Method, system, and computer program for generating a 3D model of dentures in occlusion

Abstract

According to some embodiments of the present invention, there is provided a method of generating 3D models of a first denture and a second denture, the method comprising: obtaining a set of data by x-ray scanning the first denture and the second denture in an occlusal state; generating a 3D image of the first denture and the second denture in an occluded state, the 3D image comprising voxels, from the acquired data; classifying each voxel of the generated 3D image as belonging to either the first denture, the second denture, or the space between the first and second denture; generating 3D models of the first denture and the second denture from the classified voxels; Classifying the voxels of the generated 3D image is based on features of neighboring voxels in the generated 3D image according to an iterative analysis between the neighboring voxels.

Description

[Technical Field]

[0001] The present invention relates generally to the field of dental imaging, and more particularly but not exclusively to a method, system and computer program for generating a 3D model of dentures in occlusion. [Background technology]

[0002] Radiography has proven valuable to dentists by helping to identify various problems related to a patient's teeth and supporting structures, as well as verify other measurements and observations. Extraoral imaging devices, one type of X-ray system that holds particular promise for improving dental treatment, can acquire one or more X-ray images sequentially. Here, multiple images of the patient are acquired at different angles, and these images are combined to obtain a 3D reconstruction of the patient's jaw dentition and other facial features. Various types of imaging devices have been proposed to provide this type of volumetric image content. In these systems, a radiation source and an image detector are maintained at a known distance from each other (e.g., fixed or variable) and rotate synchronously around the patient at various angles, capturing a series of images by directing and detecting radiation passing through the patient at the different rotation angles. For example, volumetric images (e.g., 3D or volumetric image reconstructions) showing the shape and dimensions of the head and jaw structures can be obtained using computed tomography (CT), such as cone-beam computed tomography (CBCT), or other volumetric imaging techniques. The resulting volumetric images have proven particularly valuable for providing useful information to aid in diagnosis and treatment.

[0003] These techniques have proven effective for imaging the jaw, but are also used for other dental applications, for example, in preparation for orthodontic treatment and for replicating dentures.

[0004] It should be noted here that dentures are divided into two parts: the inner arch and the outer arch. The inner arch is the inner surface of the denture that comes into contact with the patient's gums or oral structures (also called the bearing surface, mating surface, or support surface). The outer arch is the outer surface of the denture that corresponds to the teeth and gums of the reconstructed dentition. The outer arch surface includes the occlusal surface, which is the surface that comes into contact with the opposing dental arch.

[0005] Below are the general steps performed by a practitioner to create dentures. ·Occlusal surface position adjustment - Create a double-jaw imp using remaining teeth and gums · Create a provisional positive physical model of the denture using jaw imps Adjusting the position of the dentures (vertically and laterally for each tooth) according to aesthetic criteria (e.g., the patient's face, gums, lips)

[0006] Patients may experience bone resorption, which locally deforms the denture support surfaces. Therefore, situations arise where the denture needs to be adapted to these changes. In such cases, practitioners must duplicate the dentures before adjusting the internal arch surfaces. Denture duplication refers to duplicating the existing dentures, regardless of whether the existing dentures are modified. It should be noted that when duplicating dentures, the occlusal and interocclusal surfaces may be modified, for example, to account for natural bone resorption and / or occlusal changes. Denture duplication is a laborious and complex process. To address these drawbacks, digital tools exist that can reduce chair time by approximately 75%. This also reduces manufacturing costs and improves the patient experience by improving comfort and reducing the number of visits. These tools are typically based on using a desktop scanner to scan the dentures.

[0007] Desktop scanners are optical scanners that allow scanning the denture surface. However, it is not possible to scan it completely in a single step. For each denture, both sides corresponding to the inner and outer arc surfaces must be scanned. Then, with the denture in occlusion, a third acquisition (vestibular) is made, and the acquired inner arc surface and the acquired outer arc surface are co-registered to obtain the complete denture surface.

[0008] Cone beam computed tomography (CBCT) can also be used to scan dentures. Unless the dentures are in occlusion, each denture must be scanned and then the 3D image of the dentures aligned, which can be time-consuming for practitioners. Scanning dentures in occlusion using CBCT produces low-quality results because it is difficult to separate the representation of each denture in the 3D image using traditional thresholding methods. Furthermore, the orientation of the occlusal plane is usually unknown (and therefore difficult to estimate). Denture materials can also vary between manufacturers and even within the denture itself (which can result in different X-ray attenuation between parts of the denture, which can affect the accuracy of mesh extraction during 3D model generation). Therefore, there is a need to improve the process of replicating dentures, particularly to reduce costs and improve patient comfort. [Prior art documents] [Patent documents]

