Neural network-based generation and placement of dental appliances for tooth restorations

Abstract

A technique for automating the design of dental restorations using a neural network is described, in which an exemplary computing device receives deformation information associated with a dental restoration patient's current dental structure, provides the deformation information associated with the dental restoration patient's current dental structure as input to a neural network that is trained with the deformation information indicative of placement of dental prosthetic components for one or more teeth of the corresponding dental structure, where the dental prosthetics have been used in dental restoration treatment for the one or more teeth, and runs the neural network using the input to generate placement information for the dental prosthetic components for the dental restoration patient's current dental structure.

Description

[Technical Field]

[0001] FIELD OF THE DISCLOSURE The present disclosure relates to dental restorative appliances for restoring teeth. [Background technology]

[0002] Dentists often use dental appliances to reshape or restore a patient's dental structure. Dental appliances are typically constructed from a model of the patient's dental structure that has been expanded to the desired dental structure. The model may be a physical model or a digital model. Designing dental appliances is often a manual, time-consuming, and non-rigorous process. For example, practitioners typically design dental appliance models by trial and error. For example, practitioners may add, remove, reposition, rearrange, and / or resize features until the dental appliance model is satisfactory to the practitioner. Summary of the Invention

[0003] The present disclosure relates to techniques for automating the design of dental restorations for restoring a given patient's dental structure and / or the proposed placement of dental restorations for the patient's dental restorative treatment. A computing system configured according to aspects of the present disclosure implements a neural network-based system that is trained on any of a variety of data sets to generate a dental restoration design (or "shape") and / or a placement characteristic (or "transform") of a dental restoration for a particular patient. To generate a custom shape and / or transformation information for a patient-specific dental restoration, the computing system of the present disclosure implements a neural network that is trained on dental restoration information available from a digital library of dental structures and / or predefined appliance shapes and / or prefabricated appliance "ground truth" (e.g., appliance shapes created manually by a skilled practitioner or automatically via a rule-based system described in WO 2020 / 240351, filed May 20, 2020, the entire contents of which are incorporated herein by reference).

[0004] The dental appliance shapes are represented by a digital three-dimensional (3D) mesh that incorporates the features of the generated shapes. In various examples, the neural network-based techniques of the present disclosure may generate new shapes based on the patient's dental structure and a dataset(s) of "ground truth" appliance shapes on which the neural network is trained. In some examples, the neural network-based techniques of the present disclosure can generate placement data for shapes selected from a digital library of dental appliance shapes, and then place shapes from the library relative to or as portions of the patient's teeth based on the patient's dental structure and the dataset(s) of deformation matrices on which the neural network is trained.

[0005] In some examples, the neural network-based techniques of the present disclosure may generate placement data for dental appliances selected from a digital library of dental appliance shapes. Example placement data may relate to one or more of the position, orientation, scale, or shear map of dental appliances to be placed during a patient's dental restorative treatment. The computing system of the present disclosure may implement a simple neural network (e.g., with a relatively small number of hidden layers or no hidden layers), a graph convolutional neural network (GCNN), a generative adversarial network (GAN), a conditional generative adversarial network (cGAN), an encoder-decoder-based CNN, a U-NetCNN, a GAN with a PatchGAN discriminator, and / or another deep neural network (DNN) to perform the various techniques described herein.

[0006] In this manner, the computing system of the present disclosure trains and executes a neural network to automate the selection / generation of dental appliance component 3D meshes and the placement of dental appliance library component 3D meshes, all of which are later combined and assembled into the finished dental appliance. The computing system of the present disclosure can train the neural network using various data, such as dental structure landmarks, patient-to-mesh mapping, and two-dimensional (2D) dental structure images, to define the dental restoration components and their placement. The shape and placement information generated by the neural network trained on the dataset(s) selected according to the techniques of the present disclosure automatically generates a mesh (a 3D digital custom model) that can be used to manufacture the patient-specific dental restoration, such as by 3D printing the restoration from the mesh.

[0007] In one example, the computing device includes an input interface and a neural network engine. The input interface is configured to receive deformation information associated with a current dental structure of a dental restoration patient. The neural network engine is configured to provide the deformation information associated with the current dental structure of the dental restoration patient as input to a neural network trained with the deformation information indicative of placement of dental appliance components for one or more teeth of the corresponding dental structure, the dental appliances being used in the dental restoration treatment for the one or more teeth. The neural network engine is further configured to execute the neural network using the input to generate placement information of the dental appliance components for the current dental structure of the dental restoration patient.

[0008] In another example, a method includes receiving deformation information associated with a dental restoration patient's current dental structure. The method further includes providing the deformation information associated with the dental restoration patient's current dental structure as input to a neural network that is trained with the deformation information indicative of placement of dental appliance components relative to one or more teeth of the corresponding dental structure, the dental appliances being used in the dental restoration treatment of the one or more teeth. The method further includes running the neural network using the input to generate placement information for the dental appliance components relative to the dental restoration patient's current dental structure.

[0009] In another example, an apparatus includes means for receiving deformation information associated with a dental restoration patient's current dental structure; means for providing the deformation information associated with the dental restoration patient's current dental structure as input to a neural network that is trained with the deformation information indicative of placement of dental prosthetic components on one or more teeth of the corresponding dental structure, where the dental prosthetics have been used in dental restoration treatment for the one or more teeth; and means for running the neural network using the input to generate placement information of the dental prosthetic components on the dental restoration patient's current dental structure.

[0010] In another example, a non-transitory computer-readable storage medium is encoded with instructions that, when executed, cause one or more processors of a computing system to receive deformation information associated with a dental restoration patient's current dental structure, provide the deformation information associated with the dental restoration patient's current dental structure as input to a neural network that is trained with the deformation information indicative of placement of dental device components relative to one or more teeth of the corresponding dental structure, the dental devices being used in the dental restoration treatment for the one or more teeth, and run the neural network using the input to produce placement information of the dental device components relative to the dental restoration patient's current dental structure.

[0011] In one example, the computing device includes an input interface and a neural network engine. The input interface is configured to receive one or more three-dimensional (3D) tooth meshes associated with a dental restoration patient's current dental structure and 3D component meshes representing generated shapes of dental prosthetic components. The neural network engine is configured to provide the one or more 3D tooth meshes and 3D component meshes received by the input interface as inputs to a neural network trained with training data including ground truth dental prosthetic component shapes and corresponding 3D tooth meshes of dental restoration cases. The neural network engine is further configured to run the neural network using the provided inputs to generate updated models of the dental prosthetic components for the dental restoration patient's current dental structure.

[0012] In another example, a method includes receiving, at an input interface, one or more three-dimensional (3D) tooth meshes associated with a dental restoration patient's current dental structure and 3D component meshes representing generated shapes of dental appliance components. The method further includes providing, by a neural network engine communicatively coupled to the input interface, the one or more 3D tooth meshes and 3D component meshes received by the input interface as inputs to a neural network trained with training data including ground truth dental appliance component shapes and corresponding 3D tooth meshes of dental restoration cases. The method further includes running, by the neural network engine, the neural network using the provided inputs to generate updated models of the dental appliance components for the dental restoration patient's current dental structure.

[0013] In another example, an apparatus includes means for receiving one or more three-dimensional (3D) tooth meshes associated with a dental restoration patient's current dental structure and 3D component meshes representing generated shapes of dental prosthetic components; means for providing the one or more 3D tooth meshes and 3D component meshes received by the input interface as inputs to a neural network trained with training data including ground truth dental prosthetic component shapes and corresponding 3D tooth meshes of dental restoration cases; and means for running the neural network using the provided inputs to produce updated models of the dental prosthetic components for the dental restoration patient's current dental structure.

[0014] In another example, a non-transitory computer-readable storage medium is encoded with instructions that, when executed, cause one or more processors of a computing system to receive one or more three-dimensional (3D) tooth meshes associated with a current dental structure of a dental restoration patient and 3D component meshes representing generated shapes of dental appliance components, provide the one or more 3D tooth meshes and 3D component meshes received by the input interface as inputs to a neural network trained with training data including ground truth dental appliance component shapes and the 3D tooth meshes of corresponding dental restoration cases, and run the neural network using the provided inputs to produce updated models of the dental appliance components for the current dental structure of the dental restoration patient.

[0015] In one example, the computing device includes an input interface and a neural network engine. The input interface is configured to receive two-dimensional (2D) images of the dental restoration patient's current dental structure. The neural network engine is configured to provide the 2D images of the dental restoration patient's current dental structure as input to a neural network trained with training data including 2D images of the pre-restoration dental structure and corresponding 2D images of the post-restoration dental structure of previously performed dental restoration cases. The neural network engine is further configured to run the neural network using the input to generate 2D images of the proposed dental structure of the dental restoration patient, the proposed dental structure being associated with a post-restoration outcome of the dental restoration plan for the dental restoration patient.

[0016] In another example, a method includes receiving a two-dimensional (2D) image of a current dental structure of a dental restoration patient. The method further includes providing the 2D image of the current dental structure of the dental restoration patient as input to a neural network trained with training data including 2D images of pre-restoration dental structures and corresponding 2D images of post-restoration dental structures of previously performed dental restoration cases. The method further includes running the neural network using the input to generate a 2D image of a proposed dental structure of the dental restoration patient, the proposed dental structure being associated with a post-restoration outcome of the dental restoration plan for the dental restoration patient.

[0017] In another example, an apparatus includes means for receiving a two-dimensional (2D) image of a current dental structure of a dental restoration patient; means for providing the 2D image of the current dental structure of the dental restoration patient as input to a neural network trained with training data including 2D images of pre-restoration dental structures and corresponding 2D images of post-restoration dental structures of previously performed dental restoration cases; and means for running the neural network using the input to generate a 2D image of a proposed dental structure of the dental restoration patient, the proposed dental structure being associated with a post-restoration outcome of the dental restoration plan for the dental restoration patient.

[0018] In another example, a non-transitory computer-readable storage medium is encoded with instructions that, when executed, cause one or more processors of a computing system to receive a two-dimensional (2D) image of a current dental structure of a dental restoration patient, provide the 2D image of the current dental structure of the dental restoration patient as input to a neural network trained with training data including 2D images of pre-restoration dental structures and corresponding 2D images of post-restoration dental structures of previously performed dental restoration cases, and run the neural network using the input to produce a 2D image of a proposed dental structure of the dental restoration patient, the proposed dental structure being associated with a post-restoration outcome of the dental restoration plan for the dental restoration patient.

[0019] In one example, the computing device includes an input interface and a neural network engine. The input interface is configured to receive one or more three-dimensional (3D) tooth meshes associated with a dental restoration patient's current dental structure. The neural network engine is configured to provide the one or more 3D tooth meshes received by the input interface as inputs to a neural network trained with training data including ground truth dental appliance component shapes and corresponding 3D tooth meshes of dental restoration cases. The neural network engine is further configured to run the neural network using the provided inputs to generate custom shapes of dental appliance components for the dental restoration patient's current dental structure.

[0020] In another example, a method includes receiving one or more three-dimensional (3D) tooth meshes associated with a dental restoration patient's current dental structure. The method further includes providing the one or more 3D tooth meshes associated with the dental restoration patient's current dental structure as inputs to a neural network trained with training data including ground truth dental appliance component shapes and corresponding dental restoration case 3D tooth meshes. The method further includes running the neural network using the provided inputs to generate custom shapes of the dental appliance components for the dental restoration patient's current dental structure.

[0021] In another example, an apparatus includes means for receiving one or more three-dimensional (3D) tooth meshes associated with a dental restoration patient's current dental structure; means for providing the one or more 3D tooth meshes associated with the dental restoration patient's current dental structure as input to a neural network trained with training data including ground truth dental prosthetic component shapes and corresponding 3D tooth meshes of dental restoration cases; and means for running the neural network using the provided input to create custom shapes of dental prosthetic components for the dental restoration patient's current dental structure.

[0022] In another example, a non-transitory computer-readable storage medium is encoded with instructions that, when executed, cause one or more processors of a computing system to receive one or more three-dimensional (3D) tooth meshes associated with a dental restoration patient's current dental structure, provide the one or more 3D tooth meshes associated with the dental restoration patient's current dental structure as input to a neural network trained with training data including ground truth dental device component shapes and corresponding 3D tooth meshes of dental restoration cases, and run the neural network using the provided input to create custom shapes of dental device components for the dental restoration patient's current dental structure.

[0023] The techniques and practical applications described herein can provide certain advantages. For example, by automatically determining the shape and placement of a 3D mesh for forming a dental prosthesis component, or a comprehensive model of a dental prosthesis, for a patient's restorative treatment, the disclosed computing system can improve data accuracy and conserve resources. For example, by generating more accurate 3D mesh components of a dental prosthesis, the disclosed computing system can improve the function and effectiveness of the dental prosthesis when used in restorative treatment.

[0024] When a computing system predicts dental appliance placement information using a neural network with fewer layers, the computing system may reduce the use of computational resources by implementing a neural network with fewer hidden layers. When a computing system generates a 3D mesh shape using a GAN, GCNN, cGAN, encoder-decoder-based CNN, U-NetCNN, GAN with PatchGAN discriminator, or other deep neural network, the computing system improves the process by reducing iterations that result from providing defective or suboptimal dental appliances to dentists when performing restorative treatment on patients. In this way, the neural network-based dental appliance configuration technique of the present disclosure improves speed, accuracy, and predictability.

[0025] Earlier and / or more accurate restoration of a patient's dental structure can improve functionality (e.g., reducing bruxism or interference between teeth) and can improve the patient's quality of life by, for example, reducing pain resulting from suboptimal dental form, integrity, and functionality. In some instances, more accurate restoration of a patient's dental structure can improve the appearance of the patient's dental structure, further improving the patient's experience and / or quality of life. Furthermore, by creating an accurate, fast, and predictable process for restoring dental structures with shapes and / or placements generated by neural networks, the computing system of the present disclosure provides increased efficiency for a wider range of dentists and improved affordability for a wider range of patients.

