Automated 3D root shape prediction using deep learning methods

A computer-implemented method for training a 3D deep learning neural network to generate a prediction of 3D root shape comprising:

• a computer receiving training data, the training data including clinical 3D data comprising voxel representations of crowns and associated roots or on the basis of voxel representations of crowns and associated teeth wherein at least part of the voxel representations being derived from segmented 3D X-ray data;
• and/or, derived from optically scanned complete teeth;
• offering a voxel representation of a crown to the input of the 3D deep neural network and the 3D deep neural network generating a voxel representation of a predicted root;
• optimizing values of one or more network parameters of the 3D deep neural network by minimizing a loss function representing a deviation between the voxel representation of a predicted root or a predicted complete tooth and the voxel representation of a root or a complete tooth that is associated with the voxel representation of the crown that was offered to the input of the 3D deep neural network;
• and,
• storing the optimized values in a computer readable storage medium, the optimized values defining one or more network parameters of a trained neural network configured to, when provided with a voxel representation of a crown, predict an anatomically accurate voxel representation of a root corresponding to the crown or a voxel representation of a complete tooth.

A computer system adapted to automatically predict a 3D root shape comprising:

• a computer readable storage medium having computer readable program code embodied therewith, the program code including a pre-processing algorithm and at least a trained 3D deep neural network;
• and a processor 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:
• a pre-processor receiving 3D data defined in a voxel space, the voxel space defining a 3D volume, the voxel space comprising a voxel representation of a crown, and processing the voxel representation such that it is in a scale, position and orientation that corresponds to voxel representations that are used to train a first 3D deep neural network;
• the pre-processor providing the voxel representation of the crown to the input of the first 3D deep neural network, the first 3D deep neural network being trained on the basis of a training set of pre-processed clinical 3D data defining 3D representations of real teeth, the trained deep neural network being configured to predict an anatomically accurate voxel representation of a root corresponding to the crown or a voxel representation of a complete tooth;
• the first 3D deep neural network generating a voxel representation of a predicted root or of a complete tooth comprising the predicted root on the basis of the voxel representation of the crown, wherein the generation of the voxel representation of the predicted root or the complete tooth includes:
• determining voxel activations for voxels in a voxel space of the output of the first 3D deep learning network, each voxel activation representing a probability measure defining the probability that a voxel is part of the root or of the complete tooth;
• and,
• determining whether a voxel activation is part of the root or of the complete tooth by comparing the voxel activation with a voxel activation threshold value.