Dental image segmentation and registration with machine learning
A system and method are disclosed for representing and studying anatomy in the oral region such as parts of a subject's teeth and adjoining tissues. Types of inputs are used to form segmented outputs representing the teeth and can include segmented crown and root portions of the teeth. Machine learning methods are used for optimum and accurate results and to generate data objects corresponding to respective anatomical features of the subject.
1. A method for automated processing of a three-dimensional (3D) dental image containing one or more bone structures and teeth within a teeth volume in a processor-based machine using machine learning steps, so as to provide results for treatment planning and follow-up, the method comprising:
acquiring CT image data of said bone structures and teeth;
acquiring intra-oral scan (IOS) data of said teeth;
forming said three-dimensional dental image from both the acquired CT image data and the IOS data;
receiving the three-dimensional dental image including an image of the one or more bone structures and the teeth within said teeth volume;
performing a sequence of automated segmentation of anatomical features of the received dental image, including by tiling of configurably sized and overlapped tile regions, to produce segmented anatomical features, the segmentation based on machine learning from previously-analyzed dental images;
creating a segmented data file corresponding to one or more of the segmented anatomical features;
formatting the segmented data file so that the one or more segmented anatomical features are viewable by a user on a display device; and
providing the segmented data file to the user,
wherein the sequence of automated segmentation comprises:
segmenting all of said one or more bone structures collectively as a first class, creating a collective segmented data file;
segmenting all of said teeth collectively as a single class, creating a total segmented teeth volume data file that represents a total segmented teeth volume;
predicting multiple predicted 3D bounding boxes for each tooth of a patient to provide a respective estimated location for each tooth within the total segmented teeth volume;
collapsing, for each tooth, the multiple predicted 3D bounding boxes into a respective single 3D bounding box;
segmenting each tooth separately using the respective single 3D bounding box; and
generating a segmented tooth data file for each tooth.
2. The method of claim 1 , wherein receiving the 3D dental image comprises receiving the 3D dental image of the one or more bone structures over a cloud-based communication network.
3. The method of claim 1 , wherein the sequence of automated segmentation comprises using a convolutional neural network (CNN) process executed in the processor-based machine, and employing information from the previously-analyzed dental images, so as to segment the anatomical features in the 3D dental image.
4. The method of claim 1 , wherein the 3D dental image comprises data from a dental cone beam computerized tomography (CBCT) device.
5. The method of claim 1 , wherein the 3D dental image comprises data from one or more of a dental cone beam computerized tomography (CBCT) device, an intra-oral data acquisition device, and/or a scan of a denture model.
6. The method of claim 1 , further comprising in addition to said anatomical features at least one of: roots, crowns, maxillae, mandibles and nerve canals.
7. The method of claim 1 , wherein said teeth are segmented individually after said teeth are segmented collectively.
8. The method of claim 1 , wherein said individual tooth data files of individual teeth are merged into a total tooth volume representation that includes the total tooth volume data file.
9. The method of claim 1 , wherein the sequence of automated segmentation further comprises extracting features of interest from the segmented teeth data file.