[0009] [Patent Document 1] European Patent Application Publication No. 3654289 Summary of the Invention [Problem to be solved by the invention]

[0010] The present invention has been devised to address one or more of the above-mentioned concerns. In this regard, a method, a system and a computer program are provided that allow for the automatic generation of a 3D model of dentures in occlusion. [Means for solving the problem]

[0011] According to one aspect of the present invention, there is provided a method for generating 3D models of a first denture and a second denture, the method comprising: obtaining a set of data by x-ray scanning the first denture and the second denture in an occlusal state; generating a 3D image of the first denture and the second denture in an occluded state, the 3D image comprising voxels, from the acquired data; classifying each voxel of the generated 3D image as belonging to either the first denture, the second denture, or the space between the first and second denture; generating 3D models of the first denture and the second denture from the classified voxels; Classifying the voxels of the generated 3D image is based on features of neighboring voxels in the generated 3D image according to an iterative analysis between the neighboring voxels.

[0012] The method according to the present invention allows automatic generation of a 3D model of dentures in occlusion, even without any knowledge of the occlusal orientation, and allows for easy, fast, and reliable acquisition of occlusion information. To this end, the entire denture (upper and lower jaws) is scanned just once, ensuring perfect fit to the patient (occlusion and bearing surface), and the resulting data is split into two 3D models (3D surfaces or 3D meshes). Furthermore, the meshes are represented in the same reference system, so the practitioner does not need to adjust the mesh alignment.

[0013] According to some embodiments of the present invention, the classification of each voxel of the generated 3D image is based on a watershed algorithm.

[0014] Further according to some embodiments of the present invention, classifying each voxel of the generated 3D image comprises: obtaining a 3D image showing the voxel's belonging to the dentures; applying a shrinkage algorithm to obtain a first set of voxels belonging to the first denture and a second set of voxels belonging to the second denture, separated by voxels not belonging to the denture; and applying a dilation algorithm to identify the contact surface between the first and second dentures.

[0015] According to some embodiments of the present invention, the method further comprises obtaining an outer surface of the denture that limits the expansion algorithm. Further according to some embodiments of the present invention, classifying each voxel of the generated 3D image is based on a deep neural network. Further according to some embodiments of the present invention, the deep neural network is a convolutional neural network. Further according to some embodiments of the present invention, at least a portion of the generated 3D model is obtained directly from the generated 3D image. Further according to some embodiments of the present invention, at least one of the generated 3D models is obtained using a Marching Cubes algorithm. According to some embodiments of the present invention, the method further comprises applying a smoothing algorithm to at least one portion of the generated 3D model. Further according to some embodiments of the present invention, the generated 3D model is a 3D surface or a 3D mesh. According to some embodiments of the present invention, the method further comprises displaying, transmitting and / or storing the generated 3D model. Further according to some embodiments of the present invention, the scan is a cone beam computed tomography, CBCT, scan. Further according to some embodiments of the present invention, at least the first or second dentures are made of materials with different X-ray attenuation.

[0016] According to another aspect of the present invention, there is provided a device comprising a processing unit configured to perform the steps of the above method. The other aspects of the present disclosure also have advantages similar to those of the above-mentioned one aspect.

[0017] At least part of the methods of the present invention may be computer-implemented. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, microcode, etc.), or an embodiment combining software and hardware aspects, which may be referred to collectively herein as a "circuit," "module," or "system." Furthermore, the present invention may take the form of a computer program product embodied in any tangible medium of expression having computer-usable program code embodied therein.

[0018] Since the present invention can be realized in software, it can be embodied as computer-readable code for provision to a programmable apparatus on any suitable carrier medium. Tangible carrier media can include storage media such as floppy disks, CD-ROMs, hard disk drives, magnetic tape devices, or solid-state memory devices. Transient carrier media can include signals such as electrical, electronic, optical, acoustic, magnetic, or electromagnetic signals, e.g., microwave or RF signals. [Brief explanation of the drawings]

