System and methods for restorative dentistry treatment planning using adversarial learning

1. A method for diagnosis of dental pathologies comprising:

receiving, by a computer system, a plurality of training data entries, each training data entry of the plurality of training data entries including a dental image, one or more tooth labels of one or more teeth represented in the dental image, and a target pathology label; and

for each training data entry of the plurality of training data entries:

processing, by the computer system, the dental image and the one or more tooth labels of the each training data entry using a generator machine learning model to obtain a synthetic pathology label;

comparing, by the computer system, the synthetic pathology label to the target pathology label of the each training data entry; and

updating, by the computer system, the generator machine learning model according to the comparing of the synthetic pathology label to the target pathology label of the each training data entry;

wherein the dental image of each training data entry of the plurality of training data entries is an image of dental anatomy according to an imaging modality selected from the group consisting of full mouth series X-rays, dental cone beam computed tomography (CBCT), cephalometric X-ray, intra-oral optical image, panoramic dental X-ray, dental magnetic resonance imaging (MRI) image, dental light detection and ranging (LIDAR) image.

2. The method of claim 1 , further comprising:

process, by the computer system, the synthetic pathology label and a real pathology label from a repository of pathology labels with a discriminator machine learning model to obtain a realism estimate; and

updating, by the computer system, the generator machine learning model and the discriminator machine learning model according to the realism estimate.

3. The method of claim 1 , wherein processing the dental image and the one or more tooth labels of the each training data entry using a generator machine learning model to obtain the synthetic pathology label comprises:

concatenating, by the computer system, a style matrix associated with a human generator of the target pathology label with outputs of a first intermediate layer of the generator machine learning model to obtain a concatenated output; and

inputting, by the computer system, the concatenated output to a second intermediate layer following the first intermediate layer in the generator machine learning model.

4. The method of claim 3 , wherein the generator machine learning model comprises an encoder convolution neural network (CNN) coupled to a decoder CNN, the first intermediate layer being a last layer of the decoder CNN.

5. The method of claim 1 , wherein each training data entry of the plurality of training data entries further includes a target surface indicating a tooth surface on which a pathology represented in the target pathology label of the each training data entry is located, the method further comprising, for each training data entry of the plurality of training data entries:

processing, by the computer system, the dental image, one or more tooth labels, and the target pathology label of the each training data entry using a surface machine learning model to obtain a surface estimate;

comparing, by the computer system, the surface estimate to the target surface of the each training data entry; and

updating, by the computer system, the surface machine learning model according to the comparing of the surface estimate to the target surface of the each training data entry.

6. The method of claim 5 , wherein each training data entry of the plurality of training data entries further includes a restoration label indicating one or more restorations represented in the dental image of the each training data entry, the method further comprising, for each training data entry of the plurality of training data entries:

processing, by the computer system, the dental image, one or more tooth labels, the target pathology label, and the restoration label of the each training data entry using the surface machine learning model to obtain the surface estimate.

7. The method of claim 5 , wherein the target surface of each training data entry of the plurality of training data entries is one of mesial, occlusal, distal, buccal, and lingual.

8. The method of claim 1 , wherein each training data entry of the plurality of training data entries further includes a treatment label indicating an appropriate treatment for a pathology represented in the target pathology label of the each training data entry, the method further comprising, for each training data entry of the plurality of training data entries:

processing, by the computer system, the dental image, one or more tooth labels, and the target pathology label of the each training data entry using a treatment machine learning model to obtain a treatment estimate;

comparing, by the computer system, the treatment estimate to the treatment label of the each training data entry; and

updating, by the computer system, the treatment machine learning model according to the comparing of the treatment estimate to the treatment label of the each training data entry.

9. The method of claim 8 , wherein the treatment label designates any of a filling, a crown, extraction, implant, bridge, and root canal therapy.

10. The method of claim 8 , wherein each training data entry of the plurality of training data entries further includes a restoration label labeling one or more restorations represented in the dental image of the each training data entry and a defect label indicating one or more defects in the one or more restorations represented in the dental image of the each training data entry;

wherein the method further comprises, for each training data entry of the plurality of training data entries;

processing, by the computer system, the dental image, one or more tooth labels, the target pathology label, and the restoration label of the each training data entry using a defect machine learning model to obtain a defect estimate;

comparing, by the computer system, the defect estimate to the defect label of the each training data entry; and

updating, by the computer system, the defect machine learning model according to the comparing of the defect estimate to the defect label of the each training data entry.

