Training machine learning models to perform aligner damage prediction

Abstract

Embodiments relate to an aligner breakage solution that tests damage to an aligner using machine learning. A method includes of training a machine learning model to predict damage to an orthodontic aligner includes gathering a training dataset comprising digital designs for a plurality of orthodontic aligners, wherein each digital design is associated with a respective orthodontic aligner of the plurality of orthodontic aligners, and wherein each digital design comprises metadata indicating whether the associated respective orthodontic aligner was damaged during manufacturing of the associated respective orthodontic aligner. The method further includes training the machine learning model using the training dataset, wherein the machine learning model is trained to process data from a digital design for an orthodontic aligner and to output a probability that the orthodontic aligner associated with the digital design will be damaged during manufacturing of the orthodontic aligner.

What is claimed is: 
     
         1 . A method of training a machine learning model to predict damage to an orthodontic aligner, the method comprising:
 gathering a training dataset comprising digital designs for a plurality of orthodontic aligners, wherein each digital design is associated with a respective orthodontic aligner of the plurality of orthodontic aligners, and wherein each digital design comprises metadata indicating whether the associated respective orthodontic aligner was damaged during manufacturing of the associated respective orthodontic aligner; and   training the machine learning model using the training dataset, wherein the machine learning model is trained to process data from a digital design for an orthodontic aligner and to output a probability that the orthodontic aligner associated with the digital design will be damaged during manufacturing of the orthodontic aligner.   
     
     
         2 . The method of  claim 1 , wherein one or more of the digital designs comprises metadata indicating a location of damage that occurred for the associated respective orthodontic aligner. 
     
     
         3 . The method of  claim 1 , wherein gathering the training dataset comprises:
 receiving the digital designs for the plurality of orthodontic aligners, wherein the digital designs for the plurality of orthodontic aligners comprise at least one of digital models of the plurality of orthodontic aligners or digital models of molds used to manufacture the plurality of orthodontic aligners;   receiving information indicating one or more of the plurality of orthodontic aligners that were damaged during manufacturing; and   for each digital design associated with an orthodontic aligner that was damaged during manufacturing, adding the metadata to the digital design indicating that the orthodontic aligner was damaged during manufacturing.   
     
     
         4 . The method of  claim 1 , wherein the digital designs for the plurality of orthodontic aligners comprise at least one of digital models of the plurality of orthodontic aligners or digital models of molds used to manufacture the plurality of orthodontic aligners, the method further comprising performing the following for each digital design of the digital designs:
 extracting a plurality of characteristics of the associated respective orthodontic aligner from the digital design, the plurality of characteristics comprising at least one of geometrical characteristics, clinical characteristics or treatment related characteristics that are represented as structured or tabular data;   selecting a subset of the plurality of characteristics; and   generating an embedding for the digital design based on the subset of the plurality of characteristics, wherein the training dataset comprises embeddings for the digital designs for the plurality of orthodontic aligners.   
     
     
         5 . The method of  claim 1 , further comprising:
 periodically repeating the gathering and the training using digital designs for recently manufactured orthodontic aligners.   
     
     
         6 . The method of  claim 1 , wherein the machine learning model is a random forest classifier or a gradient boosted decision tree classifier. 
     
     
         7 . The method of  claim 1 , further comprising:
 processing additional digital designs for a plurality of additional orthodontic aligners using a first numerical simulation that simulates removal of an orthodontic aligner from a mold of a dental arch of a patient or a second numerical simulation that simulates loading around weak spots in the orthodontic aligner;   for each digital design of the additional digital designs for the plurality of additional orthodontic aligners, performing the following comprising:
 determining whether any probable points of damage for the orthodontic aligner are predicted based on a result of the processing; and 
 adding information about the probable points of damage as metadata to the digital design; and 
   adding the additional digital designs for the plurality of additional orthodontic aligners to the training dataset.   
     
     
         8 . A non-transitory computer readable medium comprising instructions that, when executed by a processing device, cause the processing device to perform operations comprising:
 gathering a training dataset comprising digital designs for a plurality of orthodontic aligners, wherein each digital design is associated with a respective orthodontic aligner of the plurality of orthodontic aligners, and wherein each digital design comprises metadata indicating whether the associated respective orthodontic aligner was damaged during manufacturing or subsequent use of the associated respective orthodontic aligner; and   training a machine learning model using the training dataset, wherein the machine learning model is trained to process data from a digital design for an orthodontic aligner and to output a probability that the orthodontic aligner associated with the digital design will be damaged during manufacturing or subsequent use of the orthodontic aligner.   
     
     
         9 . The non-transitory computer readable medium of  claim 8 , wherein one or more of the digital designs comprises metadata indicating a location of damage that occurred for the associated respective orthodontic aligner. 
     
     
         10 . The non-transitory computer readable medium of  claim 8 , wherein gathering the training dataset comprises:
 receiving the digital designs for the plurality of orthodontic aligners, wherein the digital designs for the plurality of orthodontic aligners comprise at least one of digital models of the plurality of orthodontic aligners or digital models of molds used to manufacture the plurality of orthodontic aligners;   receiving information indicating one or more of the plurality of orthodontic aligners that were damaged during manufacturing; and   for each digital design associated with an orthodontic aligner that was damaged during manufacturing, adding the metadata to the digital design indicating that the orthodontic aligner was damaged during manufacturing.   
     