[0019] Embodiments of the present invention will now be described, by way of example only, with reference to the following drawings, in which: [Figure 1a] Shown are the first and second dentures in occlusion, CBCT 3D images of the first and second dentures in occlusion, and coronal and sagittal sections of the CBCT 3D images. [Figure 1b] Shown are the first and second dentures in occlusion, CBCT 3D images of the first and second dentures in occlusion, and coronal and sagittal sections of the CBCT 3D images. [Figure 1c]Shown are the first and second dentures in occlusion, CBCT 3D images of the first and second dentures in occlusion, and coronal and sagittal sections of the CBCT 3D images. [Figure 1d] Shown are the first and second dentures in occlusion, CBCT 3D images of the first and second dentures in occlusion, and coronal and sagittal sections of the CBCT 3D images. [Figure 2] 1 is a schematic diagram illustrating an imaging apparatus for CBCT imaging of an object. [Figure 3] 1 shows an example of steps in a method for generating a 3D model of dentures in occlusion. [Figure 4] 2 shows an example of steps for identifying voxels belonging to a first denture and a second denture according to a first embodiment. [Figure 5] Voxels belonging to the dentures and voxels belonging to the background are illustrated in the axial, coronal, and sagittal planes. [Figure 6] The boundary between voxels belonging to dentures and voxels belonging to the background is illustrated in the axial, coronal, and sagittal planes. [Figure 7] After applying a shrinkage algorithm to distinguish between sets of voxels representing the upper and lower jaws, voxels belonging to the dentures and voxels belonging to the background are illustrated in axial, coronal and sagittal planes. [Figure 8] We illustrate a deep neural network that can be used to process the generated 3D images of the first and second dentures in their occlusion state to identify voxels belonging to the first denture and voxels belonging to the second denture. [Figure 9] 1 illustrates steps that can be used to improve the accuracy of the 3D models of the first and second dentures when generating them. [Figure 10] An example of smoothing the contact surface between the upper and lower jaws is shown. [Figure 11] 1 is a schematic block diagram of a computing device for implementing one or more embodiments of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0020] The following is a detailed description of specific embodiments of the present invention, with reference to the drawings in which like reference numerals identify like elements of structure in each figure.

[0021] In the drawings and description that follow, like elements are designated by like reference numerals, and like descriptions of elements already described and the arrangement and interaction of elements are omitted. The terms "first," "second," etc., used do not necessarily represent any order or priority relationship, and may simply be used to more clearly distinguish one element from another, unless otherwise specified.

[0022] In the context of this disclosure, the terms "viewer," "operator," and "user" are considered equivalent and refer to a practitioner, technician, or other human viewing an image who may acquire, view, and manipulate the x-ray image on a display monitor. "Operator commands," "user commands," or "viewer commands" result from explicit commands entered by the viewer, such as by clicking a button on the system hardware, or by using a computer mouse or touch screen or keyboard entry.

[0023] In the context of this disclosure, the phrase "in signal communication" indicates that two or more devices and / or components can communicate with each other via signals traveling over some type of signal path. The signal communication can be wired or wireless. The signals can be communication signals, power signals, data signals, or energy signals. The signal path can include physical, electrical, magnetic, electromagnetic, optical, wired, and / or wireless connections between a first device and / or component and a second device and / or component. The signal path can further include additional devices and / or components between the first device and / or component and the second device and / or component.

[0024] According to some embodiments of the present invention, cone beam computed tomography (CBCT) is used to scan a first denture (e.g., an upper denture) and a second denture (e.g., a lower denture) in occlusion to obtain 3D images that are processed to obtain 3D models of each denture in occlusion. The occlusal plane may be oriented in any direction. According to other embodiments, CBCT is used to scan the dentures and jaw mold in occlusion, thereby obtaining 3D images that are processed to obtain 3D models of the dentures and jaw mold in occlusion. For clarity, the description will be based on scanning the first and second dentures in occlusion to obtain 3D models, e.g., 3D surfaces or 3D meshes, of each denture. The occlusal relationship corresponds to the intercuspal position.

[0025] Unless otherwise specified, the term "3D model" refers to a 3D mesh or a 3D surface, and the term "3D image" refers to a 3D volumetric image. 1a to 1d show the first and second dentures in occlusion, CBCT 3D images of the first and second dentures in occlusion, and coronal and sagittal sections of the CBCT 3D images, respectively.

[0026] As shown in FIG. 1a, the first and second dentures 100, 105 (here, the upper denture 100 and the lower denture 105) each include portions corresponding to teeth 110 and gums 115. These portions are made of different materials, such as acrylic composite, ceramic, titanium, and zirconia. These different materials absorb X-rays in different ways, resulting in different gray levels in the image, as shown in FIG. 1b. Therefore, it is difficult to accurately define the contours of the various portions of the dentures using a single threshold value.

[0027] According to some embodiments of the present invention, a set of X-ray images of the first and second dentures in an occlusal position, such as X-ray image 120 of FIG. 1b, are used to accurately, quickly, and cost-effectively generate 3D models of each of the first and second dentures.

[0028] As shown, the first and second dentures in occlusion can be viewed in a coronal plane 125 as shown in FIG. 1c, and in axial and sagittal planes 130 and 135 as shown in FIG. 1d. FIG. 2 is a schematic diagram showing an imaging apparatus for CBCT imaging of an object.

