Subsequently, in step S102, the digital dental object 100-1 is automatically assigned to a predefined class based on its shape using a self-learning algorithm. For this purpose, this algorithm was previously trained using a large number of known digital dental objects 100-1 and their respective dental classifications and / or indications. Each indication is a subset of a class. The self-learning algorithm can optimize the parameters of a function that maps the geometric information of the digital dental object 100-1 to the respective class.

[0046] This recognition process allows the three-dimensional data sets for the digital dental objects 100-1 to be identified automatically and without any attached metadata during import into the CAM software. Instead of using a static algorithm to explicitly identify specific features of the digital dental object 100-1, a dynamic machine learning algorithm is used, which learns the features implicitly.

[0047] The result depends on the three-dimensional structure and the number of available digital dental objects 100-1 with an indication that were used for training. In a further step, a verification and plausibility check of the classification can be performed based on simple geometric features, such as the dimensions of the digital dental object 100-1 (bounding box, centroid, local curvature).

[0048] The plausibility check is performed after classification, once the classification result has already been determined, to double-check it. For example, the volume of dental object 100-1 can be used to check plausibility. Based on this volume, it can be checked, for example, whether the previous classification result "single crown" is correct. Based on this volume, it can be determined whether it is actually a single crown or whether the volume is too large for a single crown and more closely corresponds to the volume of a bridge.

[0049] Furthermore, features such as the dimensions of a bounding box, a maximum bounding box size, the aspect ratio of the bounding box, the curvature of the dental object, or the position of the dental object's center of mass can be used for the plausibility check. Combinations of multiple features can also be used for this purpose.

[0050] To enable the optimal orientation and positioning of the digital dental object 100-1 in the dental workflow using the CAM-AM software, a classification of the digital dental object 100-1 should be known. In this case, for example, rules for the orientation of the digital dental object can be defined so that an optimized support structure can be generated automatically.

[0051] If the class, i.e., the indication, of the digital dental object 100-1 is known to the CAD-CAM software, further measures can be implemented to optimize the manufacturing workflow. These measures include, for example, optimizing the orientation of the digital dental object so that the manufactured dental object requires as few support structures as possible and a large portion of the dental object's surface no longer requires reworking.

[0052] For this purpose, a digital reference object based on the determined class can be retrieved from a database. For example, if it were determined that the digital dental object 100-1 is a full denture, a digital reference object of the "full denture" class can be retrieved from the database. The digital dental object 100-1 can then be transformed to the reference object in the coordinate system. For example, an orientation, position, or size of the digital dental object 100-1 is adapted to the reference object. Furthermore, a manufacturing process used to manufacture the digital dental object can be provided for the reference object of the respective class.

[0053] The recognition method for the dental object 100-1 can be implemented in a manufacturing device 200 with which the real dental objects 100-2 can be produced based on the digital dental objects 100-1, such as a 3D printing device or a milling device. For this purpose, the manufacturing device 200 comprises, for example, a computer with a digital memory and a processor for executing a computer program that implements the recognition method.

[0054] All features explained and shown in connection with individual embodiments of the invention can be provided in different combinations in the subject matter according to the invention in order to simultaneously realize their advantageous effects.

[0055] All method steps can be implemented by devices suitable for performing the respective method step. All functions performed by physical features can be a method step of a method.

[0056] The scope of the present invention is given by the claims and is not limited by the features explained in the description or shown in the figures. LIST OF REFERENCE SYMBOLS

[0057] 100-1 digital dental object 100-2 dental object to be manufactured 101 classification dataset 103 class 109 artificial neural network 111 layer 113 neuron 115 point 117 point cloud 200 manufacturing devices

Claims

1. A recognition method for a dental object (100-1), comprising the steps of: - providing (S101) a digital dental object (100-1) in a coordinate system describing a shape of the dental object (100-2) to be manufactured; and - automatically assigning (S102) the digital dental object (100-1) to a specified class based on the shape by a self-learning algorithm; - assigning a manufacturing method to the digital dental object (100-1) based on the assigned class; and - adding further spatial structures to the digital dental object (100-1) based on the assigned manufacturing method.

2. The recognition method according to claim 1, wherein a number of points on the surface of the digital dental object (100-1) is detected.

3. The recognition method according to claim 2, wherein the points on the surface of the digital dental object (100-1) are randomly selected.

4. The recognition method according to claim 2 or 3, wherein the coordinates of the detected points form an input for an artificial neural network (109).

5. The recognition method according to any one of claims 2 to 4, wherein the artificial neural network (109) has been trained by a plurality of training data sets.

6. The recognition method according to any one of claims 2 to 5, wherein the class is output by the artificial neural network (109).

7. The recognition method according to any one of the preceding claims, wherein a digital reference object is assigned to the digital dental object (100-1) based on the assigned class.

8. The recognition method according to any one of the preceding claims, wherein the digital dental object (100-1) is transformed based on the assigned class and / or the reference object in the coordinate system.

9. The recognition method according to claim 1, wherein the dental object (100-2) is manufactured by the manufacturing method.

10. The recognition method according to claim 9, wherein the manufacturing method is an additive or subtractive manufacturing method.

11. A manufacturing apparatus (200) for a dental object (100-2), which is configured to perform the recognition method according to any one of claims 1 to 10.

12. A computer program, comprising instructions which, when the computer program is executed by a computer, cause the computer to execute the method according to any one of claims 1 to 8.