     
         11 . The non-transitory computer readable medium of  claim 8 , wherein the digital designs for the plurality of orthodontic aligners comprise at least one of digital models of the plurality of orthodontic aligners or digital models of molds used to manufacture the plurality of orthodontic aligners, the operations further comprising performing the following for each digital design of the digital designs:
 extracting a plurality of characteristics of the associated respective orthodontic aligner from the digital design, the plurality of characteristics comprising at least one of geometrical characteristics, clinical characteristics or treatment related characteristics that are represented as structured or tabular data;   selecting a subset of the plurality of characteristics; and   generating an embedding for the digital design based on the subset of the plurality of characteristics, wherein the training dataset comprises embeddings for the digital designs for the plurality of orthodontic aligners.   
     
     
         12 . The non-transitory computer readable medium of  claim 8 , the operations further comprising:
 periodically repeating the gathering and the training using digital designs for recently manufactured orthodontic aligners.   
     
     
         13 . The non-transitory computer readable medium of  claim 8 , wherein the machine learning model is a random forest classifier or a gradient boosted decision tree classifier. 
     
     
         14 . The non-transitory computer readable medium of  claim 8 , the operations further comprising:
 processing additional digital designs for a plurality of additional orthodontic aligners using a first numerical simulation that simulates removal of an orthodontic aligner from a mold of a dental arch of a patient or a second numerical simulation that simulates loading around weak spots in the orthodontic aligner;   for each digital design of the additional digital designs for the plurality of additional orthodontic aligners, performing the following comprising:
 determining whether any probable points of damage for the orthodontic aligner are predicted based on a result of the processing; and 
 adding information about the probable points of damage as metadata to the digital design; and 
   adding the additional digital designs for the plurality of additional orthodontic aligners to the training dataset.   
     
     
         15 . A system comprising:
 a memory; and   a processing device operatively connected to the memory, the processing device to:
 gather a training dataset comprising digital designs for a plurality of orthodontic aligners, wherein each digital design is associated with a respective orthodontic aligner of the plurality of orthodontic aligners, and wherein each digital design comprises metadata indicating whether the associated respective orthodontic aligner was damaged during manufacturing or subsequent use of the associated respective orthodontic aligner; and 
 train a machine learning model using the training dataset, wherein the machine learning model is trained to process data from a digital design for an orthodontic aligner and to output a probability that the orthodontic aligner associated with the digital design will be damaged during manufacturing or subsequent use of the orthodontic aligner. 
   
     
     
         16 . The system of  claim 15 , wherein one or more of the digital designs comprises metadata indicating a location of damage that occurred for the associated respective orthodontic aligner. 
     
     
         17 . The system of  claim 15 , wherein gathering the training dataset comprises:
 receiving the digital designs for the plurality of orthodontic aligners, wherein the digital designs for the plurality of orthodontic aligners comprise at least one of digital models of the plurality of orthodontic aligners or digital models of molds used to manufacture the plurality of orthodontic aligners;   receiving information indicating one or more of the plurality of orthodontic aligners that were damaged during manufacturing; and   for each digital design associated with an orthodontic aligner that was damaged during manufacturing, adding the metadata to the digital design indicating that the orthodontic aligner was damaged during manufacturing.   
     
     
         18 . The system of  claim 15 , wherein the digital designs for the plurality of orthodontic aligners comprise at least one of digital models of the plurality of orthodontic aligners or digital models of molds used to manufacture the plurality of orthodontic aligners, and wherein the processing device is further to perform the following for each digital design of the digital designs:
 extract a plurality of characteristics of the associated respective orthodontic aligner from the digital design, the plurality of characteristics comprising at least one of geometrical characteristics, clinical characteristics or treatment related characteristics that are represented as structured or tabular data;   select a subset of the plurality of characteristics; and   generate an embedding for the digital design based on the subset of the plurality of characteristics, wherein the training dataset comprises embeddings for the digital designs for the plurality of orthodontic aligners.   
     
     
         19 . The system of  claim 15 , wherein the processing device is further to:
 periodically repeat the gathering and the training using digital designs for recently manufactured orthodontic aligners.   
     
     
         20 . The system of  claim 15 , wherein the processing device is further to:
 process additional digital designs for a plurality of additional orthodontic aligners using a first numerical simulation that simulates removal of an orthodontic aligner from a mold of a dental arch of a patient or a second numerical simulation that simulates loading around weak spots in the orthodontic aligner; and   for each digital design of the additional digital designs for the plurality of additional orthodontic aligners, perform the following comprising:
 determine whether any probable points of damage for the orthodontic aligner are predicted based on a result of the processing; and 
 add information about the probable points of damage as metadata to the digital design; and 
   adding the additional digital designs for the plurality of additional orthodontic aligners to the training dataset.