[0029] As shown, imaging device 200 can be used to acquire, process, and display CBCT images of an object on a support, such as dentures positioned in occlusion on a tray. A transport device 205 at least partially rotates a generator having a detector 210 and an X-ray source 215 around the support position to acquire multiple 2D projection images used to generate (or reconstruct) a 3D volumetric image (referred to as a 3D image). A control logic processor, such as a control logic processor in a server 220, provides power to the X-ray source 215, the detector 210, the transport device 205, and other imaging devices to acquire the image content necessary for 3D imaging of the object. The control logic processor may include memory and is in signal communication with a display and / or a remote computer, such as a laptop computer 225, for inputting operator instructions and displaying image results. The server 220 and the laptop computer 225 may be in signal communication with the X-ray source 215, the detector 210, the transport device 205, and other imaging devices via a communications network 230.

[0030] 3D model generation of dentures FIG. 3 shows an example of steps in a method for generating a 3D model of a denture in occlusion. As shown, the first step is to place a first denture in occlusal relationship with a second denture (step 300), where the first and second dentures are intended to fit a particular patient, and the occlusal relationship corresponds to an intercuspal position. As disclosed above, the first or second denture can be replaced with another element, such as a dental stone cast. According to some embodiments, the dentures are placed in occlusion on a tray within a CBCT imaging device, as described with reference to FIG.

[0031] It should be noted that due to the occlusal relationship between the scanned dentures, the practitioner does not need to align a first 3D model (e.g., a first 3D mesh) corresponding to a first denture onto a second 3D model (e.g., a second 3D mesh) corresponding to a second denture after the 3D volume is acquired.

[0032] Next, as described with reference to FIG. 2, a data set is acquired for the first and second dentures in occlusal relationship using, for example, a CBCT imaging device (step 305).

[0033] The acquired data is then processed, for example, using standard algorithms, to generate one or more 3D images (step 310). That is, one or more sets of voxels forming one or more 3D volumes are generated. Preferably, all voxels are the same size. For illustrative purposes, the length of a voxel edge may be between 75 and 300 micrometers. Optionally, if the dentures contain metal, a metal artifact reduction (MAR) algorithm may be used when generating the one or more 3D images. An example of this algorithm is described in EP 3654289. Optionally, other algorithms, such as noise reduction and contrast enhancement algorithms, may also be used when generating the one or more 3D images.

[0034] Next, each voxel belonging to the first denture and each voxel belonging to the second denture are identified. For example, each voxel in the generated 3D image is classified as belonging to the first denture, the second denture, or the air surrounding the two dentures (step 315). Thus, in the generated 3D image being analyzed, first, second, and third sets of voxels corresponding to the first denture, the second denture, and the air surrounding the dentures (also called background) are identified, respectively.

[0035] While distinguishing between voxels belonging to dentures and voxels belonging to the background can be easily achieved, for example, by comparing the value of a voxel to a threshold, possibly weighted with the values ​​of neighboring voxels, distinguishing between voxels belonging to a first denture and voxels belonging to a second denture can be more difficult, especially when the first and second dentures are in contact and when the contacting materials have the same or similar X-ray attenuation. For illustrative purposes, distinguishing between voxels belonging to a first denture and voxels belonging to a second denture can be based on a watershed technique, as described with reference to Figures 4 to 7, or on an artificial intelligence (AI)-based technique, such as a deep neural network, as described with reference to Figure 8. It should be noted that other methods (e.g., graph cuts, etc.) can also be used.

[0036] Next, a 3D model of the first denture, e.g., a 3D surface or a 3D mesh, is generated using all voxels belonging to the first denture (step 320). This can be achieved, for example, using known algorithms that allow for the generation of 3D surfaces corresponding to the boundary of an object represented by a 3D volume. Similarly, a 3D model of the second denture, e.g., a 3D surface or a 3D mesh, can be generated using all voxels belonging to the second denture. For illustrative purposes, step 320 can involve applying the known Marching Cubes algorithm (which extracts a polygonal mesh of isosurfaces from a three-dimensional discrete scalar field) to the 3D image. Each voxel represents a category of voxels: first denture, second denture, and background. Another example of generating a 3D model of dentures is disclosed with reference to FIG. 9.

[0037] The acquired 3D surface or 3D mesh may be displayed, saved, and / or transmitted (step 325). For example, the acquired 3D surface or 3D mesh may be saved on server 220 of FIG. 2 and sent to laptop 225 for local saving and / or display.

[0038] Voxel classification using the watershed algorithm According to some embodiments, the watershed algorithm is used to determine which voxels belong to the first denture and which belong to the second denture. It should be noted that the watershed algorithm was introduced by S. Beucher and C. Lantuejoul in 1979. The basic idea is to place a water source (seed) at the lowest point of the relief of the area, flood the entire relief from this water source (seed), and build walls wherever different water sources meet. The series of walls thus obtained constitutes a flooded watershed. FIG. 4 shows an example of steps according to the first embodiment for identifying voxels belonging to the first denture and the second denture, as disclosed with reference to step 315 of FIG.

