Dental image processing protocol for dental aligners

Abstract

The present disclosure relates to a method of pixel-based classification of medical images. The method includes training a neural network to perform the pixel-based classification, the training comprising performing a first training on a classifier using a first training set of medical images, pixels of the first training set of medical images being manually labeled as a biological structure type, applying the classifier after the first training to a second training set of medical, identifying and correcting incorrectly classified pixels of the second training set of medical images, and performing a second training on the classifier after the first training using the manually labeled first training set and the correctly classified second training set, and applying the classifier after the second training to the medical images to perform the pixel-based classification.

1 . A method of pixel-based classification of medical images, comprising:
 training a neural network to perform the pixel-based classification of the medical images, the training comprising
 performing a first training on a classifier using a first training set of medical images, pixels of the first training set of medical images being manually labeled, 
 applying the classifier after the first training to a second training set of medical images to classify pixels of the second training set of medical images, 
 identifying and correcting incorrectly classified pixels of the second training set of medical images, and 
 performing a second training on the classifier after the first training using the manually labeled first training set of medical images and the correctly classified second training set of medical images; and 
   applying the classifier after the second training to the medical images to perform the pixel-based classification, the pixel-based classification of the medical images including assigning pixels of each medical image to a biological structure type.   
     
     
         2 . The method of  claim 1 , wherein the training the neural network further comprises
 generating a third training set of medical images that includes the manually labeled first training set of medical images and the correctly classified second training set of medical images,   allocating each medical image of the third training set of medical images to one of a first subset of the third training set of medical images or a second subset of the third training set of medical images, and   performing a third training on the classifier after the second training using the first subset of the third training set of medical images.   
     
     
         3 . The method of  claim 2 , wherein the training the neural network further comprises
 applying the classifier after the third training to the second subset of the third training set of medical images,   identifying images of the second subset of the third training set of medical images that are incorrectly classified by the classifier, the identified images having a pixel classification error rate above a pixel classification threshold,   reallocating a number of the identified incorrectly classified images of the second subset of the third training set of medical images to the first subset of the third training set of medical images, and reallocating a number of images, corresponding to the number of images reallocated to the first subset of the third training set of medical images, from the first subset of the third training set of medical images to the second subset of the third training set of medical images, and   training the classifier after the third training using the first subset of the third training set of medical images including the reallocated images of the second subset of the third training set of medical images.   
     
     
         4 . The method of  claim 3 , wherein a cycle including the applying, the identifying, the reallocating, and the training is performed two or more times. 
     
     
         5 . The method of  claim 2 , wherein the training the neural network further comprises
 applying the classifier after the third training to the second subset of the third training set of medical images,   identifying images of the second subset of the third training set of medical images that are incorrectly classified, the identified images having a pixel classification error rate above a pixel classification threshold,   reallocating images between the first subset of the third training set of medical images and the second subset of the third training set of medical images based on a comparison of a quantity of the identified incorrectly classified images and an identification threshold, and   training the classifier after the third training using the first subset of the third training set of medical images including the reallocated images the third training set of medical images.   
     
     
         6 . The method of  claim 5 , wherein a cycle including the applying, the identifying, the reallocating, and the training is performed two or more times. 
     
     
         7 . The method of  claim 1 , wherein the neural network is a fully convolutional neural network. 
     
     
         8 . The method of claim I , wherein the biological structure type is one of a hard tissue or a soft tissue. 
     
     
         9 . An apparatus for pixel-based classification of medical images, comprising:
 processing circuitry configured to   train a neural network to perform the pixel-based classification of the medical images, the training comprising
 performing a first training on a classifier using a first training set of medical images, pixels of the first training set of medical images being manually labeled, 
 applying the classifier after the first training to a second training set of medical images to classify pixels of the second training set of medical images, 
 identifying and correcting incorrectly classified pixels of the second training set of medical images, and 
 performing a second training on the classifier after the first training using the manually labeled first training set of medical images and the correctly classified second training set of medical images, and 
   apply the classifier after the second training to the medical images to perform the pixel-based classification, the pixel-based classification of the medical images including assigning pixels of each medical image to a biological structure type.   
     
     
         10 . The apparatus of  claim 9 , wherein the processing circuitry is further configured to train the neural network by
 generating a third training set of medical images that includes the manually labeled first training set of medical images and the correctly classified second training set of medical images,   allocating each medical image of the third training set of medical images to one of a first subset of the third training set of medical images or a second subset of the third training set of medical images, and   performing a third training on the classifier after the second training using the first subset of the third training set of medical images.   
     
