Automated tooth implant planning

Abstract

A computer-implemented method for automated planning of a tooth implant in a dento-maxillofacial complex is described, wherein the method includes receiving or determining 3D virtual patient data representing a dento-maxillofacial complex comprising anatomical structures and at least one area of interest for placing the tooth implant, the 3D virtual patient data including 3D representations of anatomical structures within a vicinity of the at least one area of interest; determining a tooth implant model and / or one or more implant parameters defining a tooth implant model to be positioned at the at least one area of interest, using an implant planning engine, the implant planning engine being configured to receive the 3D representations of anatomical structures within a vicinity of the area of interest and to determine the tooth implant model and / or one or more implant parameters defining a tooth implant model based on clinical constraints, the clinical constraints including distance requirements of the tooth implant with regard to at least part of the 3D representations of the anatomical structures; and, rendering the implant model on a display or transforming the implant model into a data format for controlling a manufacturing apparatus, such as a 3D printer, to manufacture the tooth implant.

Description

[0001] Automated tooth implant planning

[0002] Technical field

[0003] The present disclosure relates to automated tooth implant planning, and in particular, though not exclusively, to methods and systems for automated tooth implant planning in a dento-maxillofacial complex and a computer program product for executing such methods.

[0004] Background

[0005] Accurate implant planning is a critical component of dental treatment. Placing and dimensioning an implant in a dento-maxillofacial complex can influence both functional and aesthetic outcomes. Generally, two primary methodologies guide implant planning, each with its specific focus depending on the clinical scenario: prosthetic-driven planning and bone-driven planning. When using prosthetic-driven planning, a final restorative outcome is prioritized. The implant position, orientation, and dimensions are determined primarily by the planned prosthetic restoration, ensuring ideal functional and esthetic results. The implant is aligned according to the future crown's position, favoring occlusal stability and harmonious integration into the dental arch. This method is widely regarded as the gold standard in modern implantology. When using bone-driven planning, the implant position is determined primarily by the quantity and quality of the available jawbone in which the implant should be placed. This approach is particularly useful when anatomical limitations or compromised bone availability are present.

[0006] To ensure the success and longevity of dental implants, established clinical guidelines should be followed. These guidelines define clinical constraints which help minimize complications, promote osseointegration, and maintain the health of surrounding hard and soft tissues. The clinical constraints may include a sufficient implant support by surrounding bone, a minimum distance to neighboring teeth and / or implants and sufficient distance to nerves and the nasal cavity. Besides that, there are mechanical requirements, as, for instance, the implant and a crown attached to it have certain limitations regarding deviation of their axis and / or their posterior and anterior region alignment.

[0007] As clinical constraints require positioning and dimensioning of an implant with high accuracy, thorough and accurate implant planning is essential before surgery. While software systems exist to assist clinicians, determining the ideal implant position, orientation, and dimensions remains a time-consuming and iterative process requiring significant experience.US 2023 / 419631A1 discloses a guided implant planning system including a prediction model which is trained to predict a so-called phantom crown and match the predicted phantom crown to pre-existing implant and / or crown libraries. Implant planning by such system is restoration-centric and library-dependent, with implant parameters being selected from a discrete set of predefined components. In this approach implant parameters are derived from a predicted phantom crown.

[0008] Consequently, there is a need for improved methods and systems for automatic implant planning which allows generation of tooth implants that are anatomically consistent with the area in the dentition in which the tooth implant needs to be implanted.

[0009] Summary

[0010] As will be appreciated by one skilled in the art, aspects of the present invention may be embodied as a system, method or computer program product. Accordingly, aspects of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, micro-code, etc.) or an embodiment combining software and hardware aspects that may all generally be referred to herein as a “circuit”, “module”, “model”, or “system”. Functions described in this disclosure may be implemented as an algorithm executed by a microprocessor of a computer.

[0011] Furthermore, aspects of the present invention may take the form of a computer program product embodied in one or more computer readable medium(s) having computer readable program code embodied, e.g., stored, thereon.

[0012] Any combination of one or more computer readable medium(s) may be utilized. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.A computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0013] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber, cable, RF, etc., or any suitable combination of the foregoing. Computer program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object-oriented programming language such as Java(TM), Smalltalk, C++ or the like and conventional procedural programming languages, such as the “C” programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer, or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0014] Aspects of the present invention are described below with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor, in particular a microprocessor or central processing unit (CPU), of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer, other programmable data processing apparatus, or other devices create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0015] These computer program instructions may also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other devices to function in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture includinginstructions which implement the function / act specified in the flowchart and / or block diagram block or blocks.

[0016] The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. Additionally, the Instructions may be executed by any type of processors, including but not limited to one or more digital signal processors (DSPs), general purpose microprocessors, application specific integrated circuits (ASICs), field programmable logic arrays (FPGAs), or other equivalent integrated or discrete logic circuitry.

[0017] The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustrations, and combinations of blocks in the block diagrams and / or flowchart illustrations, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

[0018] In an aspect, the embodiments in this disclosure relate to a computer-implemented method for automated planning of a tooth implant in a dento-maxillofacial complex, the method comprising: receiving or determining 3D virtual patient data representing a dento-maxillofacial complex comprising anatomical structures and at least one area of interest for placing the tooth implant, the 3D virtual patient data including 3D representations of anatomical structures within a vicinity of the at least one area of interest. The method may further comprise determining an implant model and / or one or more implant parameters of an implant model that needs to be positioned at the at least one area of interest, wherein the implant model and / or the one or more implant parameters defining the implant model is determined based on a implant planning engine, which is configured to receive the 3D representations of anatomical structures within a vicinity of the area of interest and to determine the implant model and / or the one or more implant parameters based onclinical constraints, the clinical constraints including distance requirements of the tooth implant with regard to the 3D representations of the anatomical structures.

[0019] In an embodiment, the method may include determining an implant model, e.g. a mesh representation, based on the implant parameters.

[0020] In an embodiment, the method may include rendering the implant model and, optionally, the 3D virtual patient data and the crown design on a display.

[0021] In an embodiment, the method may include transforming the implant model into a data format for controlling a manufacturing apparatus, such as a 3D printer, to manufacture the tooth implant. Hence, a data file may be generated which includes the determined tooth implant data and control information for controlling the manufacturing apparatus.

[0022] Hence, the method allows automatic planning and design of a tooth implant for an area of interest of a dento-maxillofacial complex taking the clinical constraints associated with the area of interest into account. An implant planning engine is used to directly determine a tooth implant model and / or one or more implant parameters defining a tooth implant model from 3D representations of anatomical structures within a vicinity of the area of interest, while taking clinical constraints into account. This way, a direct functional relationship between anatomical structures and the tooth implant model and / or implant parameters under clinical constraints is established.

[0023] The method uses 3D virtual patient data, which represent at least part of a dento-maxillofacial complex. The dento-maxillofacial complex comprises several anatomical structures and at least one area of interest. In the present disclosure, the term “area of interest" refers to an area where a tooth implant can be placed. The 3D virtual patient data can comprise various structures in the dento-maxillofacial complex, including the several anatomical structures and the at least one area of interest. Within a vicinity of the at least one area of interest, the 3D virtual patient data can include volumetric representation of anatomical structures adjacent to the area of interest. Volumetic representation may include mesh representations, point cloud representation and any other format for representing anatomical elements of a dento-maxillofacial complex.

[0024] In case of several areas of interest, volumetric representations can be present for anatomical structures adjacent to each of the several areas of interest. The method can determine one or more implant parameters for a tooth implant to be positioned at the area of interest or at one of the areas of interest. The implant parameters can describe various attributes of the tooth implant. According to one embodiment, the implant parameters describe a position of the tooth implant within the area of interest and / or an orientation of the tooth implant and / or dimensions of the tooth implant.The one or more implant parameters are determined by an implant planning engine, where the implant planning engine can be configured to receive the 3D representations of anatomical structures within a vicinity of the area of interest and to determine and output the one or more implant parameters based on clinical constraints. The clinical constraints define at least some requirements a tooth implant placed in a dento-maxillofacial complex needs to fulfill. The clinical constraints may include distance requirements of the tooth implant with regard to the anatomical structures. In that way, the method can determine one or more implant parameters defining a tooth implant to be placed in a dento-maxillofacial complex. The determined implant parameters can fulfill each or at least some, preferably most of the clinical constraints. When outputting, e.g. displaying the determined implant parameters or a tooth implant model, e.g. a mesh, based on the implant parameters to a user, the user can evaluate the results and use them for further planning steps.

[0025] In some embodiments, determining 3D virtual patient data may comprise: receiving 2D and / or 3D maxillofacial data representing at least parts of the dento-maxillofacial complex, received 2D and / or 3D maxillofacial data preferably comprise (CB)CT data, IOS data, panoramic radiographs and / or extraoral scan data; and aligning received 2D and / or 3D data with each other, thus forming 3D virtual patient data as combined 3D representation of the maxillofacial complex. In this way, a combined 3D representation of the maxillofacial complex and, thus, 3D virtual patient data can be determined. The combined 3D data can be high resolution data and represent anatomical structures that are not visible at some dento-maxillofacial data.

[0026] In some embodiments, determining 3D virtual patient data comprises segmenting received 2D and / or 3D maxillofacial data and / or the combined 3D representation into anatomical structures. Hence, alignment of the maxillofacial data from various sources can be facilitated.

[0027] In some embodiments, the method may comprise receiving the area of interest via an input interface or determining the area of interest by analyzing the 3D virtual patient data regarding certain features, and inputting a received or determined area of interest into the implant planning engine, the implant planning engine being configured to base determining of the one or more implant parameters additionally on the received / determined area of interest. In this way, the efficiency of the method can be improved and determining suitable implant parameters can be speeded up. Embodiments of the “certain features” may include a dentition gap, a rapture tooth, or a tooth cavity.

[0028] In some embodiments, the method may comprise receiving a designed crown, the designed crown being designed to be put within the area of interest and being attached to the tooth implant, inputting the designed crown to the implant planning engine, anddetermining the one or more implant parameters additionally based on a received designed crown. Hence, a crown that was designed by a separate system or entity can be used and the constraints of the tooth implant with regard to the designed crown can be considered.

[0029] In some embodiments, the clinical constraints may comprise at least two categories of constraints, a category of constraints defining an importance that a constraint of the category is met, the implant planning engine being configured to determine the one or more implant parameters in such a way that constraints of a high importance category are met, while constraints of a low importance category may be violated.

[0030] In an embodiment, there are at least two categories of constraints, where a high importance category define hard constraints and a low importance category define soft constraints, wherein hard constraints have to be met, while soft constraints may be violated.

[0031] In some embodiments, the method additionally comprises preprocessing of the volumetric representations of anatomical structures. In an embodiment, the preprocessing may include transforming a 3D representation into a representation that can be processed by the implant planning engine. For example, in an embodiment, a mesh representation may be transformed into a point cloud which can be processed by a trained model, e.g. a trained deep neural network, of the implant planning engine. Hence, the anatomical structures can be represented in such a way that the implant planning engine can process them in a suitable way. The preprocessing means can be regarded as part of the implant planning engine.

[0032] In some embodiments, the method additionally comprises receiving data regarding commercially available tooth implants, determining the one or more implants parameters additionally constrained by the received data so that one or more determined implant parameters describe one of the commercially available tooth implants. Hence, a method can provide exactly those tooth implants which a user can purchase. The user does not have to find a suitable tooth implant and / or adapt the planning so that the purchased tooth implant really suits the automated planning result.

[0033] In some embodiments, the implant planning engine may be based on an optimization algorithm, wherein operation of the implant planning engine may comprise: receiving or generating an initial set of implant parameters; determining costs of the initial set of implant parameters based on a cost function, the cost function computing the costs based on the clinical constraints; and optimizing one or more of the implant parameters by adapting the initial set of implant parameters until the cost determined by the cost function is minimized and / or a stopping criteria is met.

[0034] In some embodiments, the implant planning engine comprises a trained predictive deep learning model, the trained predictive deep learning model being trained to receive mesh representations of the anatomical structures or representations derived fromthe mesh representations of the anatomical structures and to output a prediction for the one or more implant parameters representing a tooth implant meeting clinical constraint.

[0035] In some embodiments, the implant planning engine comprises a trained generative deep learning model, the trained generative deep learning model being trained to receive mesh representations of the anatomical structures or representations derived from the mesh representations of the anatomical structures and to generate and output the one or more implant parameters representing a tooth implant meeting the clinical constraints.

[0036] In some embodiments, the trained predictive deep learning model or the trained generative deep learning model is configured to receive sets of 3D virtual patient data and associated approved implant parameters a ground truth and to finetune model parameters based on the received sets of 3D virtual patient data and the approved implant parameters, the approved implant parameters being generated at earlier usages of the implant planning engine and being approved by a user. Hence, an already trained deep learning model can be adapted to results of earlier plannings. In this way, data that were generated after the initial training of the deep learning model can be used to improve the planning results of the implant planning engine.

[0037] In some embodiments, the method additionally comprises determining sleeve parameters describing an implant sleeve, the sleeve parameters being determined by a sleeve generator based on the implant parameters. Hence, a user of the method can get even more support for a surgery for placing the implant.

[0038] In some embodiments, the implant planning engine is configured to plan two or more implants, the two or more implants being planned successively and each planned implant forming a new anatomical structure at the 3D virtual patient data. In this way, the method for planning a single tooth implant can be used for multiple implant planning without considerable adaption. It may be beneficial if the sequence of the implant planning reflects a planning strategy. It can be increasingly difficult to place a further tooth implant after placing other tooth implants before. This can be reflected by planning important and / or readily visible areas of interest first and then proceeding to less important and / or less visible areas of interest.

[0039] In some embodiments, the implant planning engine is configured to plan two or more implants, the two or more implants being planned in parallel, the implant planning engine using mutual dependencies between the several implants as further clinical constraints. In this way, the mutual dependencies between the tooth implants can be considered straight away.