[0026] The details of one or more examples are set forth in the accompanying drawings and the description below. Other features, objects, and advantages will be apparent from the description and drawings, and from the claims. [Brief explanation of the drawings]

[0027] [Figure 1]FIG. 1 is a block diagram illustrating an exemplary system for designing and manufacturing dental appliances for restoring a patient's dental structure, according to various aspects of the present disclosure. [Figure 2] 2 is a flowchart illustrating an exemplary process that the system of FIG. 1 may perform to generate a digital model of a dental appliance by executing a neural network trained according to an embodiment of the present disclosure. [Figure 3] FIG. 10 is a flow diagram illustrating an exemplary use of a neural network to position library components of dental appliances, according to various aspects of the present disclosure. [Figure 4] FIG. 10 is a flow diagram illustrating an example of neural network-based component shape generation according to various aspects of the present disclosure. [Figure 5] 1 is a flowchart illustrating a process that a computing device may implement to generate component shapes using GANs, according to aspects of the present disclosure. [Figure 6] 10 is a rendering illustrating an exemplary center clip placement performed in accordance with the neural network-based placement technique of the present disclosure. [Figure 7] 10 is a rendering showing an example of a bonding pad (e.g., for a lingual bracket) customized to the shape of a corresponding tooth. [Figure 8] 1 is a rendering showing an example set of components that make up a lingual bracket. [Figure 9] FIG. 10 is a flow diagram illustrating another example of neural network-based component shape generation according to aspects of the present disclosure. [Figure 10] FIG. 1 is a conceptual diagram illustrating the symbiotic training process of a generative network and a discriminative network of a cGAN configured to render a 2D image of a patient's proposed dental structure, according to an embodiment of the present disclosure. [Figure 11A] 1 shows the inputs and outputs of a cGAN-trained generative network configured to generate 2D images of proposed dental structures using 2D renderings of a patient's current dental structures. [Figure 11B]1 shows a comparison between a current dental structure image, a proposed dental structure image of the present disclosure, and a ground truth restoration image. [Figure 12] 1 illustrates a menu that the computing device of FIG. 1 may display as part of a graphical user interface (GUI) that includes current and / or proposed dental structure images of the present disclosure. [Figure 13A] 1 is a conceptual diagram illustrating an exemplary mold parting surface according to various aspects of the present disclosure. [Figure 13B] 1 is a conceptual diagram illustrating an exemplary mold parting surface according to various aspects of the present disclosure. [Figure 14] 1 is a conceptual diagram illustrating an exemplary gingival trim surface according to various aspects of the present disclosure. [Figure 15] FIG. 1 is a conceptual diagram illustrating an exemplary facial ribbon, according to various aspects of the present disclosure. [Figure 16] FIG. 1 is a conceptual diagram illustrating an exemplary lingual shelf, according to various aspects of the present disclosure. [Figure 17] FIG. 1 is a conceptual diagram illustrating an exemplary door and window according to various aspects of the present disclosure. [Figure 18] FIG. 1 is a conceptual diagram illustrating an exemplary rear snap clamp according to various aspects of the present disclosure. [Figure 19] FIG. 1 is a conceptual diagram illustrating an exemplary door hinge according to various aspects of the present disclosure. [Figure 20A] FIG. 1 is a conceptual diagram illustrating an exemplary door snap according to various aspects of the present disclosure. [Figure 20B] FIG. 1 is a conceptual diagram illustrating an exemplary door snap according to various aspects of the present disclosure. [Figure 21] FIG. 1 is a conceptual diagram illustrating an exemplary incisal ridge, according to various aspects of the present disclosure. [Figure 22] 1 is a conceptual diagram illustrating an exemplary center clip according to various aspects of the present disclosure. [Figure 23] FIG. 1 is a conceptual diagram illustrating an exemplary door vent according to various aspects of the present disclosure. [Figure 24] FIG. 1 is a conceptual diagram illustrating an exemplary door, according to various aspects of the present disclosure. [Figure 25] FIG. 1 is a conceptual diagram illustrating an exemplary dental space matrix, according to various aspects of the present disclosure. [Figure 26] 1 is a conceptual diagram illustrating an exemplary manufacturing case frame and an exemplary dental appliance according to various aspects of the present disclosure. [Figure 27] FIG. 1 is a conceptual diagram illustrating an exemplary dental appliance including a custom label, according to various aspects of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0028] 1 is a block diagram illustrating an exemplary system for designing and manufacturing dental appliances for restoring a patient's dental structure, according to various aspects of the present disclosure. In the example of FIG. 1, system 100 includes a clinic 104, an appliance design facility 108, and a manufacturing facility 110.

[0029] A dentist 106 may treat the patient 102 at the clinic 104. For example, the dentist 106 may create a digital model of the patient's 102's current dental structure. The dental structure may include the crown or root of one or more teeth in the dental arch, gums, periodontal ligament, alveolar bone, cortical bone, bone grafts, implants, endodontic fillings, artificial crowns, bridges, veneers, dentures, orthodontic appliances, or any portion of any structure (natural or synthetic) that may be considered part of the patient's 102 pre-treatment, treatment, or treatment dental structure.

[0030] In one example, the digital model of the current dental structure includes a three-dimensional (3D) model of the current (pre-treatment) dental structure of the patient 102. The clinic 104 may, in various examples, include an intraoral scanner, a cone-beam computed tomography (CBCT) scanning (e.g., 3D X-ray) device, an optical coherence tomography (OCT) device, a magnetic resonance imaging (MRI) machine, or any other 3D image capture system that the dentist 106 may use to generate a 3D model of the dental structure of the patient 102.

[0031] 1 , the clinic 104 includes a computing system 190. Computing system 190 may represent a single device or a group of securely interconnected devices. In these examples, the individual devices of computing system 190 may form a secure interconnection by being contained entirely within the logical domain of the clinic 104 (e.g., by physical connections within the clinic 104, such as using a local area network or “LAN”) and / or by virtual private network (VPN) tunneling-based encrypted communications communicated securely over a public network such as the Internet. Computing system 190 may include one or more user-facing computing devices, such as a personal computer (e.g., a desktop computer, a laptop computer, a netbook, etc.), a mobile device (e.g., a tablet computer, a smartphone, a personal digital assistant, etc.), or any other electronic device configured to provide end-user computing capabilities, such as by presenting resources in a human-understandable form (e.g., visual images such as medical / dental imaging, readable output, symbolic / depictive output, audible output, tactile output, etc.).

[0032] The dental practitioner 106 can store a digital model of the patient's 102's current dental structure in a storage device included in or read / write accessible via the computing system 190. In some examples, the computing device 190 can also store a digital model of a proposed dental structure for the patient 102. The proposed dental structure represents the intended function, integrity, and form of the dental structure to be achieved by applying the dental appliances 112 as part of the patient's 102's restorative dental treatment.

[0033] In one example, the dental practitioner 106 may generate a physical model of the proposed dental structure and generate a digital model of the proposed dental structure using an image capture system (e.g., as described above). In another example, the dental practitioner 106 may perform modifications to the digital model of the patient's 102's current structure (e.g., by adding material to the surface of one or more teeth of the dental structure or by other methods) to generate a digital model of the proposed dental structure for the patient 102. In yet another example, the dental practitioner 106 can use the computing system 190 to modify the digital model of the patient's 102's current dental structure to generate a model of the patient's 102's proposed dental structure.

[0034] In one scenario, computing device 190 outputs digital model(s) representing patient 102's current and / or proposed dental structure to another computing device, such as computing device 150 and / or computing device 192. While described herein as executing locally on computing systems 190, 150, and 192, it will be understood that in some examples, one or more of computing systems 190, 150, and 192 can leverage cloud computing capabilities and / or software-as-a-service (SaaS) functionality to perform the underlying processing for the functionality described herein. As shown in FIG. 1 , in some examples, computing device 150 at design facility 108, computing device 190 at clinic 104, and computing device 192 at manufacturing facility 110 may be communicatively coupled to one another via network 114. Network 114, in various examples, can represent or include a federation (e.g., a dental services network, etc.) or a private network associated with another entity or group of entities.

[0035] In other examples, network 114 may represent or include a public network such as the Internet. While shown as a single entity in FIG. 1 for ease of explanation only, it will be understood that network 114 may include a combination of multiple public and / or private networks. For example, network 114 may represent a private network implemented using a public network infrastructure, such as a VPN tunnel implemented over the Internet. Thus, network 114 may include one or more of a wide area network (WAN) (e.g., the Internet), a LAN, a VPN, and / or another wired or wireless communication network. Network 114 may include wired or wireless network components conforming to one or more standards, such as over Ethernet, Wi-Fi™, Bluetooth, 3G, 4G LTE, 5G, etc.

[0036] 1 , computing device 150 is implemented by or in design facility 108. Computing device 150 is configured to automatically design dental appliances and / or generate dental appliance placement information for reshaping dental structures of patient 102. Computing device 150 (or components thereof) implements neural network techniques to determine the shape and / or placement information of the dental appliances. In the example shown in FIG. 1 , computing device 150 includes one or more processors 172, one or more user interface (UI) devices 174, one or more communication units 176, and one or more storage devices 178.

[0037] UI device(s) 174 may be configured to receive input data from a user of computing device 150 and / or provide output data to a user of computing device 150. One or more input components of UI device(s) 174 may receive input. Examples of input include tactile input, voice input, kinetic input, and optical input, to name just a few. For example, UI device(s) 174 may include one or more of a mouse, a keyboard, a voice input system, an image capture device (e.g., still camera and / or video camera hardware), physical or logical buttons, a control pad, a microphone or microphone array, or any other type of device for detecting input from a human user or another machine. In some examples, UI device 174 may include one or more presence-responsive input components, such as a resistive screen, a capacitive screen, a single-finger or multi-finger touchscreen, a stylus-sensitive screen, etc.

[0038] The output components of UI device 174 may output one or more of visual (e.g., symbolic / representative or readable) data, tactile feedback, audio output data, or any other output data understandable to a human user or another machine. The output components of UI device 174, in various examples, include one or more of a display device (e.g., a liquid crystal display (LCD) display, a touch screen, a stylus-sensitive screen, a light-emitting diode (LED) display, an optical head-mounted display (HMD), among others), a loudspeaker or loudspeaker array, a headphone or headphone set, or any other type of device capable of generating output data in a form understandable by a human or a machine.

[0039] Processor(s) 172 represent one or more types of processing hardware (e.g., processing circuitry), such as general-purpose microprocessor(s), specially designed processor(s), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), discrete logic, fixed-function circuitry, a collection of programmable circuitry (or a combination of fixed-function circuitry and programmable circuitry), or any type of processing hardware capable of executing instructions that implement the techniques described herein.

[0040] For example, storage device(s) 178 may store program instructions (e.g., software instructions or modules) that are executed by processor(s) 172 to perform the techniques described herein. In other examples, the techniques may be performed by specifically programmed circuitry of processor 172 (e.g., in the case of fixed-function circuitry or specifically programmed programmable circuitry). In these or other ways, processor(s) 172 may be configured to perform the techniques described herein, possibly by utilizing instructions and other data accessible from storage device(s) 178.

[0041] The storage device(s) 178 can store data for processing by the processor(s) 172. Some portions of the storage device(s) 178 represent temporary memory, meaning that the primary purpose of these portions of the storage device(s) 178 is not long-term storage. The short-term memory aspects of the storage device(s) 178 may include volatile memory that is not configured to retain stored contents when deactivated and reactivated (e.g., as in the case of a power cycle). Examples of volatile memory include random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), and other forms of volatile memory known in the art. In some examples, the short-term memory (e.g., RAM) of the storage device(s) 178 may include on-chip memory unit(s) collocated with a portion of the processor(s) 172 and may be formed as part of an integrated circuit (IC) or part of a system on a chip (SoC).

[0042] In some examples, the storage device(s) 178 may also include one or more computer-readable storage media. The storage device(s) 178 may be configured to store more data than volatile memory. The storage device(s) 178 may also be configured for long-term storage of data as non-volatile memory space, retaining data after activation / deactivation cycles. Examples of non-volatile memory include solid state drives (SSDs), hard disk drives (HDDs), flash memory, or forms of electrically programmable memories (EPROMs) or electrically erasable and programmable memories (EEPROMs). The storage device(s) 178 may store program instructions and / or data associated with the software configuration and / or operating system of the computing device 150.

[0043] 1 , storage device(s) 178 include appliance feature library 164, model library 166, and physician selection library 168 (collectively, libraries 164-168). Libraries 164-168 may include relational databases, multidimensional databases, maps, hash tables, or any other data structure. In one example, model library 166 includes 3D models of the patient's current and / or proposed dental structures. In some cases, libraries 164-168 may be stored locally on computing device 150 or accessed via networked file shares, cloud storage, or other remote data stores accessible using the network interface hardware of communication unit(s) 176.

[0044] The short-term memory of storage device(s) 178 and processor(s) 172 may collectively provide a computing platform for executing operating system 180. Operating system 180 may represent, for example, an embedded real-time multitasking operating system, or any other type of operating system. Operating system 180 provides a multitasking operating environment for executing one or more software components 182-186. In some examples, operating system 180 may execute any of components 182-188 as an instance of a virtual machine or within a virtual machine instance running on underlying hardware. While shown separately from operating system 180 as a non-limiting example, it will be understood that any of components 182-188 may be implemented as part of operating system 180 in other examples.

[0045] In accordance with the techniques of the present disclosure, the computing device 150 uses one or more types of neural networks trained with dental structure-related data and / or appliance feature data associated with the patient 102 and / or other patients (real or virtual) to automatically or semi-automatically generate a digital model of the dental appliances 112 for generating the dental structure of the patient 102. The pre-processor 182 is configured to pre-process the digital model of the proposed dental structure of the patient 102.

[0046] In one example, pre-processor 182 performs pre-processing to identify one or more teeth in the proposed dental structure of patient 102. In some cases, pre-processor 182 may identify a local coordinate system for each individual tooth and identify a global coordinate system that includes each tooth of the proposed dental structure (e.g., in one or both arches of the proposed dental structure). As another example, pre-processor 182 may pre-process a digital model of the proposed dental structure to identify the root structures of the dental structure.

[0047] In another example, pre-processor 182 may identify the gingiva of the gums in the proposed dental structure, thereby identifying and delineating the portion of the proposed dental structure that includes the gums and the portion of the proposed dental structure that includes the teeth. As yet another example, pre-processor 182 may pre-process the digital model of the proposed dental structure by extending the tooth roots and identifying the root apex of each tooth. Pre-processor 182 may perform one, some, or all of the various exemplary functions described above in various use case scenarios, depending on requests made by dental practitioner 106 based on the availability of data regarding patient 102 and / or other patient(s), and potentially other factors.

[0048] Computing device 150 (or its hardware / firmware components) can invoke or invoke neural network engine 184 to determine placement information for dental appliances 112 during dental restorative treatment for patient 102. In some examples, neural network engine 184 may implement a two-hidden layer neural network trained with placement information for patient 102 and / or other patients having generally corresponding (current or proposed) dental structures.

[0049] In these examples, the neural network engine 184 can implement a neural network to accept as input individual position / orientation information for each two teeth (e.g., a pair of adjacent teeth) in the patient's 102's current dental structure and can output placement information for the dental appliances 112 during dental restorative treatment. For example, the placement information may directly or indirectly reflect one or more of the position, orientation, or size of the dental appliances 112 to be used in the patient's 102's dental restorative treatment.

[0050] In one example, the neural network engine 184 may train a neural network using a back-propagation algorithm using a single 4x4 transformation for each two adjacent teeth and another 4x4 transformation that identifies a "ground truth" {position, orientation, size} tuple for the dental appliance 112 after placement is completed as part of the dental restorative treatment of the patient 102. As used herein, the term "transformation" refers to change (or "delta") information related to a {position, orientation, size} tuple, and therefore may also be described as a {translation, rotation, scale} tuple for the dental appliance 112.

[0051] In some cases, the deformations of the present disclosure may also include additional elements, such as a shear map (or simply "shear") associated with the dental appliance 112. Thus, the deformations of the present disclosure may represent affine deformations in various instances, and may include some or all of the deformations that fall within the deformations possible under an automorphism in affine space.

[0052] The neural network engine 184 can extract specific tooth deformations from 3D mesh data describing the current and / or proposed dental structure of the patient 102. As used herein, the term "ground truth" refers to a proven or otherwise well-founded description of a dental structure feature or appliance feature. Thus, in some examples, the ground truth deformations may be manually created by the dentist 106 or a technician using CAD tools.

[0053] In other examples, the ground truth deformations can be generated automatically, such as by using the automated techniques described in International Publication No. WO 2020 / 240351, filed May 20, 2020, the entire contents of which are incorporated herein by reference. Various techniques of the present disclosure are described below with respect to the non-limiting example of determining the position, orientation, and sizing of a central clip to be placed over the gap between two adjacent teeth during a dental restoration. However, it will be understood that the neural network engine 184 may also implement the techniques of the present disclosure to generate shape and / or placement information for other types of dental appliances.

[0054] As part of generating the central clip placement information, the neural network engine 184 can identify landmarks of the proposed dental structure. Exemplary landmarks include a slice, a midpoint, a gingival margin, a nearest point between two adjacent teeth (e.g., a contact point or closest point (or nearest point) between adjacent teeth), a convex hull, a center of mass, or other landmarks. A slice refers to a cross-section of the dental structure. A tooth midpoint refers to the geometric center (also called the geometric midpoint) of a tooth within a given slice.