[0039] As shown, the first step aims to distinguish between voxels belonging to the dentures and voxels belonging to the background (step 400), and depending on the algorithm used, some of the voxels belonging to the dentures may be classified as belonging to the first denture or the second denture.

[0040] For ease of explanation, the discrimination between voxels belonging to dentures and voxels belonging to the background can be performed by comparing the value of each voxel, possibly weighted by the values ​​of neighboring voxels, with a threshold (e.g., a predetermined threshold). If the value is higher than the threshold, the voxel can be considered to belong to dentures, and if the value is lower than the threshold, the voxel can be considered to belong to the background. Of course, the reverse is also possible. In this way, a binary 3D image is generated. The value of each voxel indicates whether it belongs to dentures or to the background.

[0041] According to another example, at least one voxel belonging to a lower denture and at least one voxel belonging to an upper denture are selected. The selection can be performed automatically or manually, for example, using some knowledge of the characteristics of the upper and lower jaws (e.g., location, texture, etc.). The selection can be performed on the generated 3D image or on a slice of the generated 3D image. Once these voxels are selected, voxels belonging to the first denture and the second denture are identified step by step using an iterative growing or dilation algorithm, depending on the iterative analysis of adjacent voxels. At each iteration, each neighboring voxel of the selected voxel is analyzed to determine whether it belongs to the same structure (i.e., the first denture or the second denture). The determination can be based on the value of the voxel, possibly weighted by the values ​​of neighboring voxels.

[0042] Region growing is performed over the entire volume. The growth stops when it determines that the neighboring voxels for the selected voxel are different (e.g., they are background voxels) or that the neighboring voxels have already been selected as voxels belonging to the dentures. All voxels that do not belong to the dentures are considered to belong to the background. It is observed that if the dentures are made of different types of materials, the first voxels selected are preferably those that are close to the occlusion (i.e., on the teeth).

[0043] It is also confirmed that the first and second dentures are in an occlusal state, and that the voxels belonging to the dentures are a set of continuous voxels. FIG. 5 shows examples of voxels belonging to the dentures (colored white) and voxels belonging to the background (colored black) in axial (reference numeral 500), coronal (reference numeral 505) and sagittal (reference numeral 510) views.

[0044] Next, prior to or in parallel with this, the boundary between voxels belonging to the dentures and voxels belonging to the background is obtained (step 405). This boundary can be obtained from the 3D image generated using gradient analysis (indicated by the dotted arrow in the figure) or from the voxels classified in step 400. Figure 6 shows an example of the boundary between voxels belonging to the dentures and voxels belonging to the background in the axial (reference numeral 600), coronal (reference numeral 605) and sagittal (reference numeral 610) planes.

[0045] Next, an iterative erosion algorithm is used to progressively reduce and decompose the shape formed by the voxels belonging to the dentures in order to separate the shapes of the first and second dentures according to an iterative analysis of adjacent voxels (steps 410 and 415). At each iteration, each voxel belonging to the dentures is analyzed, and voxels belonging to the dentures that are near the boundary between the voxels belonging to the dentures and the voxels belonging to the background are discarded (the boundary moves during the iteration). According to the illustrated example, after each iteration, a test is performed (step 410) to determine whether all remaining voxels belonging to the dentures can be grouped into two separate voxels representing the upper and lower jaws (step 415).

[0046] According to some embodiments, voxels are grouped into one or more contiguous (or connected) voxel pairs, i.e., groups of voxels belonging to dentures that are in direct contact or only connected through voxels belonging to the corresponding dentures. This can be performed using a connectivity algorithm that collects connected voxels and assigns them a common label. This process is repeated as long as there is one voxel pair. If voxels are grouped into at least two separate voxel pairs, a test is performed on the voxels in each pair to determine whether they represent the upper or lower jaw. Such a test can be based on a size criterion to ignore other factors. For illustrative purposes, the upper or lower jaw can be represented by a set of voxels that have predetermined characteristics, such as a denture height of 5 millimeters or more and a denture width of 4 centimeters or more.

[0047] Once the voxels belonging to the dentures are divided into voxel sets, one representing the upper jaw and one representing the lower jaw (step 420), the voxels belonging to these two sets are used as seeds for the watershed algorithm. Figure 7 shows examples of voxels belonging to the dentures and voxels belonging to the background in axial (reference number 700), coronal (reference number 705) and sagittal (reference number 710) views after applying a shrinkage algorithm, making it possible to distinguish between a set of voxels representing the upper jaw (reference number 715) and a set of voxels representing the lower jaw (reference number 720).