[0040] According to another aspect, embodiments in this disclosure relate to a computer implemented method for training an implant planning engine comprising a predictive deep learning model or a generative deep learning model to automatically predictimplant parameters, the method comprising: receiving training data comprising sets of 3D virtual patient data, associated ground truth implant parameters, and, optionally, clinical constraints, the 3D virtual patient data representing a dento-maxillofacial complex comprising anatomical structures and at least one area of interest for placing the tooth implant, the 3D virtual patient data including a volumetric representation of the anatomical structures, and the ground truth implant parameters describing a position of the tooth implant within an area of interest and / or an orientation of the tooth implant and / or dimensions of the tooth implant; providing one or more sets of the 3D virtual patient data as inputs to the implant planning engine and obtaining from the output of the implant planning engine estimated implant parameters; and training the parameters of the implant planning engine by updating its parameters by minimizing a loss function until a stopping criteria is met, wherein the loss function penalizes deviations between the estimated implant parameters and the ground truth implant parameters, wherein the optional clinical constraints are used as input features for the model and / or incorporated into the loss function as additional constraints.

[0041] According to another aspect, embodiments in this disclosure relate to a system for automated planning of a tooth implant in a dento-maxillofacial complex, the system comprising: a computer readable storage medium having computer readable program code embodied therewith; and a processor, preferably a microprocessor, coupled to the computer readable storage medium, wherein responsive to executing the computer readable program code, the processor is configured to perform executable operations comprising: receiving or determining 3D virtual patient data representing a dento-maxillofacial complex comprising anatomical structures and at least one area of interest for placing the tooth implant, the 3D virtual patient data including volumetric representations of anatomical structures within a vicinity of the area of interest; determining one or more implant parameters for the tooth implant to be positioned at the at least one area of interest, the implant parameters describing a position of the tooth implant within the area of interest and / or an orientation of the tooth implant and / or dimensions of the tooth implant, the one or more implant parameters being determined based on an implant planning engine, the implant planning engine being configured to receive the volumetric representations of anatomical structures within a vicinity of the area of interest and to determine the one or more implant parameters based on clinical constraints, the clinical constraints including distance requirements of the tooth implant with regard to the volumetric representations of the anatomical structures.

[0042] The present disclosure may also relate to a computer program product comprising software code portions configured for, when run in the memory of a computer, executing the method steps according to any of process steps described above.The present disclosure may also relate to a computer readable medium storing data which, when executed by a manufacturing machine, like a 3D printer or a CNC machine, controls the manufacturing machine to produce an implant based on one or several implant parameters determined by the method steps according to any of the process steps described above.

[0043] The invention will be further illustrated with reference to the attached drawings, which schematically will show embodiments according to the invention. It will be understood that the invention is not in any way restricted to these specific embodiments.

[0044] Brief description of the drawings

[0045] Fig. 1 depicts a high-level overview of a system for automated tooth implant planning in a dento-maxillofacial complex according to an embodiment;

[0046] Fig. 2A-2E show exemplary tooth implant and sleeve models generated by a system for automated tooth implant planning according to an embodiment;

[0047] Fig. 3A to 3D shows examples of clinical constraints for implant planning; Fig. 4A and 4B depicts the formation of a 3D data representation of part of a dento-maxillofacial complex of a patient based on various data modalities according to one embodiment;

[0048] Fig. 5 depicts system for automated tooth implant planning in a dento-maxillofacial complex based on an optimization algorithm according to an embodiment;

[0049] Fig. 6 depicts a decision diagram for an optimization algorithm for tooth implant planning according to an embodiment;

[0050] Fig. 7A-7D show steps executed by an optimization algorithm for tooth implant planning according to an embodiment;

[0051] Fig. 8 illustrates a method for automated tooth implant planning in a dento-maxillofacial complex according to an embodiment;

[0052] Fig. 9 depicts a system for training an implant planning engine using a predictive deep learning model according to an embodiment;

[0053] Fig. 10 depicts an inference process of a system with a trained predictive deep learning model according to an embodiment;

[0054] Fig. 11 depicts a system fortraining an implant planning engine using a generative deep learning model according to an embodiment;

[0055] Fig. 12 depicts a method of a training process of an implant planning engine comprising a predictive deep learning model or a generative deep learning model according to the present disclosure;Fig. 13A and 13B depict graphical models generated by a system for automated tooth implant planning according to an embodiment;

[0056] Fig. 14 depicts a block diagram illustrating an exemplary data processing system that may be used with embodiments described in this disclosure.

[0057] Description of the embodiments

[0058] The embodiments of the present disclosure generally relate to automated implant planning of a tooth implant. The term “tooth implant” according to the present disclosure refers to an object that is used for replacing a natural tooth and that is at least partly implanted into the dento-maxillofacial complex of a patient, generally in a jawbone. Generally, a tooth implant can have various shapes and provide various details and functionalities after being implanted into the jawbone. One of the basic functionalities may be to carry an artificial part of a dentition of a patient, e.g. a crown or a bridge, hereinafter generally referred to as crown. This demands that the tooth implant is safely attached to the jawbone and to the artificial part of the dentition. According to one embodiment, a tooth implant may represent a substantially cylindric object that is placed in a hole drilled into a jawbone. The tooth implant may have a screw thread at the outer surface for screwing the tooth implant into the jawbone. A crown may be attached to the implant, e.g. by a screw or other suitable means.

[0059] Fig. 1 depicts a high-level overview of a system 100 for automated planning of a tooth implant in a dento-maxillofacial complex according to an embodiment of the present disclosure. System 100 may receive 3D virtual patient data 102 representing 3D data of a dento-maxillofacial complex of a patient. The 3D data may include volumetric data, e.g. radiographic (X-ray) data generated by an X-ray scanning device such as a CT or a CBCT scanner. Typically, 3D radiographic data include voxel representations of a dento-maxillofacial complex. The 3D data may also include optical scan data, e.g. inter-oral scanning (IOS) data, generated by an optical scanner. IOS data typically may include pointcloud or mesh representations of visual parts of the dento-maxillofacial complex, e.g. the dention.

[0060] The virtual patient data 102 are input to an implant planning engine 104, i.e. an algorithm which is configured to generate an implant model or implant parameters defining an implant model for a tooth implant treatment plan based on the input data and based on clinical constraints 106. In some embodiments, clinical constraints 106 may be input to the implant planning engine from a storage or via an interface. In other embodiments, the clinical constraints 106 may be an integral part of the implant planning engine so that they cannot be controlled and / or modified as separate parameters. In thatcase, the implant planning engine may be represented by the dotted line around the implant planning engine and the clinical constraints.

[0061] In some embodiments, the 3D virtual patient data may include a crown design. The crown design may be generated by a crown design system based on the data of a dento-maxillofacial complex of a patient. For example, an automatic crown design system as disclosed in WO2025 / 016984 A1 may be used, the content of which is incorporated herewith by reference. Alternatively, the crown design may be determined by a user using a crown design tool that are known in the art. Typically, the crown design may be represented as 3D mesh, point cloud, or volumetric data, that accurately define the shape of the crown, including occlusal surface, lingual fossa, linguoincisal edge, and other characteristics.

[0062] The implant planning engine 104 may be configured to receive a digital crown design, i.e. crown model 108, as an additional input so that it can determine the implant model and / or the one or more implant parameters based on the crown design 108 and the clinical constraints. After implantation, a tooth implant structurally and mechanically supports a crown that can be manufactured based on the crown model. The implant planning engine 104 may generate an implant model, i.e. set of point such as a point cloud or mesh representing the implant model, and / or one or more implant parameters 110, e.g. parameters values (dimensions, shape, type, etc.) that can be used to define an implant model at its output. The implant model and / or one or more implant parameters may be used as input for a tooth implant module 112 which is configured to generate a toot implant model 116 in a format that is suitable for manufacturing a physical model. In an embodiment, the implant model and / or parameters may also be input to an implant sleeve module 114 which is configured to generate a digital implant sleeve design, i.e. implant sleeve model 118. The implant sleeve model may be physically implemented as a tube of a biocompatible material that is positioned in a so-called dental template of a surgical guide and that fits securely over a surgical drill ensuring accurate drilling depth and angulation during implant surgery.

[0063] The tooth implant module and the implant sleeve module may generate models in different data formats so that they can be used one or more further operations by the implant planning system. For example, in an embodiment, the implant model and / or sleeve model may in a data format for imaging so that it can be rendered by a graphical user interface (GUI) on a display for visualization of the model. The GUI may allow a user to interact with the implant model and the generation of the implant model by the system. In another embodiment, the models may be formatted based on a data format that is suitable for 3D printing or a computer numerical control (CNC) machine. Data format may include STL, FBX, OBJ, 3MF, PLY, etc. This way the modeled tooth implant and sleeve may be produced. In a further embodiment, the tooth implant module may be configured to match the designed tooth implant with a list of commercially available tooth implants. The module maycheck if a commercially available tooth implant suits to the planned tooth implant 112 and, if not, can test which of the commercially available tooth implants suits best to the planned implant situation.

[0064] The output data 110 generated by implant planning engine 104 may be a tooth implant model and / or one or more implant parameters defining an implant model. The tooth implant model may a 3D representation of a tooth implant. Such representation may include a point cloud or a mesh representation of the tooth implant and may be associated with a position of the tooth implant within an area of interest, an orientation of the tooth implant model and / or dimensions of a tooth implant model.

[0065] An implant parameter may be associated with a structural feature of the tooth implant 112 ora specific implanting situation of the implant in a dento-maxillofacial complex. Implant parameters may include any suitable type of information for describe a tooth implant and / or a particular implant treatment. An implant parameter may be represented by a natural number, a real number, a set of several numbers (e.g. a vector indicating a place in a 3D space), a category, or the like. Implant parameters may include structural information of a 3D object representing an implant, dimensions and a curvature defining one or more outer surfaces of the tooth implant.

[0066] In some embodiments, implant parameters may include dimensions of the tooth implant model. An implant parameter may define a length of the tooth implant. Typical lengths of tooth implants range from about 6 millimeters (or even below) to 15 millimeters (and above). Another implant parameter may be the diameter of the tooth implant. Diameters of typical commercially available tooth implants range from 3.5 millimeters (or even below) to 5 millimeters (and above). Longer and thicker tooth implants can provide more support for an attached crown, but may be too big for placing them into the jawbone without violating clinical constraints.

[0067] In an embodiment, an implant parameter may include information regarding the type of tooth implant. Generally, there are several primary types of implants known in the art. The selection may be important for addressing specific clinical indications and prosthetic outcomes. Two basic primary types of implants are bone-level implants and tissue-level implants. Bone-level implants are placed at the level of the alveolar bone crest and are commonly used when a more natural emergence profile is desired. Bone-level implants allow for greater flexibility in soft tissue management and are often the system of choice for esthetically critical zones (anterior regions). Tissue-level implants feature a built-in collar and extend above the alveolar crest, with the implant-abutment connection situated at or above the soft tissue level. They are generally easier to maintain hygienically and are often used in posterior regions where esthetic demands are lower, reducing the need for additional soft tissue shaping, but functional requirements are paramount. The appropriate implant typemay be selected based on clinical and prosthetic considerations, patient-specific anatomy, and the overall treatment plan.

[0068] In an embodiment, an implant parameter may include information regarding the position and / or the orientation (the pose) of a tooth implant model in the dento-maxillofacial complex. Information regarding the position and / or orientation of the implant model is needed to determine how a tooth implant is implanted in the dento-maxillofacial complex, particularly in the jawbone. The position and / or orientation may be based on a common reference, which is shared with other data of the system. This common reference may be a reference point in the 3D virtual patient data 102. The position and / or orientation may be defined with regard to a reference point or a reference line of the tooth implant model. This reference point or reference line may be a predetermined point or line of the tooth implant. In an embodiment, the reference point or reference line may be selected as a central point at an implant platform, an apex at the lower end of the tooth implant model or a center line of a cylindric tooth implant model.

[0069] In an embodiment, an implant parameter may include a 3D representation of the tooth implant model. Such representation may include a point cloud or a mesh representation of the tooth implant. Such tooth representation may be associated with a position of the tooth implant within an area of interest, an orientation of the tooth implant model and / or dimensions of a tooth implant model.

[0070] In an embodiment, an implant parameter may include a similarity matrix describing a transformation of a default tooth implant model into a final tooth implant model, wherein the transformation matrix may include scaling, translation and rotation parameters that scales, moves and / or rotates the default tooth implant model into the final tooth implant model.

[0071] The implant sleeve module may determine an implant sleeve model 114 based on the implant parameters 110. An implant sleeve model may be represented as a cylindrical guide which is used in guided implant surgery. The purpose of the implant sleeve is to ensure precise drilling and implant placement. Embedded in a surgical guide, it can control the position, angle, and depth of the drill, reducing deviations, minimizing errors, and improving surgical accuracy. An implant sleeve model 114 may be generated based on the implant parameters using simple heuristics. Generally, the implant sleeve is a hollow cylinder defined by an inner diameter, an outer diameter and a sleeve length. In an embodiment, a simple heuristic for dimensioning the implant sleeve may be: the inner diameter is set to the smaller value between an implant diameter and a minimum of 2.5 millimeter, while the outer diameter is set to the inner diameter plus 1 mm, capped at 3.5 millimeter to align with common sleeve specifications. The sleeve length can be fixed at 5 millimeter. During an implant operation, the implant sleeve can be positioned above, e.g. 4 millimeter, the implantplatform centroid's intersection with the jawbone ridge along the occlusal direction and is aligned with the implant orientation.