11. A non-transitory computer-readable medium storing executable instructions that, when executed by a processing device, cause the processing device to:

receive a plurality of training data entries, each training data entry of the plurality of training data entries including a dental image, one or more tooth labels of one or more teeth represented in the dental image, and a target pathology label; and

for each training data entry of the plurality of training data entries:

process the dental image and the one or more tooth labels of the each training data entry using a generator machine learning model to obtain a synthetic pathology label;

compare the synthetic pathology label to the target pathology label of the each training data entry; and

update the generator machine learning model according to the comparing of the synthetic pathology label to the target pathology label of the each training data entry;

wherein the dental image of each training data entry of the plurality of training data entries is an image of dental anatomy according to an imaging modality selected from the group consisting of full mouth series X-rays, dental cone beam computed tomography (CBCT), cephalometric X-ray, intra-oral optical image, panoramic dental X-ray, dental magnetic resonance imaging (MRI) image, dental light detection and ranging (LIDAR) image.

12. The non-transitory computer-readable medium of claim 11 , wherein the executable instructions, when executed by the processing device, further cause the processing device to:

process the synthetic pathology label and a real pathology label from a repository of pathology labels with a discriminator machine learning model to obtain a realism estimate; and

update the generator machine learning model and the discriminator machine learning model according to the realism estimate.

13. The non-transitory computer-readable medium of claim 11 , wherein the executable instructions, when executed by the processing device, further cause the processing device to process the dental image and the one or more tooth labels of the each training data entry using a generator machine learning model to obtain the synthetic pathology label by:

concatenating a style matrix associated with a human generator of the target pathology label of the each training data entry with outputs of a first intermediate layer of the generator machine learning model to obtain a concatenated output; and

inputting the concatenated output to a second intermediate layer following the first intermediate layer in the generator machine learning model.

14. The non-transitory computer-readable medium of claim 13 , wherein the generator machine learning model comprises an encoder convolution neural network (CNN) coupled to a decoder CNN, the first intermediate layer being a last layer of the decoder.

15. The non-transitory computer-readable medium of claim 11 , wherein each training data entry of the plurality of training data entries further includes a target surface indicating a tooth surface on which a pathology represented in the target pathology label of the each training data entry is located; and

wherein the executable instructions, when executed by the processing device, further cause the processing device to, for each training data entry of the plurality of training data entries:

process the dental image, one or more tooth labels, and the target pathology label of the each training data entry using a surface machine learning model to obtain a surface estimate;

compare the surface estimate to the target surface of the each training data entry; and

update the surface machine learning model according to the comparing of the surface estimate to the target surface of the each training data entry.

16. The non-transitory computer-readable medium of claim 15 , wherein each training data entry of the plurality of training data entries further includes a restoration label labeling one or more restorations represented in the dental image of the each training data entry; and

wherein the executable instructions, when executed by the processing device, further cause the processing device to, for each training data entry of the plurality of training data entries:

process the dental image, one or more tooth labels, the target pathology label, and the restoration label of the each training data entry using the surface machine learning model to obtain the surface estimate.

17. The non-transitory computer-readable medium of claim 15 , wherein the surface label of each training data entry of the plurality of training data entries is one of mesial, occlusal, distal, buccal, and lingual.

18. The non-transitory computer-readable medium of claim 11 , wherein each training data entry of the plurality of training data entries further includes a treatment label indicating an appropriate treatment for a pathology represented in the target pathology label of the each training data entry;

wherein the executable instructions, when executed by the processing device, further cause the processing device to, for each training data entry of the plurality of training data entries:

process the dental image, one or more tooth labels, and the target pathology label of the each training data entry using a treatment machine learning model to obtain a treatment estimate;

compare the treatment estimate to the treatment label of the each training data entry; and

update the treatment machine learning model according to the comparing of the treatment estimate to the treatment label of the each training data entry.

19. The non-transitory computer-readable medium of claim 18 , wherein the treatment label designates any of a filling, a crown, an extraction, a bridge, an implant, and root canal therapy.

20. The non-transitory computer-readable medium of claim 18 , wherein each training data entry of the plurality of training data entries further includes a restoration label labeling one or more restorations represented in the dental image of the each training data entry and a defect label labeling one or more defects in the one or more restorations represented in the dental image of the each training data entry;

wherein the executable instructions, when executed by the processing device, further cause the processing device to, for each training data entry of the plurality of training data entries:

process the dental image, one or more tooth labels, the target pathology label, and the restoration label of the each training data entry using a defect machine learning model to obtain a defect estimate;

compare the defect estimate to the defect label of the each training data entry; and

update the defect machine learning model according to the comparing of the defect estimate to the defect label of the each training data entry.