[0048] According to the watershed algorithm, an iterative growing or dilation algorithm is used to progressively identify voxels belonging to the first denture and voxels belonging to the second tooth, starting from voxels already identified as belonging to the first and second dentures and iteratively analyzing adjacent voxels. At each iteration, each adjacent voxel of a voxel already identified as belonging to the first or second denture is analyzed to determine whether it belongs to the same structure (step 425), i.e., the first or second denture, respectively. The determination may be based on the value of the voxel, possibly weighted by the values ​​of neighboring voxels. According to some embodiments, the expansion is limited by the boundary between the dentures and the background, as determined in step 405. Therefore, no expansion of the dentures voxels can occur within the background of the generated 3D image.

[0049] If no new voxels are identified during the iteration (step 430), the process ends. If it is determined that a voxel adjacent to a voxel already identified as belonging to the first or second denture is a voxel already identified as belonging to another denture, i.e., the second denture if the voxel under consideration belongs to the first denture, or the first denture if the voxel under consideration belongs to the second denture (step 435), or if it is determined that a neighboring voxel of a voxel already identified as belonging to the first denture is a neighboring voxel of a voxel already identified as belonging to the second denture, the expansion is locally stopped. In this case, the boundary between the voxel belonging to the first denture and the voxel belonging to the second denture defines the contact points between the first denture and the second denture, i.e., the set of contact points forming the contact surface between the first denture and the second denture (step 440). Typically, the boundary between the voxel belonging to the first denture and the voxel belonging to the second denture is a watershed surface of one voxel.

[0050] Voxel Classification Using AI-Based Algorithms As disclosed above, and according to certain embodiments, a deep neural network can be used to classify voxels in the generated 3D images and distinguish between voxels belonging to the first denture and those belonging to the second denture. Such a deep neural network can be derived from a convolutional neural network known as U-Net. The U-Net network consists of a contraction path and an expansion path, which form a stepwise analysis path according to the iterative analysis of neighbor-to-neighbor voxels. The contraction path is a typical convolutional network and consists of repeated application of convolutions followed by rectified linear unit (ReLU) and max-pooling operations, respectively. During contraction, spatial information is reduced, but feature information is increased. The expansion pathway combines the features and spatial information through a series of ascending convolutions and concatenation with high-resolution features from the contraction path. Artificial neural networks of the U-net type are described, for example, in the article entitled "U-net: Convolutional networks for biomedical image segmentation", Ronneberger, O., Fischer, P. & Brox, T., Medical Image Computing and Computer-Assisted Intervention - MICCAI 2015. Lecture Notes in Computer Science, 9351, 234-241 (Springer International Publishing, 2015).

[0051] FIG. 8 shows an example of a deep neural network that can be used to process the generated 3D images of the first and second dentures in occlusion to identify voxels belonging to the first denture and voxels belonging to the second denture.

[0052] In the illustrated example, the input to the deep neural network is a generated 3D image with 256 x 256 x 256 voxels, each of which can be coded with one byte. The output consists of two 3D images, corresponding to the upper and lower jaws, each of which also has 256 x 256 x 256 voxels, each of which can be coded with one byte. Each box represents a multi-channel feature map. The x, y, and z dimensions of the feature map are indicated by the first three digits of the label associated with the box, and the number of channels is indicated by the fourth digit (in parentheses). As shown in Figure 8, the shaded boxes represent duplicated feature maps, and the arrows indicate the operation. The two output volumes are represented in the same reference frame, so occlusion information is preserved.

[0053] Note that other types and / or sizes of inputs and outputs may also be used, for example, the input may be obtained using a 3D sliding window. As shown, the deep neural network architecture includes a contraction pass (left) and an expansion pass (right). The contraction pass aims to apply 3x3 biconvolutions several times, followed by a rectified linear unit (ReLU) and a 2x2 max-pooling operation with a stride of 2 for downsampling. Each downsampling step doubles the number of feature channels. Conversely, the expansion pass aims to upsample the feature maps. This is followed by a 3x3 convolution ("ascending convolution") that halves the number of feature channels, a concatenation with the corresponding cropped feature map from the contraction pass, and a 3x3 biconvolution, each followed by a ReLU. The final layer uses a 1x1 convolution to map each 32-component feature vector into two classes, each representing a 3D model of the upper or lower jaw.

[0054] According to some embodiments, the deep neural network shown in Figure 8 has been trained using real data in which the correct identification of the upper and lower jaws has been determined by experts. For illustrative purposes, a database containing 500 3D images (e.g., CBCT volumes) in which the upper and lower jaws are present and identified may be used. This database is divided into a training set, a test set, and a validation set, which are used for training. It should be noted that although the deep neural network shown in Figure 8 has been proven effective, some parameters such as the size of the feature maps and / or the number of channels can be varied.