[0072] Fig. 2A-2E show exemplary tooth implant and sleeve models generated by a system for automated tooth implant planning according to an embodiment. These models may for example be generated by a system for implant planning as described with reference to Fig. 1. Fig. 2A shows a tooth implant 202 which is basically reduced to a cylinder with a diameter d and a length I. This tooth implant 202 is defined different implant parameter including but limited to parameters such as “length I”, “diameter d”, and “shape”. The implant platform 206 is the “top” surface of the implant, i.e. the side of the implant 202 which faces the crown. The implant apex 208 is located at the other end of the implant 202 and can be pointed for supporting placement in the jawbone. Fig. 2B shows a crown model 204 which may be provided as a crown design 108 to the input of a automated tooth implant planning system. The crown model 204 will be attached to one end of tooth implant 202, as indicated in Fig. 2C and 2E. Fig. 2D and 2E illustrate the usage of an implant sleeve 210 with an inner diameter ds an outer diameter ds,0and a sleeve length Is.

[0073] The 3D virtual patient data may be a 3D representation of at least parts of the dento-maxillofacial complex and provides a digital representation of a patient’s maxillofacial anatomy including a dentition. The 3D virtual patient data can be represented in various ways, e.g., by meshes, point clouds, voxelized volumes, etc. The dentition may include an area of interest in which a tooth implant will be placed. To position the tooth implant in the area of interest information associated with the direct surrounding, e.g. neighboring teeth, the wear of teeth (tooth wear), dental arc, etc. may be needed for accurate implant, wherein the tooth implant model and the crown should not only meet functional but also esthetic demands.

[0074] To that end, the 3D virtual patient data preferably includes all, or at least a substantial part of all, anatomical structures that may be relevant for the tooth implant planning process. Anatomical structures may include various parts of the dento-maxillofacial complex including but not limited to inferior alveolar nerve (IAN), upper and lower jawbone, gingiva, teeth, incisive canal, maxillary sinus, nasal passage, existing implants, and crowns. The 3D virtual patient data may be represented in various ways, e.g. based on volumetric data, such as a point cloud or mesh representation, which allow precise distance measurements and / or precise spatial analysis in the vicinity of the area of interest.

[0075] The area of interest area where a tooth implant should be placed. This may be a location in the dentition where a tooth is missing. Alternatively, it may be location where a tooth that has to be removed. Such a tooth may be characterized a rapture tooth, a tooth with a cavity or another distinguishing feature. This way, an area of interest may be determined based on an analysis of the 3D virtual patient data. In an embodiment, an area of interestcan be defined by a tooth value, e.g. the FDI number of a tooth that needs to be replaced by a tooth implant with associated crown.

[0076] A vicinity of the area of interest 108 describes a region around the area of interest 108. It can have a relatively arbitrary shape and can be defined in such a way that anatomical structures adjacent to the area of interest 108 and potentially relevant for the tooth implant planning are intersecting the vicinity. The vicinity of the area of interest may comprise neighboring teeth, part of the jawbone, part of a nerve, part of a nasal passage (if the area of interest is at one of the upper incisors), etc. The vicinity of the area of interest may comprise structures that are withing a predetermined distance, from the area of interest . In an embodiment, the distance may be 5 cm, preferably 3 cm, more preferably 1 cm.

[0077] The tooth implant planning system uses clinical constraints to generate a tooth implant model. To that end, the implant planning engine is configured to receive 3D data representations, e.g. mesh representations, of anatomical structures within a vicinity of the area of interest and to determine the one or more implant parameters based on clinical constraints. Generally clinical constraints provide information, e.g. guidelines, for a clinician placing an implant in a dento-maxillofacial complex. Clinical constraints are typically based on knowledge and experience of how an implant has to be dimensioned and placed so it has the least impact on the jawbone, gingiva, etc. of a patient.

[0078] Fig. 3A to 3D show some examples of clinical constraints for implant planning. Fig. 3A shows an implant 302 that is implanted in a jawbone 304. Clinical constraints may require that neighboring anatomical structures (neighboring tooth 306 and neighboring tooth 308) keep a distance to neighboring tooth 306 of at least 1.5 millimeter. In Fig. 3B, implant 302 has neighboring tooth 306 and - as another neighboring anatomical structure -neighboring implant 310. A clinical constraint may require that the distance between implant 302 and implant 310 is at least 3 millimeters. Fig. 3C and 3D illustrate a constraint relating the amount of jawbone that should present around the implant 302. For example, at least 1 millimeter of jawbone should be present at all sides of the tooth implant 302 (besides the implant platform side). Fig. 3C shows a tissue-level implant, Fig. 3D shows a bone-level implant. The figures only illustrate a few clinical constraints that could be used. Many different clinical constraints can be used. Further examples of possible constraints that could be used by the tooth implant planning system are provided below.

[0079] The clinical constraints may, for example, include distance requirements of the tooth implant with regard to adjacent anatomical structures, e.g. neighboring nerves, nasal passage, neighboring teeth, neighboring implants, jawbone rim, etc. The clinical constraints may also include mechanical requirements, for example a relative position between the tooth implant and the crown, e.g. a limited deviation of their axis and / or their posterior and anterior region alignment.Clinical constraints may include information about sufficient bone support for the tooth implant. For example, typically, a minimum of 1.5 millimeter of surrounding jawbone around the implant in all directions except coronally (above the implant platform) is required to ensure adequate bone support and minimize the risk of bone resorption. Clinical constraints may further include a distance between neighboring implants. For example, typically it is recommended to maintain a minimum of 3 millimeters between neighboring implants at the platform level and at least 2 millimeters of separation at the apex to preserve bone health and ensure soft tissue stability.

[0080] Clinical constraints can include a distance of the toot implant from adjacent teeth. A minimum distance of 1.5 millimeters may be maintained between the implant and the roots of adjacent teeth to prevent damage to surrounding tooth structures and to maintain periodontal health. Clinical constraints can include a distance from other anatomical structures: for example, a minimum 1 millimeter distance may be maintained from critical anatomical landmarks, including the nasal cavity, incisive canal, and maxillary sinus. If bone height limitations exist due to sinus proximity, the 1 millimeter distance can become a flexible guideline rather than a strict requirement and a sinus lift procedure can be considered.

[0081] Clinical constraints may include information for protection of the Inferior Alveolar Nerve (IAN): for example typically a clearance of 2 to 3 millimeters from the IAN is necessary, with 2 millimeters regarded as the minimum safe distance to avoid nerve injury and related complications.

[0082] Clinical constraints may include information for a suitable posterior region alignment. For example, implants placed in the posterior region (premolars and molars: FDI positions 4 to 8), the implant may be centered beneath the planned prosthetic crown.

[0083] Further, a maximum 0 to 20° deviation from the ideal prosthetic axis is acceptable. Moreover, the implant platform should be positioned under the central fossa of the crown, with a tolerance of up to 1 millimeter from the ideal center.

[0084] Clinical constraints may include information fora suitable interior region alignment. In the anterior region, the implant platform should align within 1 millimeter of the crown’s center of mass and the midpoint of the cingulum. Further, the implant can be angled up to 30° relative to the long axis of the crown.

[0085] Clinical constraints may include information so that the biologic width is respected. For example, the implant shoulder should be placed at a distance of 3 to 4 millimeter from the gingival emergence line to respect the biologic width, supporting healthy soft tissue integration and long-term stability.

[0086] Clinical constraints may include information for vertical positioning. For example, the implant platform should be positioned at bone level or up to 2 millimetersubcrestally (below the alveolar ridge) for single-tooth restorations, depending on esthetic and functional requirements.

[0087] Clinical constraints may include parallelism for multiple implants. When placing multiple implants, the orientation of each new implant should be aligned within 10° of adjacent implants to promote parallelism and facilitate prosthetic rehabilitation.

[0088] Clinical constraints may include information for suitable tooth implant dimensions. For example, a common guideline may be used to maximize the implant’s surface area by selecting the largest possible length and diameter, provided all spatial and anatomical constraints are respected. A larger implant generally enhances primary stability, promotes better osseointegration, and can contribute to the long-term success and load distribution of the restoration. Factors such as bone quality, anatomical limitations, prosthetic requirements, and functional load considerations may necessitate selection of shorter or narrower implants.

[0089] Apart from the clinical constraints, the implant planning engine can additionally plan the tooth implant based on aesthetic and / or prosthetic constraints. These constraints consider the optical appearance of the implanted artificial part and its functionality within the dentition of the patient. This can include placing a crown (and the tooth implant attached to it) is line and aligned with the dental arch.

[0090] Fig. 4A and 4B depicts the formation of a 3D data representation of part of a dento-maxillofacial complex of a patient based on various data modalities according to one embodiment. The 3D virtual patient data 416 can be determined based on 2D and / or 3D data and may be formed based on different image data modalities, including cone-beam computed tomography (CBCT) data, intraoral scans (IOS) data, panoramic radiographs, and extraoral scans. The 3D virtual patient data represents an accurate 3D model of the dento-maxillofacial complex of the patient.

[0091] As shown in Fig. 4A, CBCT or CT data 402 may be input to a segmentation process 404, to segment the (CB)CT data, typically voxel representations, into separate anatomic structures of the dento-maxillofacial complex. Similarly, IOS data 406 may be input to a segmentation process 408, to segment the IOS data, typically mesh or point cloud data. Suitable methods for classification and segmentation volumetric data (voxel representations and point cloud and mesh representations) are, for instance, known from WO2019 / 002631 A1 and WO2020 / 127398 A1, the content of which is incorporated herewith by reference. The data can be segmented for determining 3D models of anatomic structures of the dento-maxillofacial complex, e.g. Inferior Alveolar Nerve (IAN), upper and lower jawbone, teeth including roots, incisive canal, maxillary sinus, nasal passage, existing implants, crowns, etc.

[0092] The segmented 3D structures of the dento-maxillofacial complex may be aligned and registered to form an accurate 3D model of the dento-maxillofacial complex.Methods for alignment and registration method are described in W02020 / 007941A1, the content of which is incorporated by reference. Fig. 4B depicts an example of 3D virtual patient data 416, including 3D models of anatomic structures of the dento-maxillofacial complex that include a dentition. The 3D virtual patient data may form an accurate representation of part of a dento-maxillofacial complex of a patient.

[0093] Fig. 4A shows a dento-maxillofacial complex of a patient including a dentition comprising an missing tooth at an area of interest 604 in the lower jaw 606. The area of interest may be associated with a tooth type which can be characterized by a tooth number, e.g. an FDI value.

[0094] In some embodiment, an automatic crown design module 410 may generate a crown design for an area of interest in the dentition based on the 3D virtual patient data. An example for automatic generation of a crown design is described in WO2025 / 016984A1. To that end, the automatic crown design module 410 may be configured to receive information about an area of interest 412 in the dentition. This way, a crown design may be generated for placement in the area of interest. The crown design may be aligned and registered with the segmented 3D structures. In this way, 3D virtual patient data 416 for the input data for the implant planning engine of the tooth implant planning system of Fig. 1.

[0095] In an embodiment, 3D virtual patient data can, for instance, comprise mesh representations for: Inferior Alveolar Nerve (IAN) (left and right), Upper and Lower Jawbone meshes (representing the maxilla and mandible), Gingiva mesh, Teeth meshes, Incisive Canal mesh, Maxillary Sinus meshes (left and right), Nasal Passage mesh, Existing Implants meshes, Crowns meshes (representing implant-supported and tooth / non-tooth-supported crowns), Designed crown mesh (Optional).

[0096] Besides calculating distances between the implant and surrounding meshes to satisfy spatial clinical requirements, additional preprocessing of the 3D virtual patient data may be necessary before inputting them to an implant planning engine according to the present disclosure. These preprocessing steps may include determining key points and / or vectors that may be used for accurately measuring clinical requirements in 3D space.

[0097] In an embodiment, the virtual patient data may be preprocessed to determine a so-called wax-up crown drilling point and / or an emergence axis. In this case, the preprocessing in particular relates to the crown design, which may also be referred as a wax-up crown or in short a wax-up. The wax-up crown drilling point refers to the point where the implant abutment is going to be connect to the wax-up crown. The emergence axis represents the ideal direction for implant placement. This axis is determined by the wax-up's occlusal orientation, while also considering the available bone and the orientation of neighboring teeth roots. In cases of bone-driven planning, where a designed crown is unavailable, the imaginary wax-up emergence axis can be estimated based on theneighboring teeth and bone structure. The wax-up drilling point may be determined differently depending on the target tooth type.

[0098] In a further embodiment, the virtual patient data may be preprocessed to determine a bone ridge, which is a point where the wax-up crown's emergence axis intersects the jawbone (effectively defining the bone level).

[0099] In another embodiment, the virtual patient data may be preprocessed to determine an emergence line of the wax-up through the gingiva. This line or curve delineates the region where the wax-up crown emerges from the gingiva. It is located by identifying the overlap between the gingiva and the wax-up crown. For bone-driven planning, this line can be estimated using the desired emergence axis and the bone ridge.

[0100] Fig. 5 depicts an embodiment of a system for tooth implant planning in a dento-maxillofacial complex according to an embodiment. In this embodiment, the implant planning engine 510 may comprise an optimization algorithm. System 500 receives 3D virtual patient data 502 that represent a dento-maxillofacial complex comprising anatomical structures and at least one area of interest 506 for placing the tooth implant. The 3D virtual patient data 502 may include 3D representations, typically representation based on volumetric data, such as meshes, of anatomical structures within a vicinity of the at least one area of interest. In an embodiment, the implant planning engine 510 may receive a crown 504 design (e.g. represented as a mesh). As shown in the figure, the implant planning engine 510 may be configured to execute an optimization loop 512. Candidate implant parameters 514 may be determined initially based on initial implant parameters 516. A constraint cost function 518 (or a constrained reward function) may use the computed candidate implant parameters to compute a loss. If stopping criteria are not met, candidate implant parameters 520 may be adapted and a further iteration of the optimization loop is executed. The optimization loop 512 may be executed until the stopping criteria are met. The cost function is based on clinical constraints 522 and may incorporate weights and / or optimization hyperparameters 524. When the stopping criteria are met, the determined implant parameters are output and used to determine a tooth implant model 526 and an implant sleeve model 528.