[0055] The gingival boundary refers to the boundary between the gingiva and one or more teeth of a dental structure. The convex hull refers to a polygon whose vertices contain a subset of vertices in a given set of vertices, and the boundary of the subset of vertices circumscribes the full set of vertices. The center of mass of a tooth refers to the midpoint, center point, centroid, or geometric center of the tooth. In some cases, the neural network engine 184 can determine one or more of these landmarks expressed using the local coordinate system of each tooth.

[0056] In some examples, the neural network engine 184 determines multiple slices of the patient's proposed dental structure as part of identifying landmarks. In one example, the thickness of each slice is the same. In some cases, the thickness of one or more slices is different from the thickness of another slice. The thickness of a given slice may be predefined. In one example, the neural network engine 184 automatically determines the thickness of each slice using a simplified neural network of the present disclosure. In another example, the thickness of each slice may be user-defined, e.g., available as a ground truth input to the simplified neural network.

[0057] As part of identifying the landmarks, the neural network engine 184, in some examples, can determine the midpoint of each tooth related to the placement of the dental appliance 112. In one example, the neural network engine 184 identifies the landmarks using the midpoint of a particular tooth by calculating the shape extrema of the particular tooth based on the entire tooth (e.g., without dividing the dental structure into slices) and determining the midpoint of the particular tooth based on the shape extrema.

[0058] In some examples, the neural network engine 184 may determine the midpoint of each tooth for each slice. For example, the neural network engine 184 may determine the midpoint of a particular slice of a particular tooth by calculating the center of mass of the collection of vertices surrounding the edge of the particular tooth for that slice. In some cases, the midpoint of a particular tooth for a particular slice may be biased toward one edge of the tooth (e.g., if one edge has more points than another edge).

[0059] In other examples, the neural network engine 184 can determine the midpoint of a particular tooth in a particular slice based on the convex hull of the particular tooth in the particular slice as part of the landmark identification portion of the geometry generation. For example, the neural network engine 184 may determine the convex hull of a set of edge points of the teeth in a given slice. In some examples, the neural network engine 184 executes a neural network that determines the geometric center from the convex hull by performing a fill operation on the area enclosed by the convex hull and calculating the center of mass of the filled convex hull as part of identifying the landmarks.

[0060] In some examples, the neural network executed by the neural network engine 184 outputs the nearest point between two adjacent teeth. The nearest point between two adjacent teeth may be a contact point or a closest point. In one example, the neural network engine 184 determines the nearest point between two adjacent teeth for each slice. In another example, the neural network engine 184 determines the nearest point between two adjacent teeth based on the entire adjacent teeth (e.g., without dividing the dental structure into slices).

[0061] Using the landmarks calculated for the proposed dental structure, a neural network executed by the neural network engine 184 generates one or more custom appliance features for the dental appliance 112 based at least in part on the landmarks. For example, the custom feature generator 184 may generate the custom appliance features by determining characteristics of the custom appliance feature, such as the size, shape, position, and / or orientation of the custom appliance feature. Examples of custom appliance features include splines, mold parting surfaces, gingival trim surfaces, shells, facial ribbons, lingual shelves (also called "reinforcement ribs"), doors, windows, incisal ridges, case frame sparing, and interdental matrix wrapping, among others.

[0062] In some examples, neural network engine 184 may identify and use features other than those listed above. For example, neural network engine 184 may identify and use features that processor(s) 172 recognize and act upon within the mathematical framework of the neural network being executed. Thus, the operations performed via the neural network executed by neural network engine 184 may represent a "black box" in terms of the features used and the mathematical framework applied by the neural network during execution.

[0063] A spline refers to a curve that passes through multiple points or vertices, such as a piecewise polynomial parametric curve. A mold parting surface refers to a 3D mesh that bisects two sides of one or more teeth (e.g., separating the facial side of one or more teeth from the lingual side of one or more teeth). A gingival trim surface refers to a 3D mesh that trims the surrounding shell along the gingival margin. A shell refers to an object with a nominal thickness. In some examples, the inner surface of the shell coincides with the surface of the dental arch, and the outer surface of the shell is a nominal offset of the inner surface.

[0064] A facial ribbon refers to a reinforcing rib having a nominal thickness offset facially from the shell. A window refers to an aperture that provides access to the tooth surface so that a dental composite can be placed on the tooth. A door refers to a structure that covers the window. An incisal ridge reinforces the incisal edge of the dental appliance 112 and may be derived from the dental archform. A case frame sparing refers to a bonding material that bonds components of the dental appliance 112 (e.g., the lingual portion of the dental appliance 112, the facial portion of the dental appliance 112, and its subcomponents) to the manufacturing case frame. In this way, the case frame sparing can join components of the dental appliance 112 to the case frame during manufacturing, protecting various components from breakage or loss and / or reducing the risk of component mix-up.

[0065] In some examples, the neural network executed by the neural network engine 184 generates one or more splines based on landmarks. The neural network executed by the neural network engine 184 may generate splines based on midpoints of multiple teeth and / or nearest points between adjacent teeth (e.g., contact points between adjacent teeth or nearest points between adjacent teeth). In some cases, the neural network executed by the neural network engine 184 generates one spline for each slice. In one example, the neural network engine 184 generates multiple splines for a given slice. For example, the neural network engine 184 may generate a first spline for a first subset of teeth (e.g., right molars), a second spline for a second subset of teeth (e.g., left molars), and a third spline for a third subset of teeth (e.g., front teeth).

[0066] In some scenarios, the neural network engine 184 generates mold parting surfaces based on landmarks. In molding without undercuts, the mold parting surfaces can be used to split the surrounding shell. In some examples, the neural network engine 184 generates additional copies of the mold parting surfaces. For example, the neural network executed by the neural network engine 184 can position one or more copies of the mold parting surfaces slightly offset relative to the main parting surfaces in order to create interference conditions when the appliances are assembled (this can, for example, improve form-fitting and sealing when applying dental restorative materials to the teeth).

[0067] The appliance feature library 164 includes a set of predefined appliance features that can be included in the dental appliance 112. The appliance feature library 164 may include a series of predefined appliance features that define one or more functional characteristics of the dental appliance 112. Examples of predefined appliance features include vents, rear snap clamps, door hinges, door snaps, incisal alignment features, center clips, custom labels, manufacturing case frames, and interdental matrix handles, among others. Each vent is configured to allow excess dental composite to flow out of the dental appliance 112.

[0068] The rear snap clamp is configured to couple a facial portion of the dental appliance 112 to a lingual portion of the dental appliance 112. Each door hinge is configured to pivotally couple a respective door to the dental appliance 112. Each door snap is configured to secure a respective door in a closed position. In some examples, the incisal alignment feature comprises a pair of male and female tabs located on the incisal edge of the dental appliance 112 (e.g., along the midsagittal plane). In one example, the incisal alignment feature is used to maintain vertical alignment between the facial portion of the dental appliance 112 and the lingual portion of the dental appliance 112.

[0069] Each central clip is configured to provide vertical alignment between a lingual portion of the dental appliance 112 and a facial portion of the dental appliance 112. Each custom label includes data identifying a component of the dental appliance 112. The manufacturing case frame is configured to support one or more components of the dental appliance 112. For example, the manufacturing case frame may removably couple the lingual portion of the dental appliance 112 and the facial portion of the dental appliance 112 to one another for safe handling and transportation of the dental appliance 112 from the manufacturing facility 110 to the clinic 104.

[0070] The neural network executed by the neural network engine 184, in some examples, can determine characteristics of one or more predefined appliance features included in the predefined appliance feature library 164. For example, the one or more features accessible from the predefined appliance feature library 164 can represent component shapes obtained in one or more ways, such as by manual generation (e.g., by the dental practitioner 106 or via automated generation, such as via the techniques described in the aforementioned International Publication WO 2020 / 240351, filed May 20, 2020). Based on availability and relevance to the patient's 106's current dental structure, the neural network can be trained (at least in part) using information available from the predefined appliance feature library 164.

[0071] In one example, the predefined appliance features are configured to enable or perform functions ascribed to the dental appliance 112. The characteristics of the predefined appliance features may include one or more of the deformation-related attributes described above (e.g., position, orientation, size) and / or other attributes, such as shape information. The neural network executed by the neural network engine 184 may determine the characteristics of the predefined appliance features based on one or more rules, such as rules generated and / or refined via machine learning (ML) techniques.

[0072] In some examples, the neural network engine 184 executes and determines placement information for the posterior snap clamps based on rules. In one example, the neural network engine 184 can generate placement information for positioning two posterior snap clamps along a dental arch during dental restoration treatment, with the two posterior snap clamps being placed at opposite ends of the dental arch. For example, a first snap clamp can be placed at one end of the dental arch during dental restoration treatment, and a second snap clamp can be placed at the other end of the same dental arch during dental restoration treatment.

[0073] In some examples, the neural network engine 184 can assign one or both of the posterior snap clamps a position one tooth beyond the furthest tooth to be restored. In some examples, the neural network engine 184 places the female part of the posterior snap clamp on the lingual side of the parting plane and the male part of the posterior snap clamp on the facial side. In some examples, the neural network engine 184 determines placement information for vents during dental restorative treatment based on rules. For example, the neural network engine 184 can assign a vent to a position midline of a corresponding door on the cutting side of the dental appliance 112.

[0074] In some scenarios, the neural network engine 184 determines the placement of the door hinges based on rules. In one scenario, the neural network engine 184 assigns each door hinge a position on the centerline of its corresponding door. In another scenario, the neural network engine 184 determines a positioning that positions the female portion of the door hinge to secure to a facial portion of the dental appliance 112 (e.g., toward the incisal edge of the tooth) and the male portion of the door hinge to secure to an exterior surface of the door.

[0075] In one example, the neural network engine 184 determines the placement of the door snaps based on the rules by placing the door snaps along the centerline of the corresponding door. In one example, the neural network engine 184 determines the placement of the female part of the door snap to be secured to the exterior surface of the door and extend downward toward the gums. In another example, the neural network engine 184 determines the placement of the male part of the door snap to be secured to the gum side of the facial ribbon. For example, the door snap can secure the door in a closed position by clipping the male part of the door snap to the facial ribbon.

[0076] The neural network engine 184 may determine characteristics of the pre-defined appliance features based on the selections of the dental practitioner 106. The practitioner's selection library 168 may include data indicative of one or more dental practitioner's 106 selections. Thus, the neural network engine 184 may determine placement or shape information for the dental appliance 112 using information from the practitioner's selection library 168 as training data in performing comprehensive training of the neural network associated with the dental practitioner 106.

[0077] Practitioner preferences can directly affect the characteristics of one or more appliance features of the dental appliance 112 in various use case scenarios. For example, the practitioner's preferences library 168 may include data indicating preferred sizes for various appliance features, such as the size of a vent. In some such instances, a larger vent may allow the pressure of the dental composite or resin to reach equilibrium more quickly during the filling process, but may result in a larger protuberance to be surfaced after hardening. In these instances, the neural network engine 184 may train a neural network with scaling information that determines the size of the dental appliance 112 according to the preferences attributed to the dental practitioner 106.

[0078] As another example, practitioner preferences indirectly influence the characteristics of appliance features. For example, the practitioner's preference library 168 may include data indicating a preferred stiffness of an appliance or a preferred tightness of a self-clamping feature. Such preference preferences may also influence more complex design modifications to the cross-sectional thickness of the matrix and / or the degree of activation of the clamp geometry. The neural network engine 184 can determine the characteristics of appliance features by extending rules based on which the implemented neural network is trained using the dental practitioner's 106 (or possibly other dental professionals) preferences available from the practitioner's preference library 168. In some examples, the neural network engine 184 can extend the rules with practitioner-selected data based on (e.g., Monte Carlo) simulations or finite element analyses performed using the practitioner-selected information. In some examples, feature characteristics can also be derived from the nature of the material used in the matrix, such as the type of composite the dentist prefers to use in the appliance.

[0079] The model assembler 186 uses the output of the neural network executed by the neural network engine 184 to generate a digital 3D model of the dental appliances 112 to be used to reshape the dental structure of the patient 102 (e.g., to shape the current dental structure into a proposed dental structure). In various examples, the model assembler 186 may generate the digital 3D model using custom and / or predefined appliance features that form the output of the neural network executed by the neural network engine 184. The digital 3D model of the dental appliances 112 may include, be one or more of, or be a portion of, one or more of a point cloud, a 3D mesh, or other digital representation of the dental appliances 112. In some cases, the model assembler 186 stores the digital model of the dental appliances 112 in the model library 166.

[0080] The model assembler 186 can output the digital model of the dental appliance 112 in various ways. In one example, the model assembler 186 can output the digital 3D model of the dental appliance 112 to a computing device 192 at the manufacturing facility 110 (e.g., via the network 114 using network interface hardware of the communication unit(s) 176). By providing the digital 3D model to the computing device 192, the model assembler 186 can enable one or more entities at the manufacturing facility 110 to manufacture the dental appliance 112. In other examples, the computing device 150 can send the digital model of the dental appliance 112 to a computing device 190 at the clinic 104. In these examples, the model assembler 186 can enable a dental practitioner 106 or other entity at the clinic 104 to manufacture the dental appliance 112 on-site at the clinic 104.

[0081] In some examples, the computing device 192 can invoke the network interface hardware of the communication unit(s) 176 to send the digital 3D model of the dental appliance 112 to the manufacturing system 194 over the network 114. In these examples, the manufacturing system 194 manufactures the dental appliance 112 according to the digital 3D model of the dental appliance 112 formed by the model assembler 186. The manufacturing system 194 may form the dental appliance 112 using any number of manufacturing techniques, such as 3D printing, chemical vapor deposition (CVD), thermoforming, injection molding, lost-wax casting, milling, machining, laser cutting, among others.

[0082] The dental practitioner 106 can receive the dental appliance 112 and utilize the dental appliance 112 to reshape one or more teeth of the patient 102. For example, the dental practitioner 106 can apply dental composite to the surface of one or more teeth of the patient 102 through one or more doors of the dental appliance 112. The dentist 106 or another clinician at the clinic 104 can remove excess dental composite through one or more vents.

[0083] In some examples, the model assembler 186 can store the generated digital 3D model of the dental appliance 112 in the model library 166. In these examples, the model library 166 can provide appliance model heuristics that the neural network engine 184 can use as training data when training one or more neural networks. In some examples, the model library 166 includes data indicative of appliance success criteria associated with each completed instance of the dental appliance 112. The neural network engine 184 can augment the neural network training data set with any appliance success criteria available from the model library 166. The appliance success criteria may be indicative of one or more of manufacturing print yield, physician feedback, patient feedback, customer feedback or ratings, or a combination thereof.

[0084] For example, the neural network 184 can train the neural network to use appliance success criteria determined for previously generated dental appliances to generate new or updated placement profiles and / or shapes for the dental appliances 112 via the digital 3D model. As part of the training, the neural network executed by the neural network engine 184 can determine whether the appliance success criteria meet one or more threshold criteria, such as one or more of a threshold manufacturing yield, a threshold physician-provided rating, a threshold patient satisfaction rating, etc.

[0085] In one example, the existing digital 3D model available from model library 166 is a template or reference digital model. In such an example, neural network engine 184 can train a neural network based in part on the template digital model. The template digital model may, in various examples, be associated with different characteristics of the patient's 102's current dental structure, such as a template for a patient with small teeth or an obstacle to opening their mouth beyond a certain width.

[0086] In one example, neural network engine 184 trains a neural network using previously generated digital 3D models available from model library 166. For example, neural network engine 184 utilizes one or more morphing algorithms to adapt previously generated digital 3D models accessed from model library 166 to a situation represented by fitting dental restorative treatment to the dentition of patient 102 during neural network training and / or execution.