[0055] 3D model rendering improvements FIG. 9 shows an example of steps that may be used to improve the accuracy of the 3D models of the first and second dentures when they are generated. As shown, the first step (step 900) aims to obtain the surfaces of the first denture and the second denture. To this end, voxels belonging to the second denture (e.g., obtained using the algorithm described with reference to FIG. 4 or the deep neural network described with reference to FIG. 8) are removed from the generated 3D image (e.g., by replacing the values ​​of voxels belonging to the second denture with predetermined values, such as the values ​​of voxels belonging to the background), to create a first 3D image of only the background and the first denture. Similarly, voxels belonging to the first denture (e.g., obtained using the algorithm described with reference to FIG. 4 or the deep neural network described with reference to FIG. 8) are removed from the generated 3D image (e.g., by replacing the values ​​of voxels belonging to the first denture with predetermined values, such as the values ​​of voxels belonging to the background), to create a second 3D image of only the background and the second denture.

[0056] A known marching cubes algorithm is then applied to each of the created first and second 3D images, with the result being two surfaces representing each denture, designated 1000 and 1005 in FIG. 10.

[0057] An optional next step may consist of smoothing the interface between the surfaces of the first and second dentures, i.e., the occlusal region, and correcting local discontinuities using standard smoothing algorithms (step 905). This can be done, for example, using a sliding average function. This can also be done independently for the upper and lower jaws, as shown in FIG. 10 with reference numeral 1010, and the surfaces of the first and second dentures superimposed on the generated 3D image.

[0058] Examples of smoothing algorithms are disclosed in the papers "Laplacian-isoparametric grid generation scheme", Herrmann, Leonard R., 1976, Journal of the Engineering Mechanics Division, 102(5), 749-756, and "Laplacian Surface Editing", Sorkine, O., Cohen-Or, D., Lipman, Y., Alexa, M., Rossl, C., Seidel, H.-P., 2004, Proceedings of the 2004 Eurographics / ACM SIGGRAPH Symposium on Geometry Processing. SGP '04. Nice, France: ACM. pp. 175-184.

[0059] It should be noted that the contact surface of the first denture (corresponding to the second denture) may be defined as the surface of the first denture (corresponding to the second denture) that is close to the surface of the second denture (corresponding to the first denture). For ease of explanation, a voxel of the first denture may be considered close to a voxel of the second denture if the voxels of the second denture are in contact or if only one background voxel separates these two voxels.

[0060] Figure 10 shows an example of smoothing the contact surfaces of the upper (reference number 1000) and lower (reference number 1005) jaws, as shown (reference number 1010), which allows a certain continuity to be achieved between the surfaces of the first and second dentures.

[0061] According to some embodiments, a smoothing algorithm may be applied directly to the surface generated from the 3D image, as described with reference to steps 315 and 320 of Figure 3. Each voxel represents a category of voxels: first denture, second denture, background. Other optional steps may be aimed at one or more of the following: Locally adjust the generated 3D image (e.g., the area representing dentures) according to the gray level isosurface values. Select the gray level threshold isosurface value from the threshold list based on the physical model material -Use different gray level threshold isosurface values ​​to filter the areas representing dentures and the areas representing the background Adjust the gray level threshold isosurface value after processing at least the area representing dentures

[0062] Example of hardware for performing steps of the method of the present disclosure FIG. 11 is a schematic block diagram of a computing device for performing the steps or portions of the steps described with particular reference to FIGS. 3, 4, 8 and 9 to implement one or more embodiments of the present invention. The computing device 1100 includes a communication bus that may be connected to all or some of the following elements: A central processing unit 1105, denoted as CPU, which may be, for example, a microprocessor. a random access memory 1110, shown as RAM, for storing executable code of the method of some embodiments of the invention and registers adapted to record variables and parameters required for implementing the method for generating 3D models of the first and second dentures, respectively, from a single 3D image of the first and second dentures in an occlusal state according to some embodiments of the invention, the memory capacity of which may be extended, for example, by an optional RAM connected to the expansion port; Read-only memory 1115, denoted as ROM, which stores computer programs for implementing some embodiments of the present invention. A user interface and / or input / output interface 1120 that can be used to receive input from a user, provide information to a user, and / or receive / transmit data from / to internal sensors and / or receive data from external devices, particularly sensors such as the X-ray sensor 210 of Figure 2. It can be built into the CBCT imager, which can be connected to the computing device 1100 via wired or wireless link communication. In some embodiments, an AI engine 1130.

[0063] Optionally, the communication bus of the computing device 1100 may be connected to a solid state disk 1135 (or hard disk) shown as SSD used as a mass storage device, an X-ray CBCT 1125, and / or a display 1140.