[0101] The constraint-based optimization may involve optimizing an objective function while satisfying specific restrictions on the model variables. The objective function can represent a cost or energy function to be minimized or a reward or utility function to be maximized. Constraints can be classified as hard constraints, which impose mandatory conditions on the variables, and soft constraints, which permit some violations but apply penalties in the objective function based on the degree of non-compliance.

[0102] Elements of the optimization problem may be defined as model variables: (e.g. implant position, orientation, length, and diameter) and cost function (e.g. clinical rules based on distances and angles between the implant and critical structures of the virtual patient).This way, constraint-based optimization may integrate clinical rules into the cost function to effectively guide a placement process. The constraints can ensure that the implant is positioned in accordance with best practices, while accounting for patient-specific anatomical variations.

[0103] According to some embodiments, the clinical requirements may be mapped to constraint functions that can operate on the implant mesh and the meshes of the structures composing the virtual patient. In an embodiment, the mapping may be a one-to-one mapping in which one clinical requirement is implemented in a constraint function.

[0104] A generic constraint function may be configured to take an implant model and the segmented 3D structures of the 3D virtual patient data as input data, compute a clinical measurement based on the input data, and return a cost depending on any violation by the clinical measurement of a constraint function. The final cost minimized during optimization is the weighted average of the costs returned by the constraint functions.

[0105] For a given tooth implant and a set of segmented 3D structures, e.g. mesh representations, of the 3D virtual patient data, the cost of a generic constraint function / can be defined as:

[0106] constraint cost[ = constraint function^implant, virtual patient data)

[0107] These constraint costs can have various dependencies on the respective constraint. The constraint function may link one or several constraints and the associated costs with a linear relationship. However, the constraint function can also link one or several constraints and the associated costs in an arbitrary way, e.g. cube, root, exponential, etc. The single constraints costSj can be linked in various ways. According to one embodiment, a constrained-based cost function associated with N clinical constraints can be described as:

[0108] cost function = weighti ' constraint cosf

[0109]

[0110] with weighti being a weight associated with the / -th constraint cost. The weigh can be used to rate single constraints differently in the cost function. However, if all constrain costs should be rated equally, the weights can also be set to one for all constraint costs.

[0111] The constraints can be categorized in different importance categories, which defines the importance of meeting a respective constraint. The importance category can have an influence on the constraint functions and / or the weights. In an embodiment, at least two importance categories may be used: a high importance category and a low importancecategory, where the constraints of the high importance category have to be met and the constraints of the low importance category may be violated. The constraints of the high importance category can also be regarded as hard constraints and the constraints of the low importance category as soft constraints. In further embodiment, between the high importance category and the low importance category one or more intermediate importance categories may be used.

[0112] In general, soft constraints can be regarded as constraints whose violation does not lead to unacceptable or poor implant placement. Therefore, its violation can be penalized with a finite cost, following a linear or a non-linear, e.g. exponential, penalization function. Examples of soft constraints may include: the distance between the implant and the nasal passage and / or its orientation relative to the crown. In contrast, a hard constraint is a constraint whose violation poses critical risks to the patient and the success of the implant placement. The importance of such a constraint can be expressed in the cost function by assigning an infinite cost (or costs that are considerably higher than other costs) whenever the constraint is violated. Examples of hard constraints may include: the distance between the implant and the inferior alveolar nerve (IAN) and / or a condition that the implant is embedded, preferably fully embedded, in the jawbone.

[0113] One or more of the following (non-limiting) embodiments regarding constraints and its influence on the cost function may be used in the optimization process:

[0114] - a constraint defining a distance between the implant and the inferior alveolar nerve (IAN): in an embodiment, if the shortest distance between the implant vertices of the implant mesh and the IAN vertices of the IAN mesh is below a predetermined threshold (e.g. as required by clinical guidelines), an infinite cost can be assigned to this constraint. In another embodiment, a violation cost proportional to the distance can be assigned, with the cost decreasing as the implant moves further from the nerve.

[0115] - a constraint related to an amount of bone around the implant body: in an embodiment, the total amount of bone around an implant may be computed. In other embodiments, the implant may be divided into regions. In an embodiment, the implant may be divided into regions, e.g. three regions: platform, middle, and apex, each with a length equal to one-third of the total implant length. For each region, the minimum distance between implant vertices of the implant mesh and jawbone vertices of the jawbone mesh can be calculated. These minimum distances may be compared to clinically required thresholds for each region and a cost may be assigned to each region that penalizes (linear or non-linear, e.g. exponentially) any violations.a constraint regarding implant immersion in the jawbone: for stability of the implant in the jawbone, the implant needs to be immersed in the jawbone to a certain extent. For the cost function, a ratio of implant vertices of the implant mesh inside the jawbone mesh and the length of the implant can be determined. In an embodiment, if this ratio is below the minimum required percentage, an infinite cost can be assigned. In another embodiment, a cost proportional to this ratio can be applied.

[0116] - a constraint related to a distance from neighboring teeth or pre-existing implants: a shortest distance between implant vertices of the implant and vertices of one or more neighboring structures can be determined. In an embodiment, if this minimum distance is below a predetermined threshold, an infinite cost can be assigned. In an embodiment, the cost may decrease as the distance increases. Examples of neighboring structures may include teeth and / or pre-existing implants.

[0117] - a constraint related to a distance from the jawbone ridge: a distance between the center of the implant platform and the jaw ridge along the occlusal direction can be determined, based on the implant mesh, the jaw mesh, and the wax-up occlusal direction. In an embodiment, if this distance is outside a predetermined range, a cost proportional to the deviation from the predetermined range can be assigned. In another embodiment, if the distance exceeds a predetermined threshold value, an infinite cost may be assigned. - a constraint related to a distance from the emergence line (EL) point: a minimum distance between the implant platform coordinates and the EL point along the occlusal direction can be determined. In an embodiment, if this distance is within a predetermined (acceptable) range, a negative cost can be applied. In another embodiment, if the distance is outside the predetermined range, a non-linear penalty can be applied. For example, in an embodiment, the penalty may increase exponentially as the distance deviates from the predetermined (acceptable) range. Here, the EL point may be defined as the centroid of the surface described by the EL line.

[0118] - a constraint related to an offset from the target platform position: given the implant mesh, the desired implant platform coordinates of the implant mesh, and the wax-up occlusal and buccal directions, the offset on the plane defined by the occlusal and buccal vectors, between the implant platform centroid and the desired platform coordinates can be determined. In an embodiment, if this offset exceeds a predetermined clinical threshold, an infinite cost can beassigned. In an embodiment, if the offset is within a predetermined allowable range, a cost proportional to the offset, e.g. the cube of the offset, can be assigned.

[0119] - a constraint related to a divergence from the desired orientation: given the implant mesh and a desired orientation, the angle between the implant direction and the desired orientation can be determined. In an embodiment, if this angle exceeds a predetermined threshold, an infinite cost can be assigned. In another embodiment, a cost proportional to the angle e.g. the cube of angel may be assigned. The desired implant orientation may be determined based on the wax-up orientation, and, optionally, the orientation of neighboring elements such as teeth or pre-existing implants.

[0120] - a constraint related to a distance from predetermined anatomic structures, including at least one of a nasal passage, maxillary sinuses, and an incisive canal: given the implant mesh and the meshes of the nasal passage, maxillary sinuses, and incisive canal (each of them can be regarded as anatomical structure), the shortest distance between implant vertices of the implant mesh and the vertices of one or more of the anatomical structures can be determined. In an embodiment, if the minimum distance is below a predetermined clinical threshold, an infinite cost can be assigned. In another embodiment, a cost proportional to the distance can be applied.

[0121] - a constraint related to an implant surface and stability: the surface area of the implant model, e.g. implant mesh, can be determined based on its diameter and length. To maximize this area, a negative cost proportional to the surface area can be applied. In an embodiment, an additional weight can be given to the length of the implant.

[0122] During optimization a set of parameters 524 may be used to define an optimization path that converges the optimization loop 512 and moves towards an optimal implant placement and, thus, an optimal set of implant parameters. These parameters can be regarded as hyperparameters of the algorithm, if they have been optimized and empirically chosen through meta-optimization, driven by quality metrics on a validation dataset. The hyperparameters may be grouped into two categories: weight parameters and optimization loop parameters.

[0123] Weight parameters (as already used in the cost function above) refer to the set of weights used to balance the importance of the different constraint functions when combined as a weighted average in the final total cost. Weights tune the relevance of each clinical constraint during optimization and finding the optimal implant parameter set. Themagnitude of each weight may reflect the clinical importance of the corresponding constraint. For example, the amount of bone around the implant body, which is essential for implant stability, can be weighted more heavily than a divergence in implant orientation. These weights are relevant for soft constraints, or for hard constraints when the implant is placed in a region that does not violate the hard constraints. Otherwise, a violation of hard constraints results in an infinite cost, effectively nullifying the impact of the weight parameters.

[0124] Optimization loop parameters can be specific to the optimization algorithm in use. These include parameters that are know from general optimization schemes, such as annealing. Examples of such parameters include the number of optimization steps, the number of warm-up iterations, the maximum allowed temperature, and / or the type of cooling scheme of the optimization scheme. These parameters govern the convergence process and the speed at which the algorithm converges, aiming to balance exploration and exploitation on the path towards the global optimum.

[0125] Once the constraint cost functions are implemented, any constraint-based optimization method which allows for non-linear cost functions can be used. A general approach of the optimization loop 512 is illustrated in Fig. 6. This figure depicts a decision diagram that may be used in a method for tooth implant planning in a dento-maxillofacial complex as described with reference to Fig. 5.

[0126] The process may start with determining an initial state of the implant parameters (steps 602,604). A suitable initial state can support or even ensure successful convergence of the optimization process. In an embodiment, the initial state may be determined by randomly placing a reference implant within the area of interest. In another embodiment, the initial state can be determined based on a set of basic rules and / or heuristics that take the crown design and other anatomical structures in the vicinity of the area of interest (jawbone, gingiva, and / or neighboring elements) into account. This approach may provide a well-informed initial implant pose (i.e. position and orientation) and dimensions, which may or may not fully satisfy all clinical constraints but serve as the best possible estimate before starting optimization.

[0127] In an embodiment, determining an initial state of the implant parameters may include the steps of:

[0128] - selecting the smallest possible implant for the target FDI tooth position from a catalog of allowed implant dimensions;

[0129] - aligning the orientation of the selected implant with the average orientation of neighboring pre-existing implants, teeth, and / or the crown design of the 3D virtual patient data. Here, neighboring elements may be defined as the3D anatomical structures that are located within a certain proximity radius, typically including the adjacent FDI number;

[0130] - positioning (translate / rotate) the implant so that the centroid of its platform aligns with the intersection point between the jawbone and the projected wax-up drilling point along the desired implant occlusal direction. This positioning may ensure that the implant is initially aligned with the planned wax-up.

[0131] This initialization provides a clinically reasonable starting position, improving optimization efficiency and the likelihood of achieving an optimal implant placement.

[0132] Based on the initial state, the costs (or rewards) may be calculated (step 608) based on the constrained-based cost function and evaluated (step 608). Evaluation may include comparing costs of the current state with costs of an earlier state. If the current state is acceptable (“yes”), the current state will become the new state (step 610) which then serves as a starting point for a next optimization cycle. If the current state is not accepted (“no”), the current state will be discarded and the previously “new state” will be used for the next optimization cycle. The process may be continued in step 612 in which stopping criteria are evaluated to determine if the optimization loop can be terminated or not. If the optimization loop can be terminated (“yes”), the process will terminate in step 614. The last valid state can be output as implant parameters. If the optimization loop is not terminated (“no”), the process continues with step 616, where the current state or a state of a previous optimization cycle will be adapted to become the new current state. Then, a next optimization cycle may start wherein the new state is input to step 606 for calculating costs.

[0133] The optimization, particularly the way of adapting the states, can be based on various optimization schemes. According to one embodiment, the optimization may be based on simulated annealing (SA). SA is a probabilistic optimization technique inspired by the physical process of annealing, where a material is heated and then gradually cooled to reach a state of minimal energy. In implant planning, SA is used to determine the optimal implant placement by minimizing a cost function that quantifies how well the implant satisfies various clinical constraints. The clinical constraints and their implementation have been discussed above. The optimization process begins with an initial implant placement, e.g. based on heuristics applied to the given crown, neighboring elements, jawbone, and gingiva. SA proceeds iteratively, where at each step, a small modification is made to the implant’s position, orientation, length, or diameter. The new configuration is then evaluated using the cost function. If the new placement results in a lower cost (better solution), it is accepted as the current configuration. However, if the new placement is worse, it may still be acceptedwith a certain probability, known as the Metropolis acceptance probability, which is a key feature of SA. The probability P of accepting a worse solution is given by:

[0134] P = e~*E / T

[0135] where Tl^is the increase in cost between the new and old solution and T is the temperature, which decreases over time according to the cooling schedule. An embodiment of simulated annealing may be described by the following pseudo-code:

[0136] 1. generate initial state

[0137] 2. repeat

[0138] a. compute neighbor to current state as new state b. compute cost ccof the current state

[0139] c. compute cost cnof the new state

[0140] d. i

[0141]

[0142] f cn< ccor e(c"-Cc) / T> Rcmd(0,l) then

[0143] current state := new state

[0144] e. decrease temperature T

[0145] until maximum number of iterations is reached.

[0146] Accepting worse solutions allows the algorithm to explore a broader range of potential implant placements, particularly in the early stages of optimization. This helps the algorithm escape local minima, suboptimal solutions that could trap other optimization methods, and explore more promising configurations that may lead to a better global solution. As the temperature decreases, the probability of accepting worse solutions diminishes, making the algorithm increasingly focused on fine-tuning the placement as it nears convergence. The temperature in SA can decrease gradually over time according to a cooling schedule, which governs the balance between exploration and exploitation.