     
         11 . The apparatus of  claim 10 , wherein the processing circuitry is further configured to train the neural network by
 applying the classifier after the third training to the second subset of the third training set of medical images,   identifying images of the second subset of the third training set of medical images that are incorrectly classified by the classifier, the identified images having a pixel classification error rate above a pixel classification threshold,   reallocating a number of the identified incorrectly classified images of the second subset of the third training set of medical images to the first subset of the third training set of medical images, and reallocating a number of images, corresponding to the number of images reallocated to the first subset of the third training set of medical images, from the first subset of the third training set of medical images to the second subset of the third training set of medical images, and   training the classifier after the third training using the first subset of the third training set of medical images including the reallocated images of the second subset of the third training set of medical images.   
     
     
         12 . The apparatus of  claim 11 , wherein a cycle including the applying, the identifying, the reallocating, and the training is performed two or more times. 
     
     
         13 . The apparatus of  claim 10 , wherein the processing circuitry is further configured to train the neural network by
 applying the classifier after the third training to the second subset of the third training set of medical images,   identifying images of the second subset of the third training set of medical images that are incorrectly classified, the identified images having a pixel classification error rate above a pixel classification threshold,   reallocating images between the first subset of the third training set of medical images and the second subset of the third training set of medical images based on a comparison of a quantity of the identified incorrectly classified images and an identification threshold, and   training the classifier after the third training using the first subset of the third training set of medical images including the reallocated images the third training set of medical images.   
     
     
         14 . The apparatus of  claim 13 , wherein a cycle including the applying, the identifying, the reallocating, and the training is performed two or more times. 
     
     
         15 . A non-transitory computer-readable storage medium storing computer-readable instructions that, when executed by a computer, cause the computer to perform a method of pixel-based classification of medical images, the method comprising:
 training a neural network to perform the pixel-based classification of the medical images, the training comprising   performing a first training on a classifier using a first training set of medical images, pixels of the first training set of medical images being manually labeled,   applying the classifier after the first training to a second training set of medical images to classify pixels of the second training set of medical images,   identifying and correcting incorrectly classified pixels of the second training set of medical images, and   performing a second training on the classifier after the first training using the manually labeled first training set of medical images and the correctly classified second training set of medical images; and   applying the classifier after the second training to the medical images to perform the pixel-based classification, the pixel-based classification of the medical images including assigning pixels of each medical image to a biological structure type.   
     
     
         16 . The non-transitory computer-readable storage medium of  claim 15 , wherein the training the neural network further comprises
 generating a third training set of medical images that includes the manually labeled first training set of medical images and the correctly classified second training set of medical images,   allocating each medical image of the third training set of medical images to one of a first subset of the third training set of medical images or a second subset of the third training set of medical images, and   performing a third training on the classifier after the second training using the first subset of the third training set of medical images.   
     
     
         17 . The non-transitory computer-readable storage medium of  claim 16 , wherein the training the neural network further comprises
 applying the classifier after the third training to the second subset of the third training set of medical images, p 1  identifying images of the second subset of the third training set of medical images that are incorrectly classified by the classifier, the identified images having a pixel classification error rate above a pixel classification threshold,   reallocating a number of the identified incorrectly classified images of the second subset of the third training set of medical images to the first subset of the third training set of medical images, and reallocating a number of images, corresponding to the number of images reallocated to the first subset of the third training set of medical images, from the first subset of the third training set of medical images to the second subset of the third training set of medical images, and   training the classifier after the third training using the first subset of the third training set of medical images including the reallocated images of the second subset of the third training set of medical images.   
     
     
         18 . The non-transitory computer-readable storage medium of  claim 17 , wherein a cycle including the applying, the identifying, the reallocating, and the training is performed two or more times. 
     
     
         19 . The non-transitory computer-readable storage medium of  claim 16 , wherein the training the neural network further comprises
 applying the classifier after the third training to the second subset of the third training set of medical images,   identifying images of the second subset of the third training set of medical images that are incorrectly classified, the identified images having a pixel classification error rate above a pixel classification threshold,   reallocating images between the first subset of the third training set of medical images and the second subset of the third training set of medical images based on a comparison of a quantity of the identified incorrectly classified images and an identification threshold, and   training the classifier after the third training using the first subset of the third training set of medical images including the reallocated images the third training set of medical images.   
     
     
         20 . The non-transitory computer-readable storage medium of  claim 19 , wherein a cycle including the applying, the identifying, the reallocating, and the training is performed two or more times.