[0064] The communication bus of the computing device 1100 may also be connected to a network interface 1145. The network interface 1145 is typically connected to a communications network capable of sending and receiving digital data to and from remote devices, particularly the dental information system and / or the storage device 1135. The network interface 1145 can consist of a single network interface or a set of different network interfaces (e.g., wired and wireless interfaces, or different types of wired or wireless interfaces). Data packets are written to the network interface for transmission or read from the network interface for reception under the control of software applications running on the CPU 1105.

[0065] The executable code may be stored either in the read-only memory 1115, in the solid-state device 1135 or on a removable digital medium such as, for example, a memory card. According to a variant, the executable code of the program may be received by a communications network via the network interface 1145 so as to be stored in one of the storage means of the computing device 1100, such as the solid-state device 1135, before execution.

[0066] The central processing unit 1105 is configured to control and direct the execution of instructions of programs or portions of software code according to some embodiments of the present invention. The instructions are stored in one of the aforementioned storage means. After power-on, the CPU 1105 can execute instructions from the main RAM memory 1110 after instructions associated with a software application have been loaded, for example, from the ROM 1115 or the solid-state device 1135. Such software applications, when executed by the CPU 1105, cause the steps disclosed herein to be performed.

[0067] Any of the steps disclosed herein may be implemented in software by execution of a sequence of instructions or a program by a programmable computing machine such as a PC ("personal computer"), DSP ("digital signal processor"), or microcontroller, or in hardware by a machine or dedicated component such as an FPGA ("field programmable gate array") or ASIC ("application specific integrated circuit").

[0068] Although the present disclosure has been described with reference to several specific embodiments, the present invention is not limited to these specific embodiments, and modifications within the scope of the present invention will be apparent to those skilled in the art.

[0069] Many further modifications and variations will be apparent to those skilled in the art in light of the foregoing exemplary embodiments, which are illustrative and not intended to limit the scope of the invention, which is determined solely by the appended claims. In particular, different features of different embodiments may be interchanged where appropriate.

[0070] In the claims, the word "comprising" does not exclude other elements or steps, and the indefinite articles "a" or "an" do not exclude a plurality. The mere fact that different features are recited in mutually different dependent claims should not be interpreted as meaning that a combination of these features cannot be used to advantage.

Claims

1. 1. A method for generating a 3D model of a first denture and a second denture, comprising: The steps performed by the computer are: acquiring a set of data by x-ray scanning the first denture and the second denture in an occluded state; generating a 3D image of the first denture and the second denture in an occluded state, the 3D image comprising voxels, from the acquired data; classifying each voxel of the generated 3D image as belonging to either the first denture, the second denture, or the space between the first and second denture; generating 3D models of the first denture and the second denture from the classified voxels; A method, wherein classifying voxels of the generated 3D image is based on characteristics of neighboring voxels in the generated 3D image in response to an iterative analysis between neighboring voxels.

2. The method of claim 1 , wherein classifying each voxel of the generated 3D image is based on a watershed algorithm.

3. Classifying each voxel of the generated 3D image includes: - obtaining a 3D image showing the voxel's belonging to the dentures; applying a shrinkage algorithm to obtain a first set of voxels belonging to the first denture and a second set of voxels belonging to the second denture, separated by voxels not belonging to the denture; and applying a dilation algorithm to identify the contact surface between the first and second dentures.

4. The method of claim 3, further comprising, as a step executed by the computer, obtaining an outer surface of the denture that limits the expansion algorithm.

5. The method of claim 1 , wherein classifying each voxel of the generated 3D image is based on a deep neural network.

6. The method of claim 5 , wherein the deep neural network is a convolutional neural network.

7. The method of claim 1 , wherein at least a portion of the generated 3D model is obtained directly from the generated 3D image.

8. The method of claim 1 , wherein at least one of the generated 3D models is obtained using a Marching Cubes algorithm.

9. The method of claim 1, further comprising, as a step performed by the computer, applying a smoothing algorithm to at least one portion of the generated 3D model.

10. The method of claim 1 , wherein the generated 3D model is a 3D surface or a 3D mesh.

11. The method of claim 1, further comprising, as a step performed by the computer, displaying, transmitting, and / or saving the generated 3D model.

12. The method of claim 1 , wherein the scan is a cone beam computed tomography, CBCT, scan.

13. 10. The method of claim 1, wherein at least the first or second dentures are made of materials with different x-ray attenuation.

14. 14. A computer program product for a programmable device comprising a series of instructions which, when loaded into and executed by the programmable device, perform each of the steps of the method according to any one of claims 1 to 13.

15. A device comprising a processing unit configured to perform each of the steps of the method according to any one of claims 1 to 13.

Images