[0147] The optimization can continue until one or several stopping criteria are met, such as reaching a predefined number of iterations or when the change in cost falls below a certain threshold. The final implant parameters are the ones with the lowest cost found during the optimization process.

[0148] In another embodiment, the optimization scheme may be based on Particle Swarm Optimization (PSO). PSO can model implant position, orientation, diameter, and length as particles in a swarm. Each particle adjusts its parameters based on both its own best placement and the best placement found by others, enabling a balance betweenexploration and exploitation. This method is particularly effective in navigating complex search spaces without requiring gradient information.

[0149] In yet another embodiment, the optimization scheme can be based on Genetic Algorithms (GA). GAs take an evolutionary approach, encoding implant parameters as chromosomes and iteratively refining solutions through selection, crossover, and mutation. By simulating natural evolution, GA can effectively explore diverse solutions and escape local minima, making it useful for highly nonlinear and multi-modal cost functions.

[0150] In a further embodiment, the optimization scheme can be based on Gradient Descent (GD). GD is suitable, when a differentiable cost function is available. GD optimizes implant parameters by iteratively adjusting them in the direction of the steepest cost reduction. While GD is highly efficient for smooth and convex problems, it can struggle with local minima and non-differentiable constraints, making it less suitable for complex anatomical constraints.

[0151] During the optimization loop 512, an optimization state can be defined by a similarity transformation matrix that applies translation, rotation, and scaling to adjust the initial pose (position, orientation) and dimensions of the implant to an optimized pose and dimensions. This transformation may be applied to the implant model, which may be represented in any suitable format, e.g. a mesh representation or a point cloud representation. The similarity transformation thus not only repositions the implant but also modifies its dimensions (e.g. diameter and length) to ensure proper adaptation to the clinical constraints. A generic transformation matrix for 3D similarity transformations that includes scaling can be represented as follows:

[0152]

[0153] = if :i

[0154] where sR represents the combined scaling and rotation applied to the initial implant and t is the translation vector that repositions the initial implant in 3D space.

[0155] A state transition may be responsible for generating a next candidate state in the parameter search space during the optimization phase. To create a new transformation matrix, rotation, translation, and scaling parameters can be sampled from a uniform distribution within predefined boundary values. The sampled transformation can then be applied to the current implant state, updating the implant's pose and dimensions, to determine a new candidate implant state. During simulated annealing (SA), boundary values for translation, rotation, and scaling can be scaled by a temperature at each iteration. As the temperature decreases over time, the boundary values also reduce, narrowing the search area and focusing the optimization on exploring solutions closer to the optimum.In some cases, when a designed crown has an inaccurate shape or pose, or the patient’s anatomy presents challenges, such as significant bone regression, it may be impossible to generate an initial implant state that satisfies all hard constraints. In such situations, finding a better starting position for the implant before beginning the actual optimization loop may be important for the success of the optimization process. To address this, a warmup procedure can be used that is designed to identify an initial implant state satisfying all hard clinical constraints. The warmup procedure can be a simplified version of the Simulated Annealing optimization loop, using the same clinical constraints. However, instead of optimizing for the best placement, the convergence condition or escape criterion is simply finding a state that does not result in an infinite cost, that is, a state that meets all hard constraints. Once such a state is found, it is used as the initial state for the actual optimization process.

[0156] Convergence criteria may be used, as stopping criteria, to determine when to terminate the optimization process. The optimization can be stopped when the maximum number of iterations is reached or when a state persists for more than a predefined number of consecutive iterations without any improvement or without any significant improvement (early stopping). The values for the maximum number of iterations and the threshold for early stopping can be chosen empirically by evaluating the algorithm’s performance on a set of validation cases.

[0157] Although Simulated Annealing (SA) is designed to escape local minima and has a high likelihood of reaching the global minimum given enough iterations, there is no guarantee that it will do so by the end of the optimization process, given that the optimization cannot take an infinite time. Moreover, implant placement is inherently a problem with multiple clinically relevant solutions, meaning several good local minima may exist alongside a global minimum. As a result, convergence does not necessarily ensure finding the absolute best solution but does guarantee a clinically valid placement, provided that all optimization constraints are satisfied. For addressing this, the optimization process can use a random number generator initialized with a fixed seed to ensure reproducibility of results. By changing the random seed, the optimization can explore different local minima, effectively generating multiple (sub)optimal solutions for the same case. In this way, multiple proposals can be generated and either output to the user or addressed for further automated assessment for finding the best solution.

[0158] At the end of an optimization process, the implant state with the lowest cost can be selected as the final implant state. Based on this state, the algorithm can output the predicted implant as implant parameters. In an embodiment, implant parameters can be described by a position (e.g. by a vector referring to a reference spot of the implant), a direction (e.g. the direction of a central axis of the tooth implant), and dimensions (e.g. lengthand diameter of the tooth implant). According to another embodiment, the implant parameters can be output as a similarity transformation matrix containing scaling, translation, and rotation parameters to transform a default tooth implant (initially at the origin with unit dimensions) into its final pose and size. In a further embodiment, the implant parameters can be output as a set of 3D points representing the implant in its final pose and optimized dimensions. In yet a further embodiment, the implant parameters can be output as a 3D model, e.g. a 3D mesh or a point cloud, of the tooth implant as a cylinder, already positioned and sized according to the optimization results. In a further embodiment, the implant parameters can be output as a voxelized binary volume representing the implant’s segmentation in its final placement and dimensions. Similar, output formats may be used when outputting the generated sleeve, ensuring it is properly positioned and aligned with the optimized tooth implant placement.

[0159] Fig. 7A to 7D show possible steps in an optimization scheme as described with reference to Fig. 5. In each part of Fig. 7, a wax-up 702 with a wax-up direction 704 is placed above a jawbone 706. IAN 708 is located in vicinity of the areas of interests in which the implant needs to be placed. Fig. 7A shows a placement of the implant 710 in an initial state. In this situation, the implant is too close to the bone edge (indicated by the dashed circle) which leads to infinite costs. At the first iteration step shown in Fig. 7B, the position, orientation and dimension of the implant 710 may be adapted. In this situation, another corner of the implant 710 is close to the bone edge (indicated by the dashed circle) which leads to high costs. At second iteration step shown in Fig. 7C, the position, orientation and dimension of the implant 710 may be adapted again. In this situation, the implant is too close to the IAN so that the distance requirement is violated and the cost value assumes an infinite value. The third iteration step shown in Fig. 7D leads to further change in the pose and / or dimensions of the implant 710. This solution may have a a low cost, or even minimum costs. The implant 710 is now well centered with the wax-up direction and with acceptable distance from the jaw ridge and IAN. This final state gives the lowest cost and has the largest possible surface area.

[0160] In a further embodiment, implant planning engine of Fig. 5 can be used to determine multiple implants. In an embodiment, planning multiple implants may include sequential planning wherein each implant is computed one at a time. In that case, a placement of one or more previous implant(s) may be taken into account as a constraint when planning one or more subsequent implants. This can be achieved by including the one or more previously computed implants as "existing implants" within the 3D virtual patient data.

[0161] In an embodiment, the order of the planning of the multiple implants can be determined based on clinical priority or anatomical location. For example, in an embodiment,posterior implants may be prioritized before anterior implants. In another embodiment, the order may be based on the availability of one or more neighboring teeth. This method is computationally simple and allows for flexibility in adjusting individual implant placements based on the overall treatment plan's progress.

[0162] In another embodiment, planning multiple implants may include one-shot planning which aims to determine the optimal placement for all implants simultaneously. This approach requires a cost function that takes into account the spatial relationships and interactions between all implants. Optimization algorithms such as Particle Swarm Optimization (PSO) or Genetic Algorithms (GA) are well-suited for this type of problem due to their ability to handle multiple interacting variables. One-shot planning can potentially lead to more globally optimal solutions but is computationally more demanding.

[0163] The implant planning engine 510 can additionally allow for user adjustments and for human feedback incorporation. To that end, the system of Fig. 1 may include a graphic user interface GUI (not shown) which is configured display the 3D virtual patent data, the implant model and the implant parameters on a display. The GUI may be configured to allow a user to modify one or more of the implant parameters and to re-execute the optimization process based on the modified implant parameters.

[0164] Hence, the system allows a clinician who review and evaluate the automatically generated plans for parameter adjustments of the optimization scheme. In an embodiment, the GUI may also be configured to allow a user to adjust the weights assigned to different constraints within the cost function. If a clinician frequently finds that the automated plan compromises a particular clinical requirement (e.g., bone support), the weight for the corresponding constraint can be increased. This way, the system allows customer-tailored hyperparameter tuning allows individual users or clinics to fine-tune the system to their specific preferences and workflows. This could involve offering different preset configurations or allowing users to experiment with various parameter settings and observe their effects on the generated implant plans. Then, given the user’s preferences a specific customer-tailored optimization model can be offered.

[0165] Fig. 8 illustrates a method for automated planning of a tooth implant in a dento-maxillofacial complex according to an embodiment of the present disclosure. In step 802, 3D virtual patient data are received or determined, where the 3D virtual patient data represent a dento-maxillofacial complex comprising volumetric representations, e.g. mesh representations, of anatomical structures and at least one area of interest for placing the tooth implant. In step 804, an initial implant model or an initial set of implant parameters defining an initial implant model may be received or determine. In step 806, costs of the initial implant model or the initial set of implant parameters defining an initial implant model may be determined based on a cost function, wherein the cost function computes the costs based onclinical constraints and. The clinical constraints may include at least one distance requirement between the tooth implant model and at least part of the 3D representations of the anatomical structures. In step 808, an optimized implant model or a set of implant parameters defining an optimized implant model may be determined by adapting the initial implant model or the set of implant parameters defining an implant model until the cost determined by the cost function is minimized and / or a stopping criteria is met.

[0166] While the schemes described above are described with reference to determining tooth implant parameters (and associated models), these schemes can also be used to determine parameters and a model of an implant sleeve associated with a tooth implant model.

[0167] The implant planning engine used in the system for automated planning of a tooth implant in a dento-maxillofacial complex as described with reference to the embodiments in this disclosure is not limited to optimization schemes as described with reference to Fig. 5-8. In further embodiments, the implant planning engine may use one or more trained deep neural networks for determining tooth implant parameters and a tooth implant model.

[0168] Fig. 9 illustrates an example workflow for automated planning of a tooth implant using an implant planning engine 930 comprising a predictive model 912 according to an embodiment of the present disclosure. In one embodiment, the predictive model leverages deep learning algorithms, which are trained to optimize an implant model and / or one implant parameters defining a tooth implant moel for a tooth implant treatment plan.

[0169] The training data for the predictive model may comprise a versatile range of implant planning cases performed by clinical experts to ensure adaptability and generalization across diverse clinical scenarios. The implant planning cases used as input data for training the predictive model may include single and / or multiple implants, varying in size, orientation and position, and may cover contiguous FDI positions, distant FDI positions, and fully edentulous cases. The training input data for the predictive model may comprise 3D virtual patient data of the patient's dento-maxillofacial complex. This 3D virtual patient data may be generated based on 2D and 3D imaging modalities, including CBCT, IOS, panoramic radiographs, and extraoral scans, as described in further details with reference to Fig. 4.

[0170] The 3D virtual patient data 902 may comprise volumetric representations, e.g. mesh representations, of anatomical structures, such as teeth, the upper and lower jaw, maxillary sinuses, and other relevant structures within the vicinity of the area of interest. These representations provide the anatomical context for the predictive model, enabling precise implant planning while ensuring alignment with clinical constraints, and optionally, with prosthetic and / or aesthetic considerations.Additionally, the training data comprises ground truth implants, which capture implant planning cases performed by clinical experts with varying levels of expertise. These ground truth implants serve as reference data for training the predictive model and may be represented in various formats, including meshes, transformation matrices, voxelized volumes, point clouds, and / or implant parameters. The implant parameters may describe a position of the tooth implant within the area of interest and / or an orientation of the tooth implant and / or dimensions of the tooth implant and may be derived from any of the aforementioned representations. Conversely, the other implant representations, including meshes, transformation matrices, voxelized volumes, and point clouds, may also be obtained from the implant parameters.

[0171] In one embodiment, the input data may also comprise one or more crown designs 904 for one or more areas of interest 906, particularly in the case of prosthetic-driven implant planning. The one or more crown designs may provide additional prosthetic constraints, ensuring that the automated implant planning aligns with functional and aesthetic requirements. The one or more crown designs may be represented as 3D meshes, point clouds, or other types of volumetric data, capturing their shape, occlusal surface, and positional relationship with adjacent and opposing teeth within the 3D virtual patient data.

[0172] The selection of the area of interest 906 may be based on clinical evaluation, missing teeth identification based on the identified anatomical structures in the 3D virtual patient data, prosthetic requirements, or patient-specific treatment plans, ensuring that implant planning corresponds to the intended restoration and occlusal function.

[0173] In some embodiments, clinical constraints 908 may be provided as additional input to the predictive model to further refine the predicted implant model and / or parameters defining an implant model. In an embodiment, the clinical constraints may be directly embedded into the predictive model 912 by encoding them into lower-dimensional feature representations. These encoded features may be concatenated into the predictive model at various hierarchical levels, allowing the predictive model to incorporate domain-specific knowledge throughout the training process. This encoding may be performed during the preprocessing step 910, as illustrated in Fig. 9.

[0174] The training data may be preprocessed 910 before being provided as input to the predictive model 912. The preprocessing 910 may include data standardization and normalisation, ensuring consistency across different formats of the input data, and conversion of volumetric data, e.g. meshes, of the 3D virtual patient data into a representation compatible with the predictive model 912, such as graphs, or point clouds. Additionally, feature extraction may be performed as a preprocessing step 910 to derive geometric properties, such as distances, angles, and / or spatial relationships, from the input meshes of the 3D virtual patient data. The preprocessing 910 may also involve encoding the area of interest 906 and / or clinical constraints 908 into structured representations that arecompatible with the predictive model architecture. This allows the predictive model 921 to incorporate domain-specific knowledge as additional inputs.

