Dental cad automation using deep learning

Abstract

A computer-implemented method of recognizing dental information associated with a dental model of dentition includes training a deep neural network to map a plurality of training dental models representing at least a portion of each one of a plurality of patients' dentitions to a probability vector including probability of the at least a portion of the dentition belonging to each one of a set of multiple categories. The category of the at least a portion of the dentition represented by the training dental model corresponds to the highest probability in the probability vector. The method includes receiving a dental model representing at least a portion of a patient's dentition and recognizing dental information associated with the dental model by applying the trained deep neural network to determine a category of the at least a portion of the patient's dentition represented by the received dental model.

What is claimed is: 
     
         1 . A computer-implemented method for recognizing dental information using DNNs, the method comprising:
 receiving, by one or more computing devices, a patient's scan data representing at least one portion of the patient's dentition data set;   identifying, using a trained deep neural network, one or more dental features in the patient's scan data based on one or more output probability values of the trained deep neural network;   determining locations of the identified one or more dental features including the locations of a prepared tooth; and   generating a contour of a crown for the prepared tooth based on the identified one or more dental features and the locations of the identified one or more dental features in the patient's scan data, wherein the crown is to be attached to the prepared tooth.   
     
     
         2 . The method of  claim 1 , wherein the trained deep neural network maps one or more dental features in at least one portion of each dentition training data set from a plurality of dentition training data sets to one or more highest probability values of a probability vector. 
     
     
         3 . The method of  claim 2 , wherein the trained deep neural network maps locations of the one or more dental features in the at least one portion of each dentition training data set to a highest location probability value of a location probability vector; and
 wherein determining locations of the identified one or more dental features in the patient's scan data is based on output probability values.   
     
     
         4 . The method of  claim 1 , further comprising comparing, using a discriminating deep neural network, the generated contour of the crown for the prepared tooth in the patient's scan data with an image of a real crown contour from one or more real-sample data sets to output a loss function based on the comparison, and wherein the deep neural network is configured to generate a second contour of the crown based on the output loss function of the second discriminating deep neural network. 
     
     
         5 . The method of  claim 1 , wherein one or more dental features in the at least one portion of each dentition training data set comprise one or more of a tooth surface anatomy, a tooth dentition, and a restoration type. 
     
     
         6 . The method of  claim 2 , wherein the probability vector comprises a plurality of probability values, wherein each probability value indicating a probability that the one or more dental features in at least one at least one portion of each dentition training data set belonging to one or more of a tooth surface anatomy, a tooth dentition, and a restoration type. 
     
     
         7 . The method of  claim 2 , wherein each of the training data sets is segmented into different portions, wherein each portion represents a characteristic of a tooth surface anatomy, a tooth dentition, or a restoration type. 
     
     
         8 . The method of  claim 1 , wherein the patient's scan data is segmented into different portions prior to identifying the one or more dental features in the patient's scan data, wherein each different portion represents an aspect of a tooth surface anatomy, a tooth dentition, or a restoration type. 
     
     
         9 . The method of  claim 5 , wherein the tooth surface anatomy comprises one or more features selected from a group of buccal and lingual cusps, distobucall and mesiobuccal inclines, distal and mesial cusp ridges, distolingual and mesiolingual inclines, occlusal surface, and buccal and lingual arcs. 
     
     
         10 . The method of  claim 5 , wherein the tooth dentition comprises one or more classifications selected from a group consisting of upper and lower jaws, prepared and opposing jaws, prepared tooth, and tooth numbers. 
     
     
         11 . The method of  claim 5 , wherein the restoration type comprises one or more restoration selected from a group consisting of crown, inlay, bridge, and implant. 
     
     
         12 . The method of  claim 2 , wherein the plurality of dentition training data sets have been preprocessed to generate a depth map for each training data set, wherein each training data set comprises three dimensional (3D) data. 
     
     
         13 . The method of  claim 12 , wherein the plurality of dentition training data sets are preprocessing prior to the training of the deep neural network. 
     
     
         14 . The method of  claim 12 , wherein the depth map is generated by converting 3D coordinates of each point of the 3D data into a distance value from a given plane to each point. 
     
     
         15 . A computer program product comprising a computer-readable storage medium having computer program logic recorded thereon for enabling a processor-based system to recognize dental information and to design a dental restoration from the recognized dental information, the computer program product comprising:
 a first program logic module for enabling the processor-based system to receive a patient's scan data representing at least one portion of the patient's dentition data set; and   a second program logic module for enabling the processor-based system to use a first deep neural network to identify one or more dental features in the patient's scan data based on one or more output probability values of the deep neural network,   wherein the second program logic module comprises:   logic for enabling the processor-based system to determine the locations of a prepared tooth; and   logic for enabling the processor-based system to generate a contour of a crown for the prepared tooth based on the identified one or more dental features in the patient's scan data, wherein the crown is to be attached to the prepared tooth.   
     
     
         16 . The computer program product of  claim 15 , wherein the trained deep neural network maps one or more dental features in at least one portion of each dentition training data set from a plurality of dentition training data sets to one or more highest probability values of a probability vector. 
     
     
         17 . The computer program product of  claim 15 , wherein a second trained deep neural network outputs a loss function based on a comparison of the generated contour of the crown for the prepared tooth in the patient's scan data with an image of a real crown contour from a real-sample data sets, and wherein the first neural network is configured to generate a second contour of the crown based on the output loss function of the second trained deep neural network. 
     
     
         18 . The computer program product of  claim 15 , wherein the one or more dental features in the at least one portion of each dentition training data set comprise one or more of a tooth surface anatomy, a tooth dentition, and a restoration type. 
     
     
         19 . The computer program product of  claim 15 , wherein the plurality of dentition training data sets have been preprocessed to generate a depth map for each training data set, wherein each training data set comprises three dimensional (3D) data. 
     
     
         20 . A system for fabricating a dental restoration from a patient's dentition data, the system comprising:
 a dental restoration client, wherein the dental restoration client receives, from a user interface, an input indicating a selection of a dental restoration type to be fabricated;   a 3D modeling module, wherein the 3D modeling module:
 receives a dentition data set of a patient, wherein the patient's dentition data set is generated by scanning a 3D impression or model of the patient's teeth; 
 selects a deep neural network pre-trained by a group of training data sets designed to model a specific restoration type that matches the selected dental restoration type; 
 uses the patient's dentition data set as an input to the selected pre-trained deep neural network; and 
 generates an output restoration model using the selected pre-trained deep neural network based on the patient's dentition data, wherein the restoration model is to be used on a preparation site.