[0087] For example, the neural network engine 184 may utilize a morphing algorithm to interpolate appliance feature shapes and / or generate a new digital model of the dental appliance 112 based on the design of the existing digital model. In one example, the design features of the existing digital model may include windows inset from the periphery to enable the neural network engine 184 to morph the shape of the existing digital model based on landmarks of different dental structures.

[0088] The neural network engine 184 performs (and potentially compresses) multiple intermediate steps in the process of training and running a neural network to generate a digital 3D model of the dental appliance 112 for purposes of placement and shape. The neural network engine 184 uses a 3D mesh of the patient's 102's current and / or proposed dental structure to generate a set of features that describe the dental appliance 112. The 3D mesh (or "tooth mesh") and, in examples where available, library components form the training input to the neural network.

[0089] As described above, the neural network engine 184 can train a neural network to automate one or both of component placement and / or shape generation of components of the dental appliance 112. Examples of components include center clip alignment tabs (or "beaks"), door hinges, door snaps, door vents, rear snap clamps, and various others. Examples of placement-related elements and / or components that the neural network engine 184 can generate include parting surfaces, gingival trim, doors, windows, facial ribbons, incisal ridges, lingual shelves, interdental matrices, case frames, part label(s), etc.

[0090] The neural network engine 184 implements a neural network to automate the positioning and / or shape generation operations of components, such as door hinges and center clips, that must be placed in specific locations relative to the 3D representation of the teeth of the patient's 102 dental structure in order to perform the dental restoration. Leveraging neural network technology to automate the positioning and / or shape information of these components significantly shortens the process turnaround for using the dental appliance 112 during the patient's 102 dental restoration treatment.

[0091] Additionally, by utilizing neural networks trained with the combination of datasets described in this disclosure, the neural network engine 184 automates placement and / or shape generation with improved consistency and accuracy, such as by updating the neural network training based on ongoing feedback information or other dynamically changing factors.

[0092] The disclosed neural network-based automated algorithms implemented by computing device 150 provide several advantages in the form of technological improvements in the art of restorative dental appliance construction. As an example, computing device 150 can invoke neural network engine 184 to generate placement and / or shape information for dental appliance 112 without having to explicitly calculate tooth shape landmarks at each instance.

[0093] Alternatively, the neural network engine 184 may execute a neural network trained on the various deformation data and / or ground truth data described above to generate placement and / or shape information based on these training factors. As another example, the computing device 150 may improve the data accuracy associated with the digital 3D model of the dental appliance 112 by continuously improving the output of the neural network based on treatment plans and results from previous patients, feedback from the dental practitioner 106 and / or the patient 102, and other factors that may be used to fine-tune the neural network using ML techniques.

[0094] As another example, the techniques of the present disclosure provide reusability and can also calculate improvements in resource sustainability because neural network engine 184 can implement further improvements to the algorithms through the introduction of new training data rather than modifying rule-based logic. Although primarily described with respect to appliances used in restorative dental treatment, it will be appreciated that neural network engine 184, in other examples, can also be configured to implement the algorithms of the present disclosure to generate shape and / or placement information for other types of dental appliances, such as orthodontic appliances, surgical guides, and bracket bonding templates.

[0095] While computing device 150 is described herein as performing both the training and execution of the various neural networks of the present disclosure, it will be understood that in various use case scenarios, the training of a neural network may be performed by a device or system separate from the device that executes the trained neural network. For example, a training system may use some or all of the training data described with respect to FIG. 1 in the form of a labeled training data set to form one or more trained models. Other devices may import the trained model(s) and execute the trained model(s) to produce the various neural network output(s) described above.

[0096] 2 is a flowchart illustrating an example process 200 that system 100 may perform to generate a digital model of a dental appliance by executing a neural network trained according to aspects of the present disclosure. Process 200 may begin with a training phase 201. As part of training phase 201, neural network engine 184 may train the neural network using deformation data associated with 3D models of various dental structures (202). According to various aspects of the present disclosure, neural network engine 184 may train the neural network using a back-propagation training technique.

[0097] In some examples, the neural network engine 184 may use one 4x4 deformation for each one or more teeth of the dental structure and one 4x4 deformation that defines a ground truth {position, orientation, size} tuple for the dental appliance 112 after complete placement. In the example of a dental appliance 112 representing a central clip, the neural network engine 184 may extract deformations for a pair of upper central incisors (shown as teeth "8 and 9" in the Universal Notation System for Permanent Teeth) or a pair of lower central incisors (shown as "24 and 25" in the Universal Notation System for Permanent Teeth) from the 3D mesh data describing the proposed dentition of the patient 102.

[0098] The ground truth deformations may represent "pristine" data manually produced by an engineer using computer-aided design (CAD) tools to determine the position and orientation of the center clip, or may represent pristine data automatically generated in various ways, such as by using the automated techniques described in International Publication WO 2020 / 240351, filed May 20, 2020. The neural network engine 184 trains a neural network using a backpropagation algorithm to generate multiple fully connected layers, i.e., there is a weighted connection between a given node in the first layer and each of the nodes in the next layer.

[0099] The backpropagation algorithm implemented by neural network engine 184 adjusts the weights of these inter-layer node-to-node connections over the course of training the neural network, thereby gradually encoding the desired logic into the neural network over the course of multiple training iterations or passes. In some examples of the present disclosure, the neural network may include two layers, thereby reducing computational overhead for both training and ultimate execution.

[0100] Although described herein with respect to training a neural network with variations of two teeth using data from one or more past cases, it will be understood that in other examples, neural network engine 184 may train the neural network with different types and / or amounts of training data. Augmenting the training data may depend on the availability and accessibility of such training data. For example, if accessible from model library 166 or another source, neural network engine 184 may augment the training data for training the neural network with variations of other teeth in the dental archform to which dental appliances 112 are applied and / or with variations of one or more teeth in the opposing dental archform.

[0101] In this manner, neural network engine 184 can train the neural network using training data that enables the neural network to determine positioning information for dental appliances 112 based on a more holistic assessment of the dental structure of patient 102. In these and / or other examples, neural network engine 184 can augment the training data with relevant selection information available from practitioner selection library 168, with patient feedback information, and / or with various other relevant data accessible to computing device 150.

[0102] After completing the training phase 201 of the process 200, the neural network engine 184 may begin the execution phase 203. The execution phase 203 may begin when the computing device 150 receives (204) a digital 3D model of a proposed (e.g., post-dental restorative treatment) dental structure for the patient 102. In one example, the computing device 150 receives the digital 3D model of the proposed dental structure from another computing device, such as a computing device 190 at the clinic 104. The digital 3D model of the proposed dental structure for the patient 102 may include a point cloud or a 3D mesh of the proposed dental structure.

[0103] A point cloud includes a collection of points that represent or define an object in three-dimensional space. A 3D mesh includes multiple vertices (also called points) and geometric faces (e.g., triangles) defined by the vertices. In one example, the dental practitioner 106 generates a physical model of the proposed dental structure for the patient 102 and uses an image capture system to generate a digital 3D model of the proposed dental structure from images captured from the physical model. In another example, the dental practitioner 106 modifies the digital 3D model of the current dental structure for the patient 102 (e.g., by simulating adding material to the surface of one or more teeth in the dental structure or by simulating other changes) to generate a digital 3D model of the proposed dental structure. In yet another example, the computing device 190 may modify the digital model of the current dental structure to generate a model of the proposed dental structure.

[0104] In some examples, pre-processor 182 pre-processes the 3D model of the proposed dental structure to generate a modified model by digitally extending the roots of the initial digital model of the proposed dental structure according to the root extension proposals determined by pre-processor 182, thereby more accurately modeling the complete structure of the patient's teeth (206). Step 206 is illustrated in FIG. 2 using a dashed boundary to indicate the optional nature of step 206. For example, the pre-processing functionality provided by step 206 may, in some use case scenarios, be subsumed within the functionality described herein with respect to neural network engine 184.

[0105] In some examples where pre-processor 182 performs step 206, because the tops of the roots (e.g., the regions farthest from the gingival emergence) may be at different heights, pre-processor 182 may detect vertices corresponding to the tops of the roots and then project those vertices along normal vectors, thereby digitally extending the roots. In one example, pre-processor 182 groups the vertices into clusters (e.g., using k-means). Pre-processor 182 may calculate the average normal vector for each cluster of vertices.

[0106] For each cluster of vertices, pre-processor 182 may determine the sum of the residual angular difference between the cluster's average normal vector and the vector associated with each vertex in the cluster. In one example, pre-processor 182 determines which cluster of vertices is the apex of the root based on the sum of the residual angular differences for each cluster. For example, pre-processor 182 may determine that the cluster with the smallest sum of residual angular differences defines the apex of the root.

[0107] The neural network engine 184 can obtain deformation amounts for one or more teeth based on the proposed dental structure represented in the received 3D model (208). For example, the neural network engine 184 can extract respective {translation, rotation, scaling} tuples for one or more teeth represented in the 3D model based on corresponding {position, orientation, size} tuples for the teeth in the current dental structure and dental restoration result information shown in the 3D model of the proposed (post-restoration) dental structure of the patient 102.

[0108] The neural network engine 184 can execute the trained neural network to output (210) placement information for the dental appliance 112. In the example of the dental appliance 112 representing a central clip, the neural network 184 can input deformations of two teeth (e.g., deformations describing the position, orientation, and dimensionality of two adjacent maxillary central incisors) into the two-layer neural network described above. The neural network executed by the neural network 184 can output deformations that position the central clip (which may represent a library part) between the two maxillary central incisors and in a perpendicular orientation relative to the overall arch form of the patient's 102's current, intermediate, or proposed dental structure.

[0109] The neural network engine 184 may use various underlying operation sets to generate deformations (the output of the trained neural network upon execution) of the dental appliances 112 to be applied to the dental restorative treatment of the patient 102. As one example, the trained neural network may process a 3D model of the proposed dental structure of the patient 102 to automatically detect a set of one or more landmarks of the proposed dental structure. In this example, each “landmark” represents a recognizable feature in the 3D model that is useful for determining the position and orientation relative to the surface of one or more teeth. In some examples, the landmarks calculated by the trained neural network include one or more slices of the dental structure, and each slice may include one or more additional landmarks. For example, the trained neural network may divide the 3D mesh of the proposed dental structure into multiple slices and calculate one or more landmarks for each slice, such as the midpoint of each tooth in the slice, the nearest point between two adjacent teeth (e.g., the contact point between two adjacent teeth or the closest point between two adjacent teeth), the convex hull of each tooth in the slice, etc.

[0110] The model assembler 186 generates 212 a 3D model of the dental appliance 112. In various examples of the present disclosure, the model assembler 186 can construct an entire 3D mesh of the dental appliance 112 based on the positioning information indicated by the deformation amounts output by the neural network executed by the neural network engine 184. For example, the model assembler 186 can generate the entire 3D mesh of the dental appliance 112 based on one or more of the positioning characteristics indicated by the {translation, rotation, scale} tuples of the deformations of the present disclosure.

[0111] In some examples, the model assembler 186 can also include shear or shear mapping information of the dental appliance 112 in the calculation of the output deformation if shear information is available for the input data to the neural network being executed and / or if the neural network engine 184 otherwise generates shear information of the dental appliance 112 by executing a trained model of the neural network.

[0112] In one example, the model assembler 186 can extrapolate one or more properties of a {position, orientation, size} tuple of the dental appliance 112 from the deformations output by the neural network (e.g., via integration or other similar techniques). Each 3D mesh {position, orientation, size} tuple generated by the model assembler 186 corresponds to a set of appliance features (e.g., one or both of custom and / or predefined appliance features) of the proposed overall structure of the dental appliance 112. In one example, the model assembler 186 may determine the location of the custom appliance features based on the midpoint of a particular tooth.

[0113] For example, model assembler 186 may align or otherwise position 3D meshes of windows and / or doors (as example features) based on the midpoints of the teeth. In this manner, model assembler 186 may determine the positions of pre-defined appliance features based on the deformation information output by the neural network executed by neural network engine 184. As an example, model assembler 186 may determine the positions of posterior snap clamps based on the positions of the teeth in the patient's 102's current dental structure.

[0114] In some examples, model assembler 186 determines the positions of predefined equipment features based on the positions of custom equipment features. For example, model assembler 186 may align a door hinge, door snap, and / or vent with the centerline of the door. Additionally, model assembler 186 may adjust the orientation, scale, or position of features based on an analysis of the entire model, such as performing a finite element analysis to adjust the active clamping force of a snap clamp. Model assembler 186 may also make adjustments (e.g., for fine tuning) based on expected manufacturing tolerances, such as providing appropriate clearances between features.

[0115] Similarly, the model assembler 186 may make adjustments based on the properties of the materials used to make the physical appliance, such as increasing thickness if a more flexible material is used. In various examples according to this aspect of the disclosure, the model assembler 186 may generate a digital 3D model of the dental appliance 112 to include one or more point clouds, 3D meshes, or other digital representation(s) of the dental appliance 112.

[0116] The computing device 150 outputs (214) the digital 3D model of the dental appliance 112. For example, the computing device 150 can output the digital 3D model of the dental appliance 112 to a computing device 192 at the manufacturing facility 110 by calling the network interface hardware of the communication unit(s) 176 and sending packetized data over the network 114. The manufacturing system 194 manufactures (216) the dental appliance 112. For example, the computing device 192 can control the manufacturing system 194 to manufacture the dental appliance 112 to conform to placement information generated by a trained neural network executed by the neural network engine 184 (e.g., based on the digital 3D model of the dental appliance 112 generated by the model assembler 186). In various examples, the manufacturing system 194 may generate the physical dental appliance 112 by 3D printing, CVD, machining, milling, or any other suitable technique.

[0117] In some examples, the computing system 150 receives feedback regarding the dental appliance 112 from the dental practitioner 106 (218). The optional nature of step 218 is indicated in FIG. 2 by the dashed border. For example, after the dental practitioner 106 receives the physical dental appliance 112 and uses it for dental restorative treatment of the patient 102, the dental practitioner 106 can utilize the computing system 190 to send feedback to the computing device 150. As an example, the computing device 150 may receive data indicating a request to adjust characteristics (e.g., size, positioning characteristics, orientation characteristics, etc.) of a future dental appliance designed according to deformation data output by the neural network for the patient 102, which is a typical patient cohort.

[0118] In some examples, computing system 150 updates 220 physician preference library 168 based on the received physician feedback. The optional nature of step 220 is indicated in FIG. 2 by the dashed border. In some examples, neural network engine 184 can continuously train the neural network using data available from physician preference library 168 (which is continuously updated using incoming physician feedback).

[0119] 3 is a flow diagram illustrating an exemplary use of a neural network for positioning library components of a dental appliance 112, according to an embodiment of the present disclosure. Figure 3 is illustrated with respect to an example using a two-hidden layer neural network to determine the placement of a central clip alignment tab at a specified position and orientation relative to two upper central incisors ("teeth 8 and 9" in the Universal Notation System for Permanent Teeth). The origin of the central clip is positioned approximately at the midpoint of the two upper central incisors, and the vertical (or "Y") axis of the central clip is oriented perpendicular to the dental archform in which the upper central incisors are located.

[0120] The neural network engine 184 may implement back-propagation-based training of a neural network using one 4x4 deformation for each of the maxillary central incisors and one 4x4 deformation that defines the ground truth position and orientation of the central clip after placement. The neural network engine 184 may extract the deformations of the maxillary central incisors from 3D mesh data that describes the current dentition of the patient 192. The neural network engine 184 may obtain the ground truth deformations from a variety of sources, such as manual fabrication by a technician who determines the position and orientation of the central clip using CAD tools, or automatic generation, such as by using the techniques described in WO 2020 / 240351, filed May 20, 2020.