[0175] The preprocessed input data is fed into the predictive model 912, which may employ one or more deep learning architectures designed for automated implant planning according to the embodiments described in the present disclosure.

[0176] In an embodiment, the predictive model 912 may employ a graph-based model. The graph-based model may be configured to represent 3D meshes corresponding to anatomical structures within the 3D virtual patient data 902 as a graph, where nodes correspond to anatomical landmarks or vertices of the mesh representation, and edges may define the geometric and spatial relationships between the nodes. This approach allows the predictive model 912 to process the topological structure of the 3D virtual patient data 902 by integrating local and global geometric features relevant for the automated implant planning. In an embodiment, a graph-based model may be used wherein, the nodes may encode anatomical information. The graph-based model may include mesh vertex coordinates and vertex normals, anatomical labels indicating the anatomical structure to which the node belongs, dihedral angles between a vertex and its neighbouring vertices, local curvature of the mesh, and bone density information derived from the CBCT scan at each vertex location.

[0177] The meshes of the single anatomical structures may initially be converted to sub-graphs. Once the anatomical structures are represented as sub-graphs, the sub-graphs may be integrated into a single unified graph through inter-mesh edges. These inter-mesh edges may encode geometric and spatial relationships between different anatomical structures of the 3D virtual patient data, facilitating information propagation across the graph.

[0178] In an embodiment, the model architecture may be based on a Graph Neural Network (GNN), in particular, a Graph Attention Network (GAT). GAT networks are known from the article by Velickovic, P., etal, (2018). Graph Attention Networks.,

[0179] arXiv: 1710.10903., These network leverage attention mechanisms to assign varying importance levels to neighbouring nodes, which will enhance feature learning.

[0180] In an embodiment, the model can integrate expert-defined clinical constraints as graph constraints, ensuring that the learned representations remain aligned with domainspecific knowledge. In an embodiment, graph constraints may be embedded as explicit constraints on nodes, edges, and their associated features. These constraints may be incorporated into the graph representation in multiple ways: edge constraints enforce anatomically valid connections between structures, feature constraints regulate the permissible range of node attributes, and positional constraints may ensure spatial coherence among landmarks.

[0181] In a further embodiment, clinical constraints, including the area of interest, may be encoded using Multilayer Perceptron (MLP) layers and may be incorporated into the global graph as additional nodes. This structured integration ensures that both anatomicalconstraints and user-specified preferences are captured within the model, improving the accuracy and adaptability of implant parameters predictions for tooth implant planning.

[0182] In an embodiment, the predictive model architecture may be configured to process point cloud data. Hence, in case the volumetric presentations of the 3D virtual patient data 902 are meshes, the meshes need to be preprocessed and converted to point cloud representations, which consist of an unordered set of three-dimensional (3D) points capturing the patient's anatomical structures within the 3D virtual patient data, respectively. This representation is particularly suited for processing 3D geometries, including the mesh representation of the 3D virtual patient data and the designed dental crown. Predictive model architectures for processing point could can be implemented using PointNet++ or PoinTr. PointNet++ is described in Qi, C. R., Yi, L., Su, H., & Guibas, L. J. (2017). PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space. arXiv:1706.02413. Further details of PoinTr can be obtained from Yu, L., Zhou, Q., Han, Z., & Zheng, N. (2021). PoinTr: Diverse Point Cloud Completion with Geometry-Aware Transformers., arXiv:2108.08839). PointNet++ employs a hierarchical feature learning approach, utilizing a series of set abstraction layers to extract both local and global geometric features from the point cloud, while PoinTr employs a transformer-based architecture that captures long-range dependencies across the point cloud. PoinTr's attention mechanisms may prioritize specific regions within the point cloud, facilitating the extraction of spatial relationships and enhancing the processing of complex geometric structures.

[0183] Before being processed by the predictive model configured to handle point cloud data, the meshes representing the 3D virtual patient data - along with, optionally, the designed crown and the area / s of interest - may undergo a preprocessing step 910. Rather than converting these meshes into graph representations, the predictive model may directly process the point cloud data obtained by sampling points from the surface of each mesh of the 3D virtual patient data. This conversion may involve extracting a set of 3D points from each input mesh using techniques such as uniform sampling, Poisson disk sampling, or curvature-based sampling, ensuring an even distribution of points while preserving geometric details. This way, each anatomical structure, including the patient’s jaw and the optional crown design, is represented as a set of 3D points sampled from its surface. These point clouds may then be processed by predictive models 912 configured to process point cloud data, such as PointNet++, which extracts geometric features using set abstraction layers, or PoinTr, which applies transformer-based attention layers to capture both local surface details and global geometric relationships.

[0184] In further embodiment, one or more clinical constraints 908, and / or areas of interest 906 may be encoded as feature vectors and integrated into the predictive model 912. These feature vectors can be incorporated at different processing stages within the predictive model 912 to influence implant parameters prediction. In PoinTr, the clinical constraints maybe embedded directly within the transformer attention mechanism, enabling the model to dynamically adjust focus based on the provided clinical constraints.

[0185] During training, the predictive model 912 learns to output implant models and / or one or more parameters defining an implant model 920. In case, predictive model 912 does not output implant parameters 920 according to the present disclosure directly, an optional conversion step (not shown in Fig. 9) may be required to transform an implant representations that is output by the predictive model into implant parameters. Such a conversion step may be performed directly at the output of predictive model 912 or at any suitable step before outputting the implant parameters 912. The predictive model 912 processes the input data, which comprises the 3D virtual patient data 902 (optionally preprocessed 910), and which may optionally comprise processed prosthetic crown design data 904, the area of interest 906, and clinical constraints 908. The predictive model 912 may be configured to generate implant predictions 914, e.g. an implant model and / or one o more implant parameters, in various formats, similar to the ground truth data. Formats for implant models may include meshes, transformation matrices, voxelized volumes, or point clouds. The implant representation formats are interconvertible, providing flexibility for different downstream tasks in the implant planning process.

[0186] In an embodiment, in a direct regression of a point cloud, the predictive model 912 may output the implant position as a set of 3D points, representing the implant as a point cloud within the 3D virtual patient data 902. A mesh can be reconstructed from the implant point cloud using surface reconstruction techniques such as Poisson surface reconstruction or alpha shapes, while a transformation matrix can be derived by aligning the predicted point cloud with a reference implant shape. Implant parameters such as implant’s position, orientation, and / or dimensions may be derived from the generated mesh.

[0187] In an embodiment, the predictive model 912 may be configured to regress a transformation matrix, the predictive model 912 directly predicting a similarity transformation matrix, which may encode the position (translation), orientation (rotation), and scale of the implant. By applying this transformation matrix to a canonical implant model, that may be derived from commercially available tooth implants, a mesh representation may be generated within the 3D virtual patient data coordinate system. If needed, a point cloud can be sampled from the transformed mesh.

[0188] In another embodiment, the predictive model 912 may be configured to output a direct regression of a mesh representation of an implant, including vertices and faces, providing a ready-to-use implant geometry. From this mesh, a point cloud representation may be extracted, and if a reference implant shape is available, a transformation matrix aligning it to the predicted mesh may be computed. Additionally, implant parameters, such as position, orientation, and dimensions, can be derived directly from the generated mesh.Both point-based and graph-based predictive models 912 may be configured to output implant parameters 920 corresponding to any of these implant representations, making the predictive model 912 adaptable to different requirements in an implant planning pipeline.

[0189] The predictive model 912 may be trained in an iterative process to refine the implant model and / or implant parameters. In an embodiment, the predictive model 912 receives as input training data comprising sets of 3D virtual patient data 902, associated ground truth implant parameters, and optionally, clinical constraints 908.

[0190] At each training step, the predictive model 912 generates implant prediction 914. The implant prediction 914 is evaluated using a loss function 916, which penalizes deviations between the implant prediction 914 as output of the predictive model 912 and the ground truth implant parameters. Stopping criteria 918 determine whether the predictive model 912 has converged to an optimal solution. If the stopping criteria 918 are not met, an optimizer 922 updates the parameters of the predictive model 924 to minimize the loss function 916. Once the stopping criteria 918 are satisfied, an implant model and / or one or more implant parameters defining an implant model 920 are output. This iterative optimization process continuously refines the predictive model 912, minimizing errors relative to the ground truth implant parameters and ensuring the generation of clinically viable implant parameters.

[0191] Optionally, clinical constraints may be explicitly incorporated into the loss function 916 to enforce anatomically and prosthetically valid implant parameters 920 for an implant planning case. For example, anatomical safety constraints may penalize implant placements that encroach on critical structures such as nerves or sinuses by computing distances between the predicted implant position and relevant anatomical landmarks.

[0192] Similarly, prosthetic-driven alignment constraints may ensure proper alignment between the implant axis and the designed prosthetic crown, penalizing angular or translational deviations.

[0193] The loss function 916 may be formulated as a multi-term objective function, where different clinical constraints are weighted, e.g. according to their respective importance category:

[0194] L

[0195]

[0196] total \ ^placement T 2^ anatomical T 3^prosthetic

[0197] where Lplacementpenalizes deviations from the ground truth implant parameters, Lanatomicalpenalize implant predictions 914 that encroach on critical anatomical structures such as nerves and sinuses by computing distances between predicted implants 914 and relevant anatomical structure, and Lprostheticmay enforce alignment between the predicted implant 914, that may be used to derive an implant axis, and the designed prosthetic crown usingangular or translational deviation metrics. By incorporating clinical constraints into the loss function 916, the predictive model 912 may integrate domain-specific knowledge, and it may improve both predictive accuracy of the predictive model 912 and clinical applicability.

[0198] Fig. 10 depicts a system for predicting an implant model or implant parameters defining an implant model using a trained predictive model according to an embodiment. The system 1000 is configured to execute an inference process for generating implant parameters using a predictive model that may have been trained as described with reference to Fig. 9.

[0199] The system may include one or more data storage systems comprising input data, including 3D virtual patient models 1002 which may be associated with different patient. Each 3D virtual patient model may be associated with one or more prosthetic crown models 1004, one or more areas of interest 1006 for placing one or more target implants respectively, and clinical constraints 1008. The input data provided to the input of a preprocessing module 1010 which may be configured to preprocess input data in a similar as the preprocessing that was used during training of the predictive model. The preprocessing may include normalization and standardisation of the input data and feature extraction and encoding of extract features that may be used during the inference process.

[0200] The input data, which may be preprocessed by preprocessing module 1010, may be provided to one or more trained predictive models 1012i-n, wherein each of the one or more trained predictive models may be configured to predict candidate implant parameters 1014i-n. The output of a trained predictive model 1012 may represent a candidate implant in a certain implant representation format. In some embodiments, the candidate implant may be transformed into another format as explained with reference to Fig. 9. The candidate implant parameters 1014 generated by a trained predictive model may be assessed by a validity evaluation module 1016 based on clinical constraints 1016 which may be defined in the context of the 3D virtual patient model 1002 and, optionally, the prosthetic crown model 1004. If the assessment by the validity evaluation module confirms that the candidate implant parameters satisfy the clinical constraints, the candidate implant parameters may be output by the as validated implant parameters 1018.

[0201] In an embodiment, a multi-proposal approach may be used wherein multiple candidate implant parameters 1014i.nmay be predicted by multiple trained predictive models 1012i-n. In that case, candidate implant parameters may be filtered, ranked, and / or refined before selecting and validating implant parameters 1018 corresponding to a planned tooth implant.

[0202] The validity assessment of the predicted implant parameters at step 1016 may involve post-processing and feasibility filtering to verify that the predicted candidate implant parameters 1014 comply with one or more clinical constraints defined in the context of the3D virtual patient data 1002 and, optionally, considering the prosthetic crown design data 1004.

[0203] In the multi-proposal approach, rule-based filtering may be applied to discard implant parameters that violate clinical constraints that are required to be met, according to the clinical constraints category. For example, this way, it can be ensured that the implant maintains a minimum distance of at least 2 mm from a nerve structure. In an embodiment, a ranking approach may be employed, wherein a function is used to assess the quality of the predicted implant placement by prioritizing implant parameters that exhibit minimal angular deviation from the crown axis to ensure optimal prosthetic alignment or those that maximize bone engagement to enhance implant stability.

[0204] In a single prediction approach, these validation steps may be optional.

[0205] However, if performed and the validity assessment of the candidate implant parameters 1016 is not satisfied, a further optimization based on an optimization scheme, as e.g. described with reference to Fig. 5 and 6, may be used. In the multi-proposal approach, multiple prediction techniques may be applied to generate alternative implant parameter predictions as output of the predictive model, enabling the identification of a clinically valid implant.

[0206] In an embodiment, point-cloud and / or graph-based predictive models as further described in Fig. 9 may be used to generate multiple implant parameters predictions in a multi-proposal for the same planning case using various techniques.

[0207] In an embodiment, a Monte Carlo Dropout (MC Dropout) 1024 may be used to enable multiple inference runs with dropout activated, introducing randomness and producing diverse outputs. Bootstrapping methods may be used to achieve similar results by training multiple predictive models on different randomly sampled subsets of the input training data.

[0208] Alternatively, training predictive models with different initial random seeds may ensure variations in learned parameters, leading to distinct, related predictive models 1012, 1020 and their corresponding candidate implant parameters predictions 1014, 1022.

[0209] In another embodiment, Test-Time Data Augmentation (TTA) may apply transformations to the input data during inference, generating varied implant parameters predictions as output. Stochastic Weight Averaging (SWA) aggregates model weights from different training checkpoints 1012, 1020, effectively capturing different representations and producing multiple candidate implant parameters predictions 1014, 1022.