[0121] In the example of FIG. 3, neural network engine 184 converts each of the 4x4 deformations of the upper central incisors into a respective 1x7 quaternion vector. Neural network engine 184 concatenates these two 1x7 quaternion vectors to obtain a single 1x14 feature vector. The 1x14 feature vector corresponds to the data of a single patient (hereafter a single "case"). An "n"x14 matrix can be formed by horizontally concatenating the feature vectors of "n" cases, where "n" denotes a non-negative integer value.

[0122] In this way, the neural network engine 184 can encode data from multiple cases into a single matrix that can be used as a training input for training the two-hidden-layer neural network of Figure 3. The neural network engine 184 can train the neural network using a backpropagation algorithm, in some non-limiting examples of the present disclosure. The layers of the neural network are fully connected, meaning that there is a weighted connection between node i in the first layer and each node j in the next layer.

[0123] The backpropagation training algorithm executed by neural network engine 184 adjusts these weights throughout training (e.g., via multiple training iterations or training passes and fine-tuning thereof), gradually encoding the desired logic into the neural network over multiple training iterations / passes. According to the particular example shown in Figure 3, neural network engine 184 uses a fully connected feed-forward neural network with two hidden layers. Reading Figure 3 from left to right, the first and second hidden layers have dimensions of 1x32 and 1x64, respectively, and the output dimensions are 1x7.

[0124] In other examples consistent with the techniques of this disclosure, neural network engine 184 may utilize other neural network architectures and techniques, such as recurrent neural networks (RNNs), restricted Boltzmann machines (RBMs), long short-term memories (LSTMs), convolutional neural network (CNN) techniques, and various others. In other examples, such as those in which neural network engine 184 uses a 3D mesh as input, neural network engine 184 may use a graph CNN to generate placement information for one or more library components, such as the center clip described above.

[0125] In yet another example, the neural network engine 184 can use an image of a maxillary central incisor (e.g., an image created from a rendering taken of two teeth) as input to a CNN or fully connected neural network to generate placement variations of the library components. The neural network engine 184 passes the 14 input node values ​​along weighted connections to the first hidden layer of Figure 3. The neural network engine 184 passes each node value in the first hidden layer (along with weighting factors for applicable scenarios) to each of the corresponding nodes in the second hidden layer.

[0126] Once the input has completed propagation through the neural network, the neural network 184 converts the resulting 1x7 vector (interpreted as a quaternion) into a 4x4 matrix, which represents the predicted clip alignment matrix. The neural network 184 then calculates a second layer (the "L2 norm") layer normalization of the difference between the alignment matrix and the ground truth deformations input as 4x4 deformations. The resulting difference represents a "loss" value, which the neural network engine 184 may feed back into the neural network to update the weights via backpropagation.

[0127] After multiple iterations of this backpropagation process, the neural network is trained to take as input the deformations of two teeth (i.e., describing the positions and orientations of the two teeth) and output a deformation that positions a library part (e.g., a center clip) between those teeth in an orientation perpendicular to the dental archform. The program flow for training in this particular example is shown in Figure 3. While Figure 3 is described primarily with respect to an example using a neural network with two hidden layers, the neural network engine 184 can train and run more complex computational models in other examples, such as those of deep learning (e.g., generative adversarial networks or "GANs"), according to aspects of the present disclosure.

[0128] 4 is a flow diagram illustrating an example of neural network-based component shape generation according to various aspects of the present disclosure. The neural network engine 184 can train the neural network-based system of FIG. 4 to generate custom component(s), such as the dental appliance 112 or a separate component thereof. In some examples, the neural network engine 184 can use a graph CNN to generate components, such as mold parting surfaces. FIG. 4 is described herein with respect to implementing a CNN as a generative network component of a GAN.

[0129] In the example described herein with respect to Figure 4, the dentition of patient 102 (or the patient in the preceding case) is described by a set of 3D meshes, with each tooth represented by its own individual 3D mesh. When each tooth is transformed into its proper position and orientation to reflect the particular tooth's position in the dental archform, each 3D mesh includes a list of vertices and a list of faces that describe the relationships between the vertices. In other words, each 3D mesh can specify which vertices are part of which faces.

[0130] In these examples, each face is a triangle. The neural network engine 184 inputs the 3D tooth mesh into the graph CNN shown in Figure 4. The graph CNN then generates the component shape as output. The output represents the generated components in the form of another 3D mesh containing their respective vertices and faces. The graph CNN can generate this 3D mesh in one of two ways: 1) by generating a new set of vertices that describe the generated components, or 2) by moving an existing set of vertices.

[0131] In the second technique (which involves moving a set of existing vertices), the generative graph CNN can start with a template or generalized example of a generated component and then manipulate the starting set of vertices to fit the generated component to the dentition of the patient 102 (e.g., current dental structure and / or intermediate dental structure under treatment). The neural network engine 184 then feeds the pairing of the 3D meshes representing the components generated by the graph CNN with the 3D tooth meshes that were originally input to the graph CNN into a differentiable discriminant network.

[0132] The neural network engine 184 also feeds the second pairing, i.e., the pairing of the ground truth library components and the 3D tooth meshes that were originally input to the graph CNN, to the differentiable discriminant network. The differentiable discriminant network calculates the probability that the input pair comes from the second dataset (i.e., the pairing that includes the ground truth generation component). That is, the differentiable discriminant network calculates the probability that each input dataset corresponds to the ground truth dataset that includes the original tooth mesh and the target shape ground truth mesh.

[0133] The differential discriminant network produces gradients that the neural network engine 184 can use as loss functions for the generative network (in this case, implemented as a CNN) shown in FIG. 4. In machine learning terms, a loss function quantifies the degree to which a machine learning model differs from an ideal model, and the loss function is used to guide the training of the machine learning model. The generative network may also use other loss functions, such as normalization of individual layers (e.g., L1 norm and / or L2 norm) and chamfer distance (the sum of positive distances defined for an unsigned distance function). In another example, the neural network engine 184 can input an image of a maxillary central incisor (e.g., an image produced from a rendering of two teeth) into a CNN or fully connected neural network to produce a mesh of ground truth components.

[0134] In various examples, the ground truth generation component may be created manually using CAD tools or automatically using techniques described in International Publication WO 2020 / 240351, filed May 20, 2020. In some examples, the neural network engine 184 may augment the training of the generative network with placement information for the dental appliances 112, such as the deformation output by the neural network of FIG. 2, provided that it is available.

[0135] Although primarily discussed with respect to dental appliances (such as dental appliance 112) used in dental restorative treatment as an example, it will be understood that the neural network-based placement and / or shape generation techniques of the present disclosure can also be used with other types of dental appliances. Non-limiting examples of other dental appliances that the computing device 150 can customize using the techniques of the present disclosure include lingual brackets, comprehensive lingual bracket systems, orthodontic aligners (e.g., transparent or clear aligners), bonding trays, etc. For example, the neural network engine 184 can train a GAN generator, such as the graph CNN generator of the GAN of FIG. 4, using ground truth generation components used with these other types of dental appliances. In these examples, the trained generative graph CNN can produce generated features, such as one or more mold parting surfaces, for any of these other types of dental appliances. Examples of shapes that the neural network engine 184 can use to generate features for lingual brackets are discussed below with reference to FIGS. 7 and 8.

[0136] For example, the computing device 150 can invoke the neural network 184 to generate placement and / or shape information for a lingual bracket system that would otherwise be designed by a technician using custom software. As part of the design of the lingual bracket system, the neural network engine 184 can generate bonding pad specifications for specific teeth. The neural network engine 184 can train the neural network to encompass various steps of the custom software-based generation process, such as delineating the perimeter of specific teeth, determining thicknesses to form shells, and subtracting specific teeth using Boolean operations.

[0137] The neural network engine 184 can train a neural network to select a bracket body from a library (e.g., appliance feature library or model library 166), virtually place the selected bracket body on a pad, and combine the pad and the bracket body mounted thereon using a Boolean addition operation. The neural network can adjust one or more bracket components (e.g., hooks, wings, etc.) to fit the overall bracket to the specific shape of a particular tooth and adjacent gums. When executed by the neural network engine 184, the neural network can generate a design in which the adjusted bracket components are combined with the bracket body to complete the overall bracket digital design, and the overall bracket shape can be exported.

[0138] The neural network engine 184 can encode the overall shape of the bracket in various ways for export, for example, in the form of a stereolithography (STL) file that stores 3D shape information. To train the neural network to generate lingual brackets, the neural network engine 184 can utilize past cases of a patient cohort. If several past cases of various patients are available, the neural engine 184 can train the neural network to implement the automated design of lingual brackets with relatively few (or no) retrainings, thereby saving computational resource overhead and improving accuracy to the idiosyncrasies of the individual dental structure of the patient 102, thereby providing improved data accuracy.

[0139] Examples of custom appliance features that a GAN generative network may generate include information represented by 3D meshes of splines, mold parting surfaces, gingival trim surfaces, shells, facial ribbons, lingual shelves, doors, and windows, among others. In one example, the generative network may generate one or more digital meshes representing the splines for each slice of a dental structure. The GAN generative network may generate the splines for a given slice based on the midpoints of multiple teeth in the slice and / or the nearest points between adjacent teeth in the slice (e.g., contact points between adjacent teeth in the slice or nearest points between adjacent teeth in the slice). In other words, in this example, the generative network compiles a set of points for each slice (e.g., tooth midpoints, contact points between adjacent teeth, closest points between adjacent teeth, or a combination thereof) and generates a feature representing the splines for each digital slice.

[0140] In some examples, the generative network automatically generates a mold parting surface as one exemplary feature incorporated into the overall 3D model of the dental restoration. The neural network engine 184 may execute the generative network to generate the mold parting surface based on multiple midpoints and / or nearest points between adjacent teeth. For example, the generative network can aggregate multiple points of each spline for each slice to generate the mold parting surface. As an example, in an example where the generative network divides the dental structure into four slices and generates a single spline for each slice, the points of each of the four splines may be aggregated to generate the mold parting surface.

[0141] In one scenario, the neural network engine 184 can provide the dentist's 106 selection information from the doctor's selection library 168 to the ground truth repository to be used as training data augmentation. For example, the neural network engine 184 may query the doctor's selection library 168 to determine the dentist's 106 selections. Examples of data stored in the doctor's selection library 168 include the preferred size, position, or orientation of the dentist's 106's predefined appliance features.

[0142] The neural network engine 184 can also train the generative network using data indicative of predefined appliance features by accessing and obtaining the data from one or more libraries (e.g., stored in a data store, database, data lake, file share, cloud repository, or other electronic repository) of 3D meshes representing predefined features for incorporation into the overall 3D model of the dental appliance 112. For example, the neural network engine 184 can receive this data by querying the appliance feature library 164. The appliance feature library 164 stores data defining the 3D meshes of multiple predefined appliance features, such as vents, rear snap clamps, door hinges, door snaps, and incisal alignment features (also called "beaks"), among others.

[0143] In one example, the neural network engine 184 selects one or more predefined prosthetic features from a plurality of predefined prosthetic features stored in the prosthetic feature library 164. For example, the prosthetic feature library 186 may include data defining a plurality of different predefined prosthetic features of a given type. As an example, the prosthetic feature library 164 may include data defining different characteristics (e.g., size, shape, scale, orientation) of a given type of predefined prosthetic feature (e.g., data for hinges of different sizes and / or shapes, etc.). In other words, the prosthetic feature library 164 can determine characteristics of a predefined prosthetic feature and select a feature from the predefined prosthetic library that corresponds to the determined characteristics.

[0144] In some scenarios, the neural network engine 184 selects predefined appliance features from the appliance feature library 164 based on landmarks of corresponding teeth (e.g., door hinges of a particular size), characteristics (e.g., size, type, location) of corresponding teeth (e.g., teeth that will be restored using the appliance features when the dental appliance is applied to the patient), physician selection, or both for training data augmentation.

[0145] In other examples, appliance feature library 164 includes data defining a set of required predefined appliance features. In some such examples, neural network engine 184 can retrieve, for each required predefined feature, a 3D mesh representing the predefined feature for use as additional training data. In such examples, the GAN's generative network can deform the 3D mesh for inclusion in a patient-specific dental appliance. For example, the generative network can rotate or scale (e.g., resize) the 3D mesh of a particular feature based on corresponding dental landmarks, tooth characteristics, and / or practitioner preferences.

[0146] 5 is a flowchart illustrating a process 500 that computing device 150 may implement to generate component shapes using a GAN, according to an embodiment of the present disclosure. Process 500 generally corresponds to the techniques described above with respect to FIG. 4. Process 500 may begin with a training phase 501, in which neural network engine 184 obtains 3D meshes of ground truth dental prosthesis component shapes (502). In various examples, neural network engine 184 may obtain the 3D meshes of ground truth dental prosthesis component shapes from a source that provides manually generated component shapes or a source that provides automatically generated component shapes using techniques described in International Publication WO 2020 / 240351, filed May 20, 2020.

[0147] Also as part of the training phase 501, the neural network engine 184 may train a generative network (e.g., the graph CNN of FIG. 4) on the ground truth component shapes and 3D tooth mesh using a discriminative network (504). For example, the neural network engine 184 may train the generative network by feeding {generated component shape, 3D tooth mesh} pairs and {ground truth component shape, 3D tooth mesh} pairs to the discriminative network. The neural network engine 184 may run the discriminative network to calculate the probability of each pairing and indicate whether the respective pairing is based on the ground truth component shapes.

[0148] While step 504 is shown in FIG. 5 as a single step solely for ease of illustration, it will be appreciated that the neural network engine 184 runs the discriminative network multiple times to train the generative network, continually fine-tuning the training until the generative network produces sufficiently accurate component shapes for the discriminative network to "spoof" the ground truth shape-based pairings.

[0149] Once the neural network engine 184 determines that the generative network is sufficiently trained, the neural network engine 184 may temporarily shelve, or potentially permanently discard, the discriminative network for the execution phase 503 of the process 500. To begin the execution phase 503, the neural network engine 184 may execute the trained generative network to generate component shapes using a 3D tooth mesh of the patient's 102 current dental structure as input (506).

[0150] In one non-limiting example, neural network engine 184 can execute the trained generative network to generate mold parting surfaces for dental appliance 112. Manufacturing system 194 then manufactures 508 dental appliance 112 according to the component shapes generated by the trained generative network. For example, computing device 150 can output the 3D mesh of the component shapes generated by neural network engine 184 to computing device 192 of manufacturing facility 110 by calling network interface hardware of communication unit(s) 176 and sending packetized data over network 114.

[0151] Figure 6 is a rendering illustrating an exemplary central clip placement performed in accordance with the neural network-based placement techniques of the present disclosure described above with respect to Figures 2 and 3. In the two views shown in Figure 6, the central clip is placed between (e.g., centered or substantially at the midpoint of) two maxillary central incisors (teeth 8 and 9 according to the Universal Numbering System for Permanent Teeth) and oriented perpendicular to the arch form of the proposed dental structure of patient 102.

[0152] 7 is a rendering showing an example of a bonding pad (e.g., for a lingual bracket) customized to the shape of a corresponding tooth. As mentioned above, the GAN-based techniques described with respect to FIGS. 4 and 5 can be used to generate such a bonding pad shape.

[0153] Figure 8 is a rendering showing an example set of components that make up a lingual bracket. The techniques described above with respect to Figures 1-5 can be used to assemble brackets, such as the complete bracket shown in Figure 8, and / or to generate placement information for brackets on the teeth of patient 102.