[0210] In this way, the one or more predictive models may be used to explore a range of clinically valid implant parameter predictions for the same implant planning case. The validation assessment in step 1016 may be applied to a single implant parameter prediction or to multiple implant parameter predictions. By applying these validation techniques, the implant parameters predictions can be systematically refined and ranked among multiple implant candidates, facilitating the selection of the most suitable implant parameters prediction for clinical application and output it as implant parameters 1018.Fig. 11 illustrates a system for generating an implant model according to another embodiment. As shown in the figure, the system may include a implant planning engine 1150 which comprises a generative model 1114,1128. As shown in the figure, the implant planning engine comprises a training module 1138 for training the generative model to predict an implant model and an inference module 1140 for predicting one or more implant models based on a 3D virtual patient model and one or more area of interests in a dentition of the 3D patient model that define.

[0211] Training module 1138 may be configured to receive training data in a manner similar to the predictive model described with reference to Fig. 9 and Fig. 10. The training data may comprise 3D virtual patient data defining 3D virtual patient models 1102 along with further input data, such as crowns models 1104, one or more areas of interests 1106, and clinical constraints 1108. The training data may be pre-processed 1110, which may include 3D data standardization to ensure consistency across different input formats and feature extraction.

[0212] In an embodiment, input data may be provided to a scene encoder model 1112, which is configured to encode 3D models, e.g. meshes or point clouds, of the 3D virtual patient models 1102 into a latent representation that serves as a structured input for generative model 1114. In an embodiment, the scene encoder 1112 may utilize deep neural network architectures which are configured to receive 3D models as input, e.g. such as PointNet++ neural network architectures for processing points clouds or MeshNet neural network architectures for processing meshes, and to encode these models into the latent space. The scene encoder allows extraction of hierarchical features from unordered point clouds or meshes to effectively capture complex 3D anatomical structures. In a further embodiment, the generative models 1114,1128 may be configured to directly receive at least part of a 3D virtual patient model 1102 as at its input.

[0213] In an embodiment, the generative models 1114,1128 may be based on a diffusion-based architecture in which an initial noisy prediction of an implant model and / or implant parameter is refined into an accurate implant model and / or implant parameter prediction. In an embodiment, the diffusion model may be based on deep neural architectures such as PointNet ++ or transformer-based models. Examples of such diffusion models are described in the articles by Liu, Zhen, et al. " Meshdiffusion: Score-based generative 3d mesh modeling." arXiv preprint arXiv:2303.08133 (2023) and Li, Muheng, et al. "Diffusion-sdf: Text-to-shape via voxelized diffusion ." Proceedings of the IEEE / CVF conference on computer vision and pattern recognition. 2023.

[0214] Based on the 3D virtual patient data 1102 or a latent representation of the 3D virtual patient data, the one or more areas of interest 1106, the clinical constraints 1108 and, in some embodiments, the crown model, the generative model 1114 may be trained topredict an implant model or implant parameters defining an implant model by learning to reconstruct the implant model or the implant parameters defining the implant model within an areas of interest of the 3D virtual patient model 1102. During training, the generative model 1114 may receive the ground truth implant models and implant parameters defining the implant models as part of the input training data, similar to the predictive models presented in Fig. 9.

[0215] The training module 1138 may be configured to train a generative model 1114 that is based on a diffusion model wherein an initial implant representation, such as a transformation matrix, a mesh, or other implant parameters, may be gradually corrupted by adding noise over a series of time steps. This forward diffusion process transforms the implant representation into a purely noisy state at time step T, following a predefined variance schedule. Loss calculation module 1116 may comprise a loss function for evaluating the diffusion model’s ability to predict the noise added at each step by comparing the predicted noise with the actual noise applied during the forward diffusion process and determines whether a stopping criterion 1118 have been met. Clinical constraints as described with reference to Fig. 5, may be incorporated into the loss function to train the diffusion model to predict anatomically and prosthetically valid implant models.

[0216] The training module 1138 may comprise an optimizer 1122 for updating the internal parameters of the diffusion model 1124 based on the loss, thereby iteratively refining the model’s ability to accurately predict the noise added at each timestep. In an embodiment, the stopping criterion 1118 may be met when the loss function stabilizes below a predefined threshold, indicating that the model has sufficiently learned to predict the added noise. In another embodiment, training the diffusion model may stop after a fixed number of iterations to prevent overfitting. Once training is completed, the trained diffusion model may be used by the inference module 1140 to iteratively remove noise and reconstruct the implant parameters by following the reverse diffusion process.

[0217] The inference module 1140 may use the trained generative model 1126 to generate an implant model prediction or implant parameters predictions defining an implant model for an implant planning case. Similar to the training process, the input data can be pre-processed (3D data standardization and / or feature extraction) by a preprocessing module 1110. In an embodiment, input data may be input to a scene encoder model 1126, which is configured to receive 3D models, e.g. point clouds or meshes, as input and to encode these models into the latent space (similar as described in connection with the training module). The inference module 1140 may start the inference process with sampling an initial noisy implant representation from a Gaussian distribution, which serves as the starting point for the reverse diffusion process. The reverse diffusion process iteratively removes noise, reconstructing plausible implant models while ensuring anatomical and clinical validity.To that end, candidate implant predictions 1130 may be evaluated using a validation module 1134 which is configured to determine its validity. A loss function may be used in the validity assessment of the predicted implant model to verify if the predicted candidate implant model complies with clinical constraints defined in the context of the 3D virtual patient model 1102 and, optionally, the prosthetic crown model 1104 in a similar way as explained with reference to Fig. 10.

[0218] In case the candidate implant model does not meet the constraints, latent sampling 1136 may be used to introduce at least one controlled variation into the latent space of the diffusion model 1128 or, in some embodiments, the scene encoder model 1126, allowing the diffusion model to explore further noise trajectories and to produce a further candidate implant model that is provided to the validation module for validation. This iterative process will continue until at least one candidate model meets one or more stopping criteria (e.g. all stopping criteria or mandatory stopping criteria, which typically means that the candidate model meets the clinical constraints, f the validation module determines that the candidate implant model complies with the clinical constraints, it may output the validated implant model and / or implant parameters 1132.

[0219] By leveraging diffusion models with scene encoders, the embodiment in Fig.

[0220] 11 enables the generation of precise, patient-specific implant models or implant parameters defining implant models, while allowing for multiple possible implant parameter predictions, ensuring flexibility in clinical decision-making. The iterative refinement of implant model predictions ensures that the final implant plan aligns with the patient’s anatomy and treatment requirements.

[0221] In another embodiment, the trained generative model for predicting an implant model or implant parameters defining an implant model that is used in the tooth implant planning schemes described in this disclosure may be based on an implicit function representation to predict tooth implant models. For example, in an embodiment, a distance function referred to as the Signed Distance Function (SDF)-based may be used. An example of an generative model that is based on an SDF function is DeepSDF as described in the article by Park, Jeong Joon, et al. "DeepSDF: Learning continuous signed distance functions for shape representation." Proceedings of the IEEE / CVF Conference on Computer Vision and Pattern Recognition (CVPR). 2019)). Hence, in this embodiment a generative model may be trained to directly predict the 3D surface of the implant using an implicit function representation, such as a signed distance function (SDF).

[0222] The training module 1138, may be configured to start training with receiving 3D virtual patient models 1102, along with optional inputs, such as prosthetic crown models 1104, areas of interest 1106, and clinical constraints 1108. This input data may be pre-processed by preprocessing module 1110, in a similar way as described with reference to Fig. 9 and 10.

[0223] In an embodiment, the SDF-based generative model 1114 (in short SDF-model) may have an auto-decoder architecture, which may operate directly on 3D virtual patient models 1102. In another embodiment, the SDF-model may form an auto-encoder architecture with the scene encoder model. In an embodiment, the auto-decoder and autoencoder may have a PointNet++, a DGCNN, a voxel / 3D CNN, or a graph neural network architecture. The training data may include both the 3D virtual patient models 1102 and corresponding ground truth implant models, which be represented as 3D surfaces.

[0224] During training of the SDF model, the training module 1138 may provide a set of 3D coordinates sampled from the 3D virtual patient model 1102 to the input of the SDF-model. The 3D coordinates may include points from the area of interest including surrounding anatomical structures and the corresponding ground truth implant model. For each 3D coordinate, the SDF model may be configured to predict a corresponding SDF value representing the distance to the nearest point on the implant surface. Ground truth SDF values are derived from the known implant geometry within the 3D virtual patient data, allowing the SDF model to learn an implicit representation of different implant models.

[0225] In case of an auto-decoder architecture, each implant shape may be associated with a latent code, which is optimized alongside the internal parameters (i.e. the weights) of the SDF-model to ensure accurate reconstruction of the implant surface. The auto-decoder may receive points sampled from a 3D virtual patient model and a latent code representing an implant shape at its input and compute an SDF value at its output. The final implant model may be reconstructed as the zero-level set of the predicted SDF values, which can then be converted into a mesh representation or to implant parameters. A latent shape code may serve as a compact representation of the geometry of an implant model, while the decoder model maps the latent shape code and 3D coordinates of sampled points to predict SDF values, which define the surface of the implant model by representing the distance of each coordinate to the nearest point on the surface.

[0226] Loss calculation module 1116 may be configured to evaluate the SDF’s model ability to predict an implant model and determine whether a stopping criterion 1118 has been met. Clinical constraints as described with reference to Fig. 5, may be incorporated into the loss function to train the decoder model to predict anatomically and prosthetically valid implant models.

[0227] The optimizer may be used to update the parameters of the SDF model. This way, an optimization process is executed to iteratively refines both the latent shape code and the internal parameters of the SDF-model by minimizing loss function 1116, which quantifiesthe difference between the predicted and the ground truth SDF values derived from surfaces of ground truth implant models.

[0228] During inference, inference module 1140 may use a trained decoder model 1128 to reconstruct the implant surface from a latent shape code. A candidate implant model associated with a latent code may be evaluated using validation module 1134. A loss function that takes clinical constraints into account may be used in the validity assessment of the predicted candidate implant model to verify if the predicted candidate implant model complies with the clinical constraints.

[0229] In case the candidate implant model does not meet the clinical constraints, latent sampling module 1136 may be used to introduce an updated latent code that is input to the decoder model 1128 allowing the decoder model to explore further shape representations and to produce a further candidate implant model associated with the updated latent code, which is provided to the validation module for validation. This iterative process will continue until a latent code is determined that defines a candidate implant model that meets the clinical constraints. If the validation module determines that the candidate implant model complies with the clinical constraints, it may output the validated implant model and / or implant parameters 1132.

[0230] This way, the latent code associated with a surface representing an implant model encoded in latent space may be optimized during inference allowing the decoder model to generate an SDF representation of a shape of an implant model that conforms to the patient’s anatomy. Clinical constraints may be incorporated as additional loss terms in the loss function to guide the optimization. Given a set of 3D coordinates sampled from the 3D virtual patient data, the trained decoder model predicts corresponding SDF values, which define the distance of each point from the implant surface. The final implant shape is reconstructed as the zero-level set of the predicted SDF values, and the corresponding implant parameters 1132 are derived from the predicted implant shape.

[0231] Fig. 12A illustrates an example of a training process of an implant planning engine comprising a predictive deep learning model or a generative deep learning model to automatically predict implant parameters, according to some embodiments of the present disclosure. In step 1202, training data are received. The training data may comprise sets of 3D virtual patient data, associated ground truth implant parameters, and, optionally, clinical constraints. The 3D virtual patient data represent a dento-maxillofacial complex comprising anatomical structures, e.g. 3D meshes or point clouds, and at least one area of interest for placing the tooth implant. In step 1204, one or more sets of the 3D virtual patient data are provided as inputs to the implant planning engine. The model processes this data and predicts estimated implant parameters as output. Step 1206 involves training the parameters of the implant planning engine by updating its parameters by minimizing a loss function untilone or more stopping criteria are met. The loss function penalizes deviations between the estimated implant parameters and the ground truth implant parameters. When clinical constraints are additionally provided, they can be used as input features for the model and / or can be incorporated into the loss function as additional constraints.

[0232] Fig. 12B illustrates an example of an inference process of an implant planning engine comprising a trained predictive deep learning model or a trained generative deep learning model to generate an implant model or implant parameters, according to some embodiments of the present disclosure.

[0233] In step 1208, input data are received. The input data may comprise 3D virtual patient data representing a dento-maxillofacial complex comprising anatomical structures, e.g. 3D meshes or point clouds, and at least one area of interest for placing the tooth implant. Optionally, the input data may further comprise a prosthetic crown model and / or clinical constraints. The input data may be pre-processed, for example by standardizing 3D data formats and / or extracting relevant features. In some embodiments, at least part of the 3D virtual patient data may be encoded into a latent representation, for example by a scene encoder, to provide a structured input for the implant planning engine.

[0234] In step 1210, the input data, or the latent representation thereof, are provided to the trained implant planning engine. The model processes this data and generates a candidate implant model and / or candidate implant parameters as output. In some embodiments, the candidate implant model and / or candidate implant parameters may be generated directly by the trained planning model. In other embodiments, the candidate implant model may be generated by iteratively refining an initial representation using a generative process, for example by progressively removing noise in a reverse diffusion process. In further embodiments, the candidate implant model may be generated by evaluating an implicit function representation of an implant model, such as a signed distance function, and reconstructing a surface representation therefrom.

[0235] In step 1212, the candidate implant model and / or candidate implant parameters are evaluated with respect to the clinical constraints. The evaluation may determine whether the candidate implant model satisfies one or more validity criteria, including anatomical and / or prosthetic constraints. If the candidate implant model and / or candidate implant parameters do not satisfy the validity criteria, a controlled variation may be introduced into a latent space or a latent representation may be updated to generate a further candidate implant model. The generating and evaluating steps may optionally be repeated iteratively until one or more stopping criteria are met. Once the stopping criteria are met, the implant planning engine outputs a validated implant model and / or validated implant parameters.