[0154] 9 is a flow diagram illustrating another example of neural network-based component shape generation according to an embodiment of the present disclosure. The neural network engine 184 can train and execute the GAN generative network shown in FIG. 9 to improve or fine-tune a previously generated dental appliance model to form an updated dental appliance model or an updated model of its components. The neural network engine 184 can use landmark information to update the shape of an automatically generated dental appliance model (e.g., using the techniques described in International Publication WO2020 / 240351, filed May 20, 2020) or a computer-aided design (CAD) tool to update the shape of a manually generated dental appliance model to form an updated model (e.g., an updated 3D) of the present disclosure.

[0155] GAN-generative based refinement of previously generated models results in reduced time and often improved accuracy and data precision regarding the shape modifications required to make the dental appliance model viable (the updated model represents a viable dental appliance component for use in a dental restorative procedure). Once the generative network is trained to sufficiently spoof the discriminative network and / or pass visual inspection, it is configured to gradually modify the component design to make the updated model of the dental appliance shape match a design that can be used during the dental restorative treatment of the patient 102.

[0156] 9 , in contrast to the technique described with respect to FIG. 4 (e.g., in which the component shape is designed entirely by a GAN generative network), combines computational results from a landmark-based automated tool (e.g., the automated tool described in International Publication WO 2020 / 240351, filed May 20, 2020) or manually generated shapes with neural network-based fine-tuning to complete the design, with any last-mile refinements (forming updated component models) that may be beneficial to the dental restorative treatment of patient 102. The GAN of FIG. 9 provides fast convergence times, enabling computing device 150 to generate updated component models, providing the benefits of both initial shape design using landmark-based techniques and neural network-based shape refinement in a computationally fast manner.

[0157] The GAN of FIG. 9 enables training of a generative network even when there are only a limited number of previously generated models to use as examples for training. In this way, the GAN of FIG. 9 leverages design elements of the initial design shape when training data is limited, while simultaneously implementing the benefits of neural network-based design for last-mile fine-tuning of the initial design. Compared to the GAN of FIG. 4, the GAN of FIG. 9 provides the generative network with additional input (in both the training and execution phases) using the initial orthosis shape, which may need to be fine-tuned to reach its final form (in the form of an updated model or updated component 3D meshes) for fabrication by the manufacturing system 194 (e.g., via 3D printing).

[0158] 10-12 are directed to aspects of the present disclosure describing a system configured to display a proposed dental restoration to a patient 102 via an automated design process using generative modeling. According to these aspects of the present disclosure, a neural network engine 184 utilizes data collected from a dental practitioner 106 and / or other trained clinicians / technologists to train a neural network configured to generatively model the proposed dental structure for the patient 102. For example, the neural network engine 184 may train the neural network in a data-driven manner to learn the attributes of acceptable dental restorations. Examples of the present disclosure are described with respect to generating two-dimensional (2D) images of a unique proposed restored dental structure for a single patient (patient 102 in these examples).

[0159] The neural network-based generative display techniques of the present disclosure are described below with respect to dental restorations as a non-limiting example of generating 2D images of proposed dental structures (after restoration). However, the neural network-based generative display techniques of the present disclosure can also be applied with respect to other areas, such as assisting with 3D printing of ceramic and / or composite crowns, etc. It will be appreciated that the goal of the various dental restoration treatments discussed herein is to provide the patient 102 with a low-cost, minimally invasive composite restoration of a damaged or unesthetic tooth, or other suboptimal dental restoration associated with the patient's 102 current dental structure.

[0160] A patient 102 (or any other patient) interested in a dental restoration can have their current dental structure scanned at the clinic 104. The disclosed neural network-based generative modeling technique provides a fast-to-process and data-accurate 2D image view of the proposed dental structure after restoration, customized for a given patient (patient 102 in this particular example). The disclosed neural network-based generative modeling technique significantly reduces the lead time and cost for dental restoration planning from existing solutions.

[0161] To improve data accuracy for generative modeling of post-restoration 2D imaging of proposed dental structures for patient 102, neural network engine 184 can incorporate dental restoration styles (e.g., young, aged, natural, oval, etc.) into the training data if style information is available for past cases. In these and / or other examples, neural network engine 184 can incorporate one or more of accepted "golden ratio" guidelines for tooth size, accepted "ideal" tooth shapes, patient preferences, practitioner preferences, etc. into the training data used to train the neural network. If different styles are available in the training dataset, patient 102 may be able to view different restoration options generated by the different style algorithms. In other words, neural network engine 184 can generate different style options for proposed post-restoration dental structures for patient 102 based on different stylized results in past cases.

[0162] By training the neural network with these dental restoration-related training data (typically over multiple training iterations for fine-tuning), the neural network engine 184 improves the accuracy of the data for generatively modeling the proposed dental structure of the patient 102 by reducing the iterations required to correct or fine-tune multiple attempts to plan a single dental restoration treatment for a given patient, reduces runtime computational resource consumption (by running an accurately trained neural network), and reduces the overall process time for generating a dental restoration treatment plan.

[0163] Computing devices 150 and 190 also provide various user experience-related improvements due to computing device 150's implementation of the neural network-based generative modeling techniques of the present disclosure. For example, dental practitioner 106 may present 2D images of proposed restored dental structures to patient 102 by scanning the patient's 102's current dental structure and then generating the 2D images relatively quickly (and potentially during the same patient encounter). In some examples, dental practitioner 106 may synthesize different post-restoration results (e.g., using different styles or other selection-related factors) to help patient 102 view different options and select a dental restoration plan.

[0164] In some examples, dental practitioner 106 can provide "pre-approved" goals for the generation of 3D restoration files that can be used in the design and / or manufacturing process of dental appliance 112. By providing pre-approved planning information (which neural network engine 184 can obtain from practitioner-selected library 168 or other sources), neural network engine 184 can train a neural network to generate a custom dental restoration model with a reduced amount of input from the practitioner, compressing the manufacturing process for the custom product.

[0165] The patient 102 can visualize possible post-restoration outcomes for their own dental structure, rather than past cases for other patients. Thus, the neural network engine 184 leverages training data to provide personalization to improve the user experience in these instances as well. The generative modeling techniques of the present disclosure may be applicable to areas other than dental restorations where patients are interested in unique or customized solutions, such as those related to protective masks, bandages, etc.

[0166] According to some examples of the present disclosure, the neural network engine 184 uses a GAN to generate 2D images of proposed dental structures of the patient 102 for the post-restorative treatment phase. As described above, a GAN utilizes a pairing of differentiable functions, often deep neural networks, with the goal of learning to generate an unknown data distribution. A first network, known as a generative network, produces data samples given some input (e.g., random noise, conditional class labels, etc.). A second network, known as a discriminative network, attempts to classify the data generated by the generator from actual data points coming from the true data distribution.

[0167] As part of training, the generative network continuously attempts to fool (or deceive or trick) the classifier into classifying newly generated data as "real." The more frequently the generated data is used to successfully fool the classifier, the more realistic the training output of the generative network becomes. In some examples of the generative 2D modeling techniques of the present disclosure, the neural network engine 184 uses a conditional GAN ​​(cGAN), where the generative network is conditioned on 2D rendered images of a 2D scan of the patient's 102's current dental structure.

[0168] In some non-limiting examples, the generative network, which is a CNN, takes as input a rendered 2D image of the patient's 102 current (pre-restoration) dental structure and generates a 2D image of what the proposed (post-restoration) dental structure of the patient 102 would look like based on the current state of the generative network's adversarial training. The generative network can also accept additional information as input data (depending on availability and / or relevance), such as which teeth should be restored, the restoration style (e.g., young, aged, natural, oval, etc.), etc.

[0169] In some examples, the discriminator network, which may be a CNN, receives pairs of 2D images as input. The first pair includes a rendered 2D image of a patient's pre-restoration dental structure and a rendered 2D image of a true restoration for the same patient performed by a clinician (classified as an "actual" or "ground truth" pairing). The second pair of images includes the pre-restoration rendered 2D image and the generative network-generated restoration. The generative network and the discriminator network are trained simultaneously in an iterative, alternating fashion, improving each other to reach a shared goal of an accurately trained generative network.

[0170] In some embodiments, the neural network engine 184 implements the generative network of the 2D inpainting image aspect of this disclosure as an encoder-decoder CNN. In these examples, the generative network reduces the dimensionality of the input image and then expands the dimensionality back to the original dimensionality (e.g., via a sequence of downsampling and upsampling, or otherwise). The generative network in these examples may also be referred to as a "U-Net." As used herein, "U-Net" refers to a type of encoder-decoder architecture in which feature maps from the encoder are concatenated to respective feature maps in the decoder.

[0171] In traditional GANs, the discriminator network receives either real images (coming from an input dataset of images) or synthesized images (created by a generator). The output of the discriminator network is a probability in the range [0, 1] that represents the likelihood that the input image is a real image (coming from the dataset). In some implementations of the 2D inpainting image aspect of the present disclosure, the discriminator network is a "patchGAN" discriminator network.

[0172] While a typical discriminant network outputs a single value representing the perceived realism of the input, the patchGAN discriminant network outputs an [nxn] matrix, with each element representing the perceived realism of the corresponding patch of the input. The perceived realism as represented by each element of the [nxn] output matrix represents the probability that the corresponding patch of the input image is part of the actual or ground truth image. The discriminant network is internally implemented as a CNN.

[0173] FIG. 10 is a conceptual diagram illustrating the symbiotic training process of a cGAN's generative and discriminative networks configured to render 2D images of proposed dental structures for a patient 102, according to an embodiment of the present disclosure. The embodiment of FIG. 10 also illustrates how various data are processed by the generative and discriminative networks. In FIG. 10, "G" denotes the cGAN's generative network, and "D" denotes the cGAN's discriminative network. The pre-restoration 2D image (which is the input to G and is one half of the image pairing fed from G to D) is denoted by "x." "G(x)" denotes the proposed restoration 2D image generated by G given x as the pre-restoration input. The 2D rendered image of the actual dental restoration (or a "true" or "ground truth" image of the restored dental structure) is denoted by "y."

[0174] The particular use case scenario shown in FIG. 10 is associated with an unsuccessful iteration during the multi-iteration training process of G. As shown in FIG. 10, D outputs a verdict that the combination of G(x) and x is “false.” In contrast, and as intended with respect to adversarial cGAN training, D outputs a “true” verdict when evaluating the input combination of x and y. In some examples, if D is a well-trained and improved network, after G is more accurately adversarially trained over future iterations of cGAN training, G may generate an instance of G(x) that, when fed to D with x, successfully fools D into outputting a “true” verdict.

[0175] In these examples, once G reaches this training level, the neural network 184 can begin running G to generate proposed 2D images of the patient's 102's restored dental structure from an input of x. In some examples, G and D are trained collaboratively, so that both networks may be untrained for a similar period of time. In these cases, both G and D may be trained until G passes qualitative inspection, such as by passing visual inspection by the dentist 106 or another clinician. Following the format used above, "x" represents the 2D pre-restoration image, "y" is the ground truth 2D post-restoration image, and G(x) is the image generated by G given the pre-restoration image input. The total loss term used in some examples is a combination of the L1 loss and the GAN loss, given by Equation (1) below:

number

[0176] The L1 loss is the absolute value of the difference between y and G(x), and the total loss applies to G but not to D. The calculation of the L1 loss is given by the following equation (2), where λ L1 is 10 in this particular example, while λ L1 It will be understood that may have other values ​​in other examples consistent with this disclosure.

number

[0177] By leveraging a communications connection to computing device 150 via network 114, computing device 190 may provide a chairside application that enables dental practitioner 106 to show 2D renderings of one or more proposed restoration plans to patient 106, often during the same visit in which a scan of the patient's current dental structure is taken (and sometimes shortly or immediately after the scan is taken). Rather than displaying generic models intended for other patients' past cases or hypothetical patients, computing device 190 may use cGAN executed by neural network engine 184 to output custom renderings of proposed dental structures for one or more post-restoration scenarios that are specifically applicable to patient 102's current dental structure and treatment plan(s).

[0178] That is, the neural network engine 184 can implement the generative modeling techniques of the present disclosure such that the dental practitioner 106 leverages cloud computing interaction to render 2D images of proposed dental structures for one or more dental restoration plans that are specifically applicable to the current dental structure of the patient 102. From the perspective of the clinic 104, given a scan of the current dental structure of the patient 102, the computing device 190 quickly (or near-instantly) processes the scan and leverages cloud computing power to render 2D images of one or more post-treatment dental structure images specific to the dentition of the patient 102 and treatment options available to the patient 102.

[0179] In this way, neural network engine 184 may implement the generative modeling techniques of the present disclosure entirely in the image domain, without requiring the potentially time-consuming generation of a 3D mesh. Upon approval (e.g., by patient 102 and / or dental practitioner 106), computing device 190 may communicate the generated 2D images over network 114 to computing device 192, enabling manufacturing system 194 to generate a 3D mesh of dental appliance 112 or directly manufacture dental appliance 112.

[0180] In some examples, the neural network engine 184 can generate or regenerate 2D images of the proposed dental restoration to incorporate patient-specified modifications, such as restoration style selection or other parameters. In one such example, the neural network engine 184 can implement a feedback loop within the cGAN to accommodate patient-provided or physician-provided modifications regarding restoration style, tooth shaping, etc.

[0181] In one example, the cGAN trained generative network may enable a technician to create a 3D mesh from the 2D image output by the trained generative network. In another example, the 3D mesh may be automatically generated from the 2D image of the proposed dental structure output by the trained generative network. In one or more of these examples, the 3D mesh may be used as input to the system described above with respect to FIGS. 3 and 4. In some examples, the dentist 106 or other clinician at the clinic 104 may use image capture hardware (e.g., a still camera or a video camera) to acquire a photograph of the patient 102's current dental structure. In these examples, the computing device 190 may use the captured photograph to generate a rendering of the patient 102's current dental structure.

[0182] Thus, according to various examples of the present disclosure, computing devices 190 and 150 may acquire a 2D image (dental scan or photograph) of a 3D object (in this case, the dentition of patient 102) and use the 2D image to generate another 2D image of a proposed dental structure (or portion thereof) for proposed dental restorative treatment for patient 102. In this manner, computing devices 190 and 150 may use the neural network training mechanisms of the present disclosure to enable dental restoration modeling in a computationally lightweight and fast manner while maintaining data accuracy for the dental restoration modeling.

[0183] In some examples, the neural network engine 184 may implement a trained version of the generative network G as the input generation system for the neural networks shown in Figures 3 and 4. For example, the neural network engine 184 may augment the deformation matrix input to the neural network of Figure 3 and / or the tooth mesh input to the generative network of Figure 4 with a 2D image of the proposed dental structure of the patient 102. In these examples, the neural network engine 184 may utilize the output of the trained version of the generative network G to more holistically train the neural network of Figure 3 and / or the generative graph CNN of Figure 4 to consider the effect of a greater proportion of the arch form of the dentition of the patient 102 for which configuration and / or shape information is being generated.

[0184] 11A shows the inputs and outputs of a cGAN-trained generative network configured to generate a 2D image of a proposed dental structure using a 2D rendering of the patient's 102 current dental structure. The current dental structure image 1102 shows a 2D rendering of the patient's 102 current dental structure. Once trained with the cGAN (e.g., by successfully fooling the discriminative network at least a threshold number of times), the generative network (“G”) of FIG. 10 uses the current dental structure image 1102 to generate a proposed dental structure image 1104.

[0185] Current dental structure image 1102 is a 2D rendering of the restored anterior dentition of patient 102. Proposed dental structure image 1104 is a 2D rendering of a projection of the final result of one proposed dental restoration treatment plan for patient 102. Thus, FIG. 11 illustrates an example of one use case scenario in which cGAN-trained iterations of generative network G implement the generative modeling techniques of the present disclosure.

[0186] FIG. 11B shows a comparison between a current dental structure image 1102, a proposed dental structure image 1104, and a ground truth restoration image 1106.

[0187] 12 illustrates a menu that the computing device 190 may display as part of a graphical user interface (GUI) that includes the current dental structure image 1102 and / or the proposed dental structure image 1104. The data menu 1202 presents the dentist 106 or another clinician with various options for manipulating the content of the generative modeling. In the example of FIG. 12, the data menu 1202 presents selectable test case options for building a dental restoration plan. The data menu 1202 also includes a tooth option that allows the dental practitioner 106 to select a specific tooth from the current dental structure image 1102 to be modeled for reconstruction.

[0188] The visual options menu 1204 allows the dental practitioner 106 to adjust various viewing parameters for the display of the current dental structure image 1102 and / or the proposed dental structure image 1104. The dental practitioner 106 can adjust various viewing parameters via the visual options menu 1204 to allow the patient 102 to better see the details of the proposed dental structure image 1204.

[0189] In this manner, the dental practitioner 106 or other clinician can operate the computing device 190 at the clinic 104 to provide cloud interaction over the network 114, thereby leveraging the neural network-based generative modeling capabilities provided by the computing device 150. By operating the data menu 1202, the dental practitioner 106 can provide restoration modeling parameters for use by the neural network engine 184 in generating the proposed dental structure image 1204. By operating the visual options menu 1204, the dental practitioner 106 uses the computing device 190 as an on-site display to customize the display parameters of the proposed dental structure image 1204 to suit the viewing needs and preferences of the patient 102.

[0190] 13A and 13B are conceptual diagrams illustrating an example mold parting surface according to various aspects of the present disclosure. The neural network engine 184 can generate the mold parting surface 1302 based on landmarks, such as the midpoints of each tooth in each slice and points between adjacent teeth (e.g., contact points between adjacent teeth and / or closest points between adjacent teeth). In some examples, the neural network engine can generate a 3D mesh of the mold parting surface 1302 as part of the neural network-based geometry generation techniques of the present disclosure. Additional details on how the mold parting surface 1302 can be used in connection with appliance manufacturing are described in International Publication WO 2020 / 240351, filed May 20, 2020.

[0191] 14 is a conceptual diagram illustrating an exemplary gingival trimming surface according to various aspects of the present disclosure. The gingival trimming surface 1402 may include a 3D mesh that trims the surrounding shell between the gingiva and the teeth in the dental structure shown.

[0192] 15 is a conceptual diagram illustrating an exemplary facial ribbon according to various aspects of the present disclosure. The facial ribbon 1502 is a reinforcing rib with a nominal thickness that is facially offset from the shell. In some cases, the facial ribbon follows both the archform and the gingival margin. In one example, the lower portion of the facial ribbon is not gingivally closer to the gingival trim surface.

[0193] 16 is a conceptual diagram illustrating an exemplary lingual shelf 1602, according to various aspects of the present disclosure. The lingual shelf 1602 is a reinforcing rib with a nominal thickness on the lingual side of the molded appliance, inset on the lingual side according to the dental arch form.

[0194] 17 is a conceptual diagram illustrating exemplary doors and windows according to various embodiments of the present disclosure. Windows 1704A-1704H (collectively, windows 1704) include apertures that provide access to the tooth surface so that a dental composite can be placed on the tooth. The doors include a structure that covers the window. The shape of the window may be defined as the nominal inset from the periphery of the tooth when viewed facially. In some cases, the shape of the door corresponds to the shape of the window. The doors may be inset to provide a gap between the door and window.

[0195] FIG. 18 is a conceptual diagram illustrating exemplary posterior snap clamps according to various aspects of the present disclosure. A neural network 184 can determine one or more characteristics (e.g., placement-related or shape-related characteristics) of the posterior snap clamps 1802A and 1802B (collectively, “posterior snap clamps 1802”). The posterior snap clamps 1802 may be configured to couple a facial portion of the dental appliance 112 with a lingual portion of the dental appliance 112. Exemplary characteristics include one or more of the size, shape, position, or orientation of the posterior snap clamps 1802. The position information of the posterior snap clamps 1802 may be at both ends of the dental arch form along the dental arch form (e.g., a first snap clamp at one end and a second snap clamp at the other end). In some examples, the female portion of the posterior snap clamp 1802 may be positioned on the lingual side of the parting plane, and the male portion of the posterior snap clamp 1802 may be positioned on the facial side.

[0196] FIG. 19 is a conceptual diagram illustrating an example door hinge according to various aspects of the present disclosure. The neural network engine 184 can determine one or more characteristics of the door hinges 1902A-1902F (collectively, door hinges 1902) as part of generating placement and / or shape information for the dental appliance 112 according to various aspects of the present disclosure. The door hinges 1902 may be configured to pivotally couple a door to the dental appliance 112. Example characteristics include one or more of the size, shape, position, or orientation of each door hinge(s) 1902. The neural network executed by the neural network engine 184 can position the door hinges 1902 based on the position of another predefined appliance feature in several non-limiting use case scenarios. For example, the neural network may position each door hinge 1902 at the centerline of its corresponding door. In one use case scenario, the female portion of each door hinge 1902 may be positioned to secure to a facial portion of the dental appliance 112 (e.g., toward the incisal edge of each tooth), and the male portion of the same door hinge 1902 may be positioned to secure to an exterior surface of the door.

[0197] 20A and 20B are conceptual diagrams illustrating exemplary door snaps according to various aspects of the present disclosure. The neural network engine 184 can determine one or more characteristics of the door snaps 2002A-2002F (collectively, "door snaps 2002"), such as placement characteristics and / or shape characteristics. Exemplary characteristics include one or more of the size, shape, position, or orientation of the door snaps 2002. In some examples, the neural network executed by the neural network engine 184 can determine the position of the door snaps 2002 based on the position of another predefined appliance feature. In one example, the neural network can generate a placement profile that positions each door snap 2002 at the midline of the corresponding door. In one example, the position of the female portion of a particular door snap 2002 may be anchored to the exterior surface of the door and extend downward toward the gums. In another example, the male portion of a particular door snap 2002 may be anchored to the gum side of a facial ribbon.

[0198] 21 is a conceptual diagram illustrating an exemplary incisal ridge, according to various aspects of the present disclosure. The incisal ridge 2102 provides reinforcement at the incisal edge.

[0199] 22 is a conceptual diagram illustrating an exemplary center clip according to various aspects of the present disclosure. The center clip 2202 aligns the facial and lingual portions of the dental appliance with one another.

[0200] 23 is a conceptual diagram illustrating exemplary door vents according to various aspects of the present disclosure. Door vents 2302A and 2302B (collectively, door vents 2302) transport excess dental composite from dental appliances.

[0201] 24 is a conceptual diagram illustrating an exemplary door, according to various aspects of the present disclosure. In the example of FIG. 24, the dental appliance includes a door 2402, a door hinge 2404, and a door snap 2406.

[0202] 25 is a conceptual diagram illustrating an exemplary dental space matrix according to various aspects of the present disclosure. The dental space matrix 2502 includes a handle 2504, a body 2506, and a wrapping portion 2508. The wrapping portion 2508 is configured to fit into the interproximal area between two adjacent teeth.

[0203] FIG. 26 is a conceptual diagram illustrating an exemplary fabrication case frame and an exemplary dental appliance according to various aspects of the present disclosure. The fabrication case frame 2602 is configured to support one or more components of the dental appliance. For example, the fabrication case frame 2602 may removably couple a lingual portion of the dental appliance 2604, a facial portion of the dental appliance 2606, and a septum matrix 2608 to one another via a case frame sparring 2610. In the example of FIG. 26, the case frame sparring 2610 connects or couples the components of the dental appliance 2604, 2606, and 2608 to the fabrication case frame 2602.

[0204] 27 is a conceptual diagram illustrating an example dental appliance including custom labels, according to various aspects of the present disclosure. Custom labels 2702-2708 may be printed on various parts of the dental appliance and include data identifying each part of the dental appliance (e.g., serial number, part number, etc.).