[0236] Fig. 13A and 13B depict graphical models generated by a system for automated tooth implant planning according to an embodiment. In particular, the figuresdepict an example of a representation of 3D virtual patient data of a dento-maxillofacial complex of a patient similar to the depicted in Fig. 4B. Fig. 13A and 13B show the result of a planning according to the present disclosure, where Fig. 13B is an enlargement of the relevant part of Fig. 13A.

[0237] An implant model 1310 may be determined by the automatic implant planning system according to the embodiments in this disclosure based on implant parameters derived from the 3D virtual patient data and based on a crown design 1314. The implant model may be placed in the areas of interest 1316 and graphically rendered with the 3D virtual patient data. Fig. 13A and 13B additionally shows an implant sleeve mode 1312 associated with the tooth implant, which may by the automatic implant planning system as well.

[0238] The figures further depict neighboring teeth 1318 surrounding the area of interest 1316, thereby providing anatomical context for the placement of the implant model 1310. Additionally, a mandibular nerve 1320 is shown, which may represent an anatomical structure that is taken into account as a clinical constraint during the implant planning.

[0239] Furthermore, a jawbone 1322, in particular a mandible, is depicted, providing structural support for the implant model 1310.

[0240] Fig. 14 is a block diagram illustrating an exemplary data processing system that may be used in embodiments as described in this disclosure. Data processing system 1400 may include at least one processor 1402 coupled to memory elements 1404 through a system bus 1406. As such, the data processing system may store program code within memory elements 1404. Further, processor 1402 may execute the program code accessed from memory elements 1404 via system bus 1406. In one aspect, data processing system 1400 may be implemented as a computer that is suitable for storing and / or executing program code. It should be appreciated, however, that data processing system 1400 may be implemented in the form of any system including a processor and memory that is capable of performing the functions described within this specification.

[0241] Memory elements 1404 may include one or more physical memory devices such as, for example, local memory 1408 and one or more bulk storage devices 1410. Local memory 1408 may refer to random access memory or other non-persistent memory device(s) generally used during actual execution of the program code. A bulk storage device 1410 may be implemented as a hard drive or other persistent data storage device. The processing system 1400 may also include one or more cache memories (not shown) that provide temporary storage of at least some program code in order to reduce the number of times program code must be retrieved from bulk storage device 1410 during execution.

[0242] Input / output (I / O) devices depicted as input device 1412 and output device 1414 optionally can be coupled to the data processing system. Examples of input device 1412 may include, but are not limited to, for example, a keyboard, a pointing device such asa mouse, or the like. Examples of output device 1414 may include, but are not limited to, for example, a monitor or display, speakers, or the like. Input device 1412 and / or output device 1414 may be coupled to system bus 1406 of the data processing system 1400 either directly or through intervening I / O controllers (not shown). A network adapter 1416 may also be coupled to data processing system 1400 to enable it to become coupled to other systems, computer systems, remote network devices, and / or remote storage devices through intervening private or public networks. The network adapter 1416 may comprise a data receiver for receiving data that is transmitted by said systems, devices and / or networks to said data and a data transmitter for transmitting data to said systems, devices and / or networks. Modems, cable modems, and Ethernet cards are examples of different types of network adapter that may be used with data processing system 1400.

[0243] As pictured in Fig. 14, memory elements 1404 may store an application 1418. It should be appreciated that data processing system 1400 may further execute an operating system (not shown) that can facilitate execution of the application. Application, being implemented in the form of executable program code, can be executed by data processing system 1400, e.g., by processor 1402. Responsive to executing application, data processing system may be configured to perform one or more operations to be described herein in further detail.

[0244] In one aspect, for example, data processing system 1400 may represent a client data processing system. In that case, application 1418 may represent a client application that, when executed, configures data processing system 1400 to perform the various functions described herein with reference to a "client". Examples of a client can include, but are not limited to, a personal computer, a portable computer, a mobile phone, or the like.

[0245] In another aspect, data processing system may represent a server. For example, data processing system may represent an (HTTP) server in which case application 1418, when executed, may configure data processing system to perform (HTTP) server operations. In another aspect, data processing system may represent a module, unit or function as referred to in this specification.

[0246] The techniques of this disclosure may be implemented in a wide variety of devices or apparatuses, including a wireless handset, an integrated circuit (IC) or a set of ICs (e.g., a chip set). Various components, modules, or units are described in this disclosure to emphasize functional aspects of devices configured to perform the disclosed techniques, but do not necessarily require realization by different hardware units. Rather, as described above, various units may be combined in a codec hardware unit or provided by a collection of interoperative hardware units, including one or more processors as described above, in conjunction with suitable software and / or firmware.The storage devices used at embodiments of the present disclosure can be implemented in various ways, as long as the storage device can fulfil the requirements of its respective usage. If the data in the storage device are not changed within the storage device, the storage device can be a read-only memory, e.g. a ROM, a portable compact disc readonly memory (CD-ROM), other optical storage devices, or the like. If the storage device should provide the possibility for changing stored data or if convenience suggest it, the storage device can also be a read-write memory, e.g. a portable computer diskette, a hard disk, a random access memory (RAM), an erasable programmable read-only memory (EPROM or Flash memory), a magnetic storage device, or any suitable combination of the foregoing, just to name a few possible storage devices.

[0247] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0248] The corresponding structures, materials, acts, and equivalents of all means or step plus function elements in the claims below are intended to include any structure, material, or act for performing the function in combination with other claimed elements as specifically claimed. The description of the present invention has been presented for purposes of illustration and description, but is not intended to be exhaustive or limited to the invention in the form disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the invention. The embodiment was chosen and described in order to best explain the principles of the invention and the practical application, and to enable others of ordinary skill in the art to understand the invention for various embodiments with various modifications as are suited to the particular use contemplated.

[0249] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. The conjunction “and / or” between a first feature and a second featureshould be understood in such a way that a first embodiment comprises the first feature, a second embodiment comprises the feature and a third embodiment comprises both the first feature and the second feature.

[0250] The corresponding structures, materials, acts, and equivalents of all means or step plus function elements in the claims below are intended to include any structure, material, or act for performing the function in combination with other claimed elements as specifically claimed. The description of the present invention has been presented for purposes of illustration and description, but is not intended to be exhaustive or limited to the invention in the form disclosed. The embodiment was chosen and described in order to best explain the principles of the invention and the practical application, and to enable others of ordinary skill in the art to understand the invention for various embodiments with various modifications as are suited to the particular use contemplated.

Claims

1. CLAIMS1. A computer-implemented method for automated planning of a tooth implant in a dento-maxillofacial complex, the method comprising:receiving or determining 3D virtual patient data representing a dento-maxillofacial complex comprising anatomical structures and at least one area of interest for placing the tooth implant, the 3D virtual patient data including 3D representations of anatomical structures within a vicinity of the at least one area of interest,determining a tooth implant model and / or one or more implant parameters defining a tooth implant model to be positioned at the at least one area of interest using an implant planning engine, e.g. the implant planning engine being configured to receive the 3D representations of anatomical structures within a vicinity of the area of interest and to determine the tooth implant model and / or one or more implant parameters defining a tooth implant model based on clinical constraints, the clinical constraints including distance requirements of the tooth implant with regard to at least part of the 3D representations of the anatomical structures; and, optionally,rendering the implant model on a display or transforming the implant model into a data format for controlling a manufacturing apparatus, such as a 3D printer, to manufacture the tooth implant.

2. Method according to claim 1, wherein determining 3D virtual patient data comprises:receiving 3D maxillofacial data and, optionally, 2D maxillofacial data representing at least parts of the dento-maxillofacial complex, the 3D maxillofacial data preferably comprising (CB)CT data, IOS data, panoramic radiographs and / or extraoral scan data;segmenting the 3D maxillofacial data into segmented 3D representations of anatomical structures of the maxillofacial complex; and,aligning the segmented 3D representations of anatomical structures to form the 3D virtual patient data.

3. Method according to claims 1 or 2, additionally comprising:receiving a crown design, the crown design being designed to be placed within the area of interest and being attached to the tooth implantproviding the crown design to the input of the implant planning engine which additionally determines the tooth implant model and / or one or more implant parameters defining a tooth implant model based on the crown design.

4. Method according to any of claims 1-3, wherein the clinical constraints comprise at least two categories of constraints, a category of constraints defining an importance that a constraint of the category is met, the implant planning engine being configured to determine the one or more implant parameters in such a way that constraints of a high importance category are met, while constraints of a low importance category may be violated.

5. Method according to any of claims 1-4 wherein the clinical constraints including at least one of: a distance between the tooth implant and adjacent anatomical structure, e.g. neighboring nerves, nasal passage, neighboring teeth, neighboring implants, jawbone rim; a relative position between the tooth implant model and the crown model, such as limited deviation of their axis and / or their posterior and anterior region alignment; information about bone support for the tooth implant; a distance between neighboring implants; information for protection of the Inferior Alveolar Nerve (IAN); information for a suitable posterior region alignment; information for a suitable interior region alignment; information for vertical positioning of the implant; information for suitable tooth implant dimensions.

6. Method according to any of claims 1-5, further comprising preprocessing the 3D representations of anatomical structures, the preprocessing preferably includes transforming the 3D representation, for example a mesh representation, into a 3D representation that can be processed by the implant planning engine.

7. Method according to any of claims 1-6, further comprising receiving data regarding commercially available tooth implants, determining the one or more implant parameters additionally constrained by the received data so that one or more determined implant parameters describe one of the commercially available tooth implants.

8. Method according to any of claims 1-7, wherein the implant planning engine comprises optimization algorithm, preferably a constraint-based optimization algorithm, wherein the implant planning engine is configured to:receive or generating an initial tooth implant model and / or an initial set of implant parameters defining a tooth implant model;determine costs of the an initial tooth implant model and / or an initial set of implant parameters using a cost function, which determines the costs based on the clinical constraints;determine an optimized implant model or a set of implant parameters defining an optimized implant model by adapting the initial implant model or the set of implant parameters defining an implant model until the costs determined by the cost function is minimized and / or a stopping criteria is met.

9. Method according to any of claims 1 to 8, wherein the implant planning engine comprises a trained predictive deep learning model, the trained predictive deep learning model being trained to receive a 3D representation of the anatomical structures and to output a prediction for the one or more implant parameters representing an tooth implant meeting at least part of the clinical constraints.

10. Method according to any of claims 1 to 9, wherein the implant planning engine comprises a trained generative deep learning model, preferably a diffusion model or a signed distance function (SDF) model, the trained generative deep learning model being trained to receive 3D representations of the anatomical structures and to generate and output the one or more implant parameters representing a tooth implant meeting at least part of the clinical constraints.

11. Method according to claim 9 or 10, wherein the trained predictive deep learning model or the trained generative deep learning model is configured to receive sets of 3D virtual patient data and associated approved implant parameters a ground truth and to tune model parameters based on the received sets of 3D virtual patient data and the approved implant parameters, the approved implant parameters being generated at earlier usages of the implant planning engine and being approved by a user.

12. Method according to any of claims 1 to 11, additionally comprising determining sleeve parameters describing an implant sleeve, the sleeve parameters being determined by a sleeve generator based on the implant parameters.

13. Method according to any of claims 1 to 12, wherein the one or more implant parameters include at least one of: a position of the tooth implant within the area of interest, an orientation of the tooth implant, a dimension of the tooth implant.

14. A computer implemented method for training an implant planning engine comprising a predictive deep learning model or a generative deep learning model to automatically predict implant parameters, the method comprising:receiving training data comprising sets of 3D virtual patient data, associated ground truth implant parameters, and, optionally, clinical constraints, the 3D virtual patient data representing a dento-maxillofacial complex comprising anatomical structures and at least one area of interest for placing the tooth implant, the 3D virtual patient data including a 3D representation of the anatomical structures, and the ground truth implant parameters describing a position of the tooth implant within an area of interest and / or an orientation of the tooth implant and / or dimensions of the tooth implant,providing one or more sets of the 3D virtual patient data as inputs to the implant planning engine and obtaining from the output of the implant planning engine estimated implant parameters, andtraining the parameters of the implant planning engine by updating its parameters by minimizing a loss function until a stopping criteria is met, wherein the loss function penalizes deviations between the estimated implant parameters and the ground truth implant parameters, wherein the optional clinical constraints are used as input features for the model and / or incorporated into the loss function as additional constraints.

15. System for automated planning of a tooth implant in a dento-maxillofacial complex, the system comprising:a computer readable storage medium having computer readable program code embodied therewith; and a processor, preferably a microprocessor, coupled to the computer readable storage medium, wherein responsive to executing the computer readable program code, the processor is configured to perform executable operations comprising:receiving or determining 3D virtual patient data representing a dento-maxillofacial complex comprising anatomical structures and at least one area of interest for placing the tooth implant, the 3D virtual patient data including 3D representations of anatomical structures within a vicinity of the at least one area of interest,determining a tooth implant model and / or one or more implant parameters defining a tooth implant model to be positioned at the at least one area of interest, using an implant planning engine, the implant planning engine being configured to receive the 3D representations of anatomical structures within a vicinity of the area of interest and to determine the tooth implant model and / or one or more implant parameters defining a tooth implant model based on clinical constraints, the clinical constraints including distance requirements of the tooth implant with regard to at least part of the 3D representations of the anatomical structures; and, optionally,rendering the implant model on a display or transforming the implant model into a data format for controlling a manufacturing apparatus, such as a 3D printer, to manufacture the tooth implant.determining one or more implant parameters for the tooth implant to be positioned at the at least one area of interest, , the one or more implant parameters being determined based on an implant planning engine, the implant planning engine being configured to receive the 3D representations of anatomical structures within a vicinity of the area of interest and to determine the one or more implant parameters based on clinical constraints, the clinical constraints including distance requirements of the tooth implant with regard to the 3D representations of the anatomical structures.

16. System according to claim 16 wherein the processor of the system is configured to perform executable operations comprising any of the steps of claims 2-13.

17. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry the method steps according to any of claims 1 to 13.

Images