[0205] Various embodiments have been described. These and other embodiments are within the scope of the following claims. In addition to the above embodiment, the following aspects are added. (Appendix 1) an input interface configured to receive deformation information associated with the dental restoration patient's current dental structure; Neural network engine, 1. A computing device comprising: The neural network engine providing the deformation information associated with the current dental structure of the dental restoration patient as input to a neural network trained with deformation information indicative of placement of dental appliance components on one or more teeth of the corresponding dental structure, the dental appliances having been used in dental restoration treatment for the one or more teeth; and executing the neural network using the input to generate placement information for the dental prosthetic component relative to the current dental structure of the dental restoration patient. Computing devices. (Appendix 2) the deformation information indicative of the placement of the dental prosthetic component relative to the one or more teeth of the corresponding dental structure includes a single matrix deformation amount corresponding to each of the one or more teeth of the corresponding dental structure; the training data further includes ground truth deformations indicative of a finished configuration of the dental appliance components; and wherein the neural network engine is configured to output a matrix deformation amount of the dental appliance components related to dental restorative treatment of the current dental structure to generate the placement information of the dental appliance components relative to the current dental structure of the dental restoration patient. 2. The computing device of claim 1. (Appendix 3) 3. The computing device of claim 2, wherein each single matrix deformation corresponding to each of the one or more teeth is a respective 4x4 deformation corresponding to each of the one or more teeth, and the matrix deformation output of the dental appliance component is a 4x4 deformation output of the dental appliance component. (Appendix 4) 4. The computing device of claim 3, wherein the 4x4 deformation output of the dental prosthesis component for the dental restoration treatment includes a {translation, rotation, scale} tuple indicating change information of a {position, orientation, size} tuple describing the placement information of the dental prosthesis component relative to the current dental structure of the dental restoration patient. (Appendix 5) the one or more teeth include two maxillary central incisors, and the neural network engine: converting each 4x4 deformation of each of the two maxillary central incisors into a respective 1x7 quaternion vector; concatenating the 1x7 quaternion vectors corresponding to the two maxillary central incisors to form a 1x14 feature vector; further configured to train the neural network by and wherein the neural network engine is configured to provide the 1×14 feature vector as the input to the neural network to provide the transformation information as the input to the neural network. 5. The computing device of claim 4. (Appendix 6) 6. The computing device of claim 5, wherein the dental appliance component is a central clip. (Appendix 7) 7. The computing device of any one of Claims 5 or 6, wherein the neural network engine is configured to use a backpropagation algorithm on the respective 4x4 transformations corresponding to the one or more teeth of the corresponding dental structure to train the neural network. (Appendix 8) receiving deformation information associated with a current dental structure of a dental restoration patient; providing the deformation information associated with the current dental structure of the dental restoration patient as input to a neural network trained with deformation information indicative of placement of dental appliance components on one or more teeth of the corresponding dental structure, the dental appliances being used in dental restoration treatment for the one or more teeth; running the neural network using the input to generate placement information for the dental prosthetic component relative to the current dental structure of the dental restoration patient; A method comprising: (Appendix 9) the deformation information indicative of the placement of the dental prosthetic component relative to the one or more teeth of the corresponding dental structure includes a single matrix deformation amount corresponding to each of the one or more teeth of the corresponding dental structure; the training data further includes ground truth deformations indicative of a finished configuration of the dental appliance components; generating the positioning information of the dental prosthetic components relative to the current dental structure of the dental restoration patient includes outputting a matrix deformation of the dental prosthetic components related to dental restoration treatment of the current dental structure; The method described in Appendix 8. (Appendix 10) 10. The method of claim 9, wherein each single matrix deformation corresponding to each of the one or more teeth is a respective 4x4 deformation corresponding to each of the one or more teeth, and the matrix deformation output of the dental appliance component is a 4x4 deformation output of the dental appliance component. (Appendix 11) 11. The method of claim 10, wherein the 4x4 deformation output of the dental prosthesis component for the dental restoration treatment includes a {translation, rotation, scale} tuple indicating change information of a {position, orientation, size} tuple describing the placement information of the dental prosthesis component relative to the current dental structure of the dental restoration patient. (Appendix 12) wherein the one or more teeth include two maxillary central incisors, and the method comprises: converting each 4×4 deformation of each of the two maxillary central incisors into a respective 1×7 quaternion vector; concatenating the 1x7 quaternion vectors corresponding to the two maxillary central incisors to form a 1x14 feature vector; training the neural network by providing the transformation information as the input to the neural network includes providing the 1×14 feature vector as the input to the neural network. The method described in Appendix 11. (Appendix 12) 13. The method of claim 12, wherein training the neural network includes using a backpropagation algorithm on the respective 4x4 transformations corresponding to the one or more teeth of the corresponding dental structure. (Appendix 14) 14. The method of any one of claims 8 to 13, wherein the dental prosthetic component is a central clip, and the placement information of the central clip identifies a midpoint between two teeth of the current dental structure of the dental restoration patient. (Appendix 15) 15. The method of any one of claims 8 to 14, wherein the dental prosthetic component is a posterior snap clamp, and the placement information of the posterior snap clamp identifies a midpoint between the two teeth of the dental restoration patient's current dental structure. (Appendix 16) 16. The method of claim 15, wherein the two teeth of the current dental structure of the dental restoration patient include a molar and a canine of the current dental structure of the dental restoration patient. (Appendix 17) 17. The method of claim 16, wherein the two teeth of the current dental structure of the dental restoration patient include a molar and a premolar of the current dental structure of the dental restoration patient. (Appendix 18) means for receiving deformation information associated with the dental restoration patient's current dental structure; means for providing the deformation information associated with the current dental structure of the dental restoration patient as input to a neural network trained with deformation information indicative of placement of dental appliance components on one or more teeth of a corresponding dental structure, the dental appliances being used in dental restoration treatment for the one or more teeth; means for executing the neural network using the input to generate placement information for the dental prosthetic component relative to the current dental structure of the dental restoration patient; An apparatus comprising: (Appendix 19) A non-transitory computer-readable storage medium encoded with instructions that, when executed, cause one or more processors of a computing system to: receiving deformation information associated with a current dental structure of a dental restoration patient; providing the deformation information associated with the current dental structure of the dental restoration patient as input to a neural network trained with deformation information indicative of placement of dental appliance components on one or more teeth of the corresponding dental structure, the dental appliances having been used in dental restoration treatment for the one or more teeth; running the neural network using the input to generate placement information for the dental prosthetic component relative to the current dental structure of the dental restoration patient; A non-transitory computer-readable storage medium. (Appendix 20) 1. A computing device comprising: an input interface configured to receive one or more three-dimensional (3D) tooth meshes associated with a current dental structure of a dental restoration patient; and a neural network engine; The neural network engine providing the one or more 3D tooth meshes received by the input interface as input to a neural network trained with training data including ground truth dental appliance component shapes and corresponding 3D tooth meshes of dental restoration cases; running the neural network using the provided inputs to generate a custom shape of the dental prosthetic component for the current dental structure of the dental restoration patient; A computing device configured to: (Appendix 21) 21. The computing device of claim 20, wherein the neural network is a generative adversarial network (GAN) consisting of a generative network and a discriminative network, and wherein the neural network engine is configured to provide the one or more 3D tooth meshes received by the input interface as input to the generative network of the GAN to execute the neural network to create the custom shape of the dental prosthetic component for the current dental structure of the dental restoration patient. (Appendix 22) The neural network engine Executing the generative network of the GAN to generate a generated shape of the dental appliance component using the one or more 3D tooth meshes; providing the generated shape of the dental prosthetic component and the one or more 3D tooth meshes as inputs to the classification network of the GAN; and running the discriminative network to output a probability that the generated shape represents a ground truth shape. (Appendix 23) 22. The computing device of any one of claims 20 or 21, wherein the neural network is further trained with geometry information of the ground truth dental appliance components as part of the training data. (Appendix 24) The dental appliance component comprises: Mold parting surface, Gum trim, door, window, Facial ribbon, Incisal ridge, Lingual shelf, Interdental matrix, Case frame, or Parts labels, 24. The computing device of any one of claims 19 to 23, comprising one or more of: (Appendix 25) receiving one or more three-dimensional (3D) tooth meshes associated with the dental restoration patient's current dental structure; providing the one or more 3D tooth meshes associated with the current dental structure of the dental restoration patient as input to a neural network trained with training data including ground truth dental appliance component shapes and corresponding 3D tooth meshes of dental restoration cases; running the neural network using the provided inputs to generate a custom shape of the dental prosthetic component for the current dental structure of the dental restoration patient; A method comprising: (Appendix 26) 26. The method of claim 25, wherein the neural network is a generative adversarial network (GAN) consisting of a generative network and a discriminative network, and executing the neural network to create the custom shape of the dental prosthetic component for the current dental structure of the dental restoration patient includes providing the one or more 3D tooth meshes received by the input interface as input to the generative network of the GAN. (Appendix 27) Executing the generative network of the GAN to generate a generated shape of the dental appliance component using the one or more 3D tooth meshes; providing the generated shape of the dental prosthetic component and the one or more 3D tooth meshes as inputs to the classification network of the GAN; running the discriminative network to output a probability that the generated shape represents a ground truth shape; 27. The method of claim 26, further comprising training the GAN by: (Appendix 28) 28. The method of any one of claims 25 to 27, wherein the neural network is further trained using geometry information of the ground truth dental appliance components as part of the training data. (Appendix 29) The dental appliance component comprises: Mold parting surface, Gum trim, door, window, Facial ribbon, Incisal ridge, Lingual shelf, Interdental matrix, Case frame, or Parts labels, 29. The method of any one of appendices 25 to 28, comprising one or more of the following: (Appendix 30) means for receiving one or more three-dimensional (3D) tooth meshes associated with the dental restoration patient's current dental structure; means for providing the one or more 3D tooth meshes associated with the current dental structure of the dental restoration patient as input to a neural network trained with training data including ground truth dental appliance component shapes and corresponding 3D tooth meshes of dental restoration cases; means for executing the neural network using the provided input to generate a custom shape of the dental prosthetic component for the current dental structure of the dental restoration patient; An apparatus comprising: (Appendix 31) A non-transitory computer-readable storage medium encoded with instructions that, when executed, cause one or more processors of a computing system to: receiving one or more three-dimensional (3D) tooth meshes associated with a current dental structure of a dental restoration patient; providing the one or more 3D tooth meshes associated with the current dental structure of the dental restoration patient as input to a neural network trained with training data including ground truth dental appliance component shapes and corresponding 3D tooth meshes of dental restoration cases; running the neural network using the provided inputs to generate a custom shape of the dental prosthetic component for the current dental structure of the dental restoration patient; A non-transitory computer-readable storage medium. (Appendix 32) 1. A computing device comprising: an input interface configured to receive a two-dimensional (2D) image of a current dental structure of a dental restoration patient; and a neural network engine; The neural network engine providing the 2D image of the current dental structure of the dental restoration patient as input to a neural network trained with training data including 2D images of pre-restoration dental structures and corresponding 2D images of post-restoration dental structures of previously performed dental restoration cases; executing the neural network using the input to generate a 2D image of a proposed dental structure for the dental restoration patient, the proposed dental structure being associated with a post-restoration outcome of the dental restoration plan for the dental restoration patient; A computing device configured to: (Appendix 33) 33. The computing device of claim 32, wherein the neural network is a conditional adversarial network (cGAN) consisting of a generative network and a discriminative network, and wherein the neural network engine is configured to provide the 2D image of the current dental structure of the dental restoration patient as the input to the neural network to provide the 2D image of the current dental structure of the dental restoration patient as the input to the generative network of the cGAN. (Appendix 34) The neural network engine executing the generative network of the cGAN to generate a respective proposed post-restoration 2D image corresponding to each of the pre-restoration dental structures of the training data; providing each of the proposed restored 2D images and the pre-restoration dental structure corresponding to each of the proposed restored 2D images as inputs to a discriminative network of the cGAN; running the discriminative network to output a respective probability of each proposed restored 2D image representing a ground truth image; 34. The computing device of claim 33, configured to train the cGAN by (Appendix 35) 35. The computing device of any one of claims 32 to 34, wherein the neural network is further trained with dental restoration style information of the previously performed dental restoration cases as part of the training data. (Appendix 36) 36. The computing device of any one of claims 32 to 35, wherein the neural network is further trained with physician-selected information associated with the dentist who developed the dental restoration plan for the dental restoration patient. (Appendix 37) receiving a two-dimensional (2D) image of a dental restoration patient's current dental structure; providing the 2D image of the current dental structure of the dental restoration patient as input to a neural network trained with training data including 2D images of pre-restoration dental structures and corresponding 2D images of post-restoration dental structures of previously performed dental restoration cases; running the neural network using the input to generate a 2D image of a proposed dental structure for the dental restoration patient, the proposed dental structure being associated with a post-restoration outcome of the dental restoration plan for the dental restoration patient; and A method comprising: (Appendix 38) 38. The method of claim 37, wherein the neural network is a conditional adversarial network (cGAN) consisting of a generative network and a discriminative network, and providing the 2D image of the current dental structure of the dental restoration patient as the input to the neural network includes providing the 2D image of the current dental structure of the dental restoration patient as the input to the generative network of the cGAN. (Appendix 39) executing the generative network of the cGAN to generate a respective proposed post-restoration 2D image corresponding to each of the pre-restoration dental structures of the training data; providing each of the proposed restored 2D images and the pre-restoration dental structure corresponding to each of the proposed restored 2D images as inputs to a discriminative network of the cGAN; running the discriminative network to output a respective probability of each proposed restored 2D image representing a ground truth image; 39. The method of claim 38, further comprising training the cGAN by: (Appendix 40) 40. The method of any one of claims 37 to 39, wherein the neural network is further trained with dental restoration style information of the previously performed dental restoration cases as part of the training data. (Appendix 41) 41. The method of any one of claims 37 to 40, wherein the neural network is further trained with physician-selected information associated with the dentist who developed the dental restoration plan for the dental restoration patient. (Appendix 42) means for receiving a two-dimensional (2D) image of the dental restoration patient's current dental structure; means for providing the 2D image of the current dental structure of the dental restoration patient as input to a neural network trained with training data including 2D images of pre-restoration dental structures and corresponding 2D images of post-restoration dental structures of previously performed dental restoration cases; means for executing the neural network using the input to generate a 2D image of a proposed dental structure for the dental restoration patient, the proposed dental structure being associated with a post-restoration outcome of the dental restoration plan for the dental restoration patient; and An apparatus comprising: (Appendix 43) A non-transitory computer-readable storage medium encoded with instructions that, when executed, cause one or more processors of a computing system to: receiving a two-dimensional (2D) image of a current dental structure of a dental restoration patient; providing the 2D image of the current dental structure of the dental restoration patient as input to a neural network trained with training data including 2D images of pre-restoration dental structures and corresponding 2D images of post-restoration dental structures of previously performed dental restoration cases; running the neural network using the input to generate a 2D image of a proposed dental structure for the dental restoration patient, the proposed dental structure being associated with a post-restoration outcome of the dental restoration plan for the dental restoration patient; A non-transitory computer-readable storage medium. (Appendix 44) 1. A computing device comprising an input interface and a neural network engine, The input interface is configured to receive one or more three-dimensional (3D) tooth meshes associated with a current dental structure of a dental restoration patient and 3D component meshes representing generated shapes of dental appliance components; The neural network engine providing the one or more 3D tooth meshes and the 3D component meshes received by the input interface as inputs to a neural network trained with training data including ground truth dental appliance component shapes and corresponding 3D tooth meshes of dental restoration cases; and running the neural network using the provided inputs to generate an updated model of the dental prosthetic component for the current dental structure of the dental restoration patient. It is configured as follows: Computing devices. (Appendix 45) 45. The computing device of claim 44, wherein the neural network is a generative adversarial network (GAN) consisting of a generative network and a discriminative network, and the neural network engine is configured to provide the one or more 3D tooth meshes and the 3D component meshes received by the input interface as inputs to the generative network of the GAN to execute the neural network to produce the updated model of the dental prosthetic component for the current dental structure of the dental restoration patient. (Appendix 46) The neural network engine running the generative network of the GAN to generate an updated shape of the dental appliance component using the one or more 3D tooth meshes and the 3D component mesh; providing the updated shape of the dental appliance component, the one or more 3D tooth meshes, and the 3D component mesh as inputs to the classification network of the GAN; running the discriminative network to output a probability that the updated shape represents a ground truth shape; and training the GAN by 46. ​​The computing device of claim 45. (Appendix 47) 47. The computing device of any one of claims 44-46, wherein the neural network is further trained with geometry information of the ground truth dental appliance components as part of the training data. (Appendix 48) The dental appliance component comprises: Mold parting surface, Gum trim, door, window, Facial ribbon, Incisal ridge, Lingual shelf, Interdental matrix, Case frame, Parts labels, Door hinges, Door snap, Door vents, Snap clamp, or Center clip, 48. The computing device of any one of claims 44 to 47, comprising one or more of: (Appendix 49) receiving, at an input interface, one or more three-dimensional (3D) tooth meshes associated with the dental restoration patient's current dental structure and 3D component meshes representing generated shapes of dental appliance components; providing, by a neural network engine communicatively coupled to the input interface, the one or more 3D tooth meshes and the 3D component meshes received by the input interface as inputs to a neural network trained with training data including ground truth dental appliance component shapes and corresponding 3D tooth meshes of dental restoration cases; executing, by the neural network engine, the neural network using the provided inputs to generate an updated model of the dental appliance component for the current dental structure of the dental restoration patient; A method comprising: (Appendix 50) 49. The method of claim 49, wherein the neural network is a generative adversarial network (GAN) consisting of a generative network and a discriminative network, and executing the neural network to create the updated model of the dental prosthetic components for the current dental structure of the dental restoration patient includes providing, by the neural network engine, the one or more 3D tooth meshes and the 3D component meshes received by the input interface as inputs to the generative network of the GAN. (Appendix 51) running the generative network of the GAN to generate an updated shape of the dental appliance component using the one or more 3D tooth meshes and the 3D component mesh; providing the updated shape of the dental appliance component, the one or more 3D tooth meshes, and the 3D component mesh as inputs to the discriminant network of the GAN; running the discriminative network to output a probability of the updated shape representing a ground truth shape; 51. The method of claim 50, further comprising training the GAN with the neural network engine by (Appendix 52) 52. The method of any one of claims 49 to 51, wherein the neural network is further trained with geometry information of the ground truth dental appliance components as part of the training data. (Appendix 53) The dental appliance component comprises: Mold parting surface, Gum trim, door, window, Facial ribbon, Incisal ridge, Lingual shelf, Interdental matrix, Case frame, Parts labels, Door hinges, Door snap, Door vents, Snap clamp, or Center clip, 53. The method of any one of appendices 49 to 52, comprising one or more of: (Appendix 54) means for receiving one or more three-dimensional (3D) tooth meshes associated with the dental restoration patient's current dental structure and 3D component meshes representing generated shapes of dental prosthetic components; means for providing the one or more 3D tooth meshes and the 3D component meshes received by the input interface as inputs to a neural network trained with training data including ground truth dental appliance component shapes and corresponding 3D tooth meshes of dental restoration cases; means for executing the neural network using the provided input to generate an updated model of the dental prosthetic component for the current dental structure of the dental restoration patient; An apparatus comprising: (Appendix 55) A non-transitory computer-readable storage medium encoded with instructions that, when executed, cause one or more processors of a computing system to: receiving one or more three-dimensional (3D) tooth meshes associated with a current dental structure of a dental restoration patient and 3D component meshes representing generated shapes of dental prosthetic components; providing the one or more 3D tooth meshes and the 3D component meshes received by the input interface as inputs to a neural network trained with training data including ground truth dental appliance component shapes and corresponding 3D tooth meshes of dental restoration cases; running the neural network using the provided inputs to generate an updated model of the dental prosthetic component for the current dental structure of the dental restoration patient; A non-transitory computer-readable storage medium.

Claims

1. 1. A computing device comprising an input interface and a neural network engine, The input interface is configured to receive one or more three-dimensional (3D) tooth meshes associated with a current dental structure of a dental restoration patient and 3D component meshes representing generated shapes of dental appliance components; The neural network engine providing the one or more 3D tooth meshes and the 3D component meshes received by the input interface as inputs to a neural network trained with training data including dental appliance component shapes and corresponding 3D tooth meshes of dental restoration objects; and running the neural network using the provided input to generate an updated model of the dental prosthetic component for modifying the current dental structure of the dental restoration patient. It is configured as follows: Computing devices.

2. 2. The computing device of claim 1, wherein the neural network is a generative adversarial network (GAN) consisting of a generative network and a discriminative network, and the neural network engine is configured to provide the one or more 3D tooth meshes and the 3D component meshes received by the input interface as inputs to the generative network of the GAN to execute the neural network to produce the updated model of the dental prosthesis components for the current dental structure of the dental restoration patient.

3. The neural network engine running the generative network of the GAN to generate an updated shape of the dental appliance component using the one or more 3D tooth meshes and the 3D component mesh; providing the updated shape of the dental appliance component, the one or more 3D tooth meshes, and the 3D component mesh as inputs to the identification network of the GAN; running the discriminative network to output a probability that the updated shape represents a ground truth shape; configured to train the GAN by The computing device of claim 2 .

4. The computing device of any one of claims 1 to 3, wherein the neural network is further trained with configuration information of the shape of the dental appliance components as part of the training data.

5. The dental appliance component comprises: Mold parting surface, Gum trim, door, window, Facial ribbon, Incisal ridge, Lingual shelf, Interdental matrix, Case frame, Parts labels, Door hinges, Door snap, Door vents, Snap clamp, or Center clip, A computing device according to any one of claims 1 to 4, comprising one or more of:

6. 1. A method of generating an updated model by a computing device having an input interface and a neural network engine, comprising: the input interface receiving one or more three-dimensional (3D) tooth meshes associated with a dental restoration patient's current dental structure and 3D component meshes representing generated shapes of dental appliance components; the neural network engine communicatively coupled to the input interface provides the one or more 3D tooth meshes and the 3D component meshes received by the input interface as inputs to a neural network trained with training data including shapes of dental appliance components and 3D tooth meshes of corresponding dental restoration objects; the neural network engine executing the neural network using the provided inputs to generate an updated model of the dental appliance component for modifying the current dental structure of the dental restoration patient; A method comprising:

7. 7. The method of claim 6, wherein the neural network is a generative adversarial network (GAN) consisting of a generative network and a discriminative network, and executing the neural network to create the updated model of the dental appliance components for the current dental structure of the dental restoration patient includes providing, by the neural network engine, the one or more 3D tooth meshes and the 3D component meshes received by the input interface as inputs to the generative network of the GAN.

8. running the generative network of the GAN to generate an updated shape of the dental appliance component using the one or more 3D tooth meshes and the 3D component mesh; providing the updated shape of the dental appliance component, the one or more 3D tooth meshes, and the 3D component mesh as inputs to the identification network of the GAN; running the discriminative network to output a probability of the updated shape representing a ground truth shape; The method of claim 7 , further comprising training the GAN with the neural network engine by:

9. The method according to any one of claims 6 to 8, wherein the neural network is further trained with configuration information of the shape of the dental appliance components as part of the training data.

10. The dental appliance component comprises: Mold parting surface, Gum trim, door, window, Facial ribbon, Incisal ridge, Lingual shelf, Interdental matrix, Case frame, Parts labels, Door hinges, Door snap, Door vents, Snap clamp, or Center clip, The method according to any one of claims 6 to 9, comprising one or more of:

Images