Diagnosing tooth decay using artificial intelligence

Abstract

A device for diagnosing a dental condition is disclosed. The device captures image data representative of a patient's teeth based on data obtained from a hardware device that scans the teeth. The device inputs the image data into a first supervised machine learning model and receives as output a plurality of biomarkers from the first supervised machine learning model, each biomarker corresponding to a different location on the tooth. The device inputs the plurality of biomarkers into a second supervised machine learning model and receives as output a diagnosis of the dental condition from the second supervised machine learning model.

Description

[Background technology]

[0001] (CROSS-REFERENCE TO RELATED APPLICATIONS) This application claims the benefit of U.S. Provisional Patent Application No. 63 / 219,216, filed December 17, 2021, the contents of which are incorporated herein by reference in their entirety.

[0002] (background) The risk of tooth decay (dental caries) is widespread throughout society, and every individual is at risk, especially due to a diet rich in carbohydrates and ultra-processed foods. Dental caries is a major cause of disability, pain, and healthcare costs. Effective treatments exist to prevent and reverse tooth decay, especially when it is detected in its early stages, when it is still asymptomatic. However, adherence to early diagnosis is suboptimal, leading to unnecessary caries progression, when the standard of care is dental examinations by dentists and, optionally, x-rays. While standard-of-care dental examinations by dentists and, optionally, x-rays, are effective in diagnosing tooth decay early, scaling this is difficult due to the need for, and shortage of, either highly skilled professionals (i.e., dentists) or highly skilled x-ray technicians, with requirements that vary widely by state, as well as exposure to x-rays, which can have long-term health effects, and the need for expert evaluation via some forms of telehealth, which leads to relatively high costs and reduces access. Summary of the Invention [Means for solving the problem]

[0003] (Abstract) Therefore, autonomous dental screening artificial intelligence (AI) technologies that a) do not require X-rays and b) do not require specialized knowledge to evaluate images would be highly beneficial for increasing access, reducing costs, and thereby avoiding permanent disability and costs. Specifically, deep penetration optical coherence tomography (DPOCT) is a non-radioactive optical technology based on low-coherence light interferograms that penetrate up to 4-5 mm into dental elements, including molars, allowing all surfaces of each element to be imaged, including any cavities on all surfaces. In particular, it allows the so-called interdental cavities between two elements to be imaged, which is where 60% of cavities occur and is inaccessible to visual inspection. Autonomous AI is a technology that enables real-time, point-of-care diagnosis of cavities from multi-planar DPOCT images.

[0004] Because DPOCT cannot penetrate the entire tooth, especially large molars, imaging of each accessible surface may be required. This can be achieved through multiplanar DPOCT, using a small probe or multiple probes covering at least the facial, lingual, and occlusal surfaces of the molar row, illuminating each plane in the so-called swept-source or domain OCT interference pattern with low-coherence light and collecting the reflected low-coherence light into a DPOCT device that calculates an interference image using a Michelson interferometry approach for each plane in the interference pattern. It obtains one or more B-scans (two-lateral scans) from these three planes coaxial with the probe, and as the probe is advanced back and forth manually or servo-assisted, a multiplanar 3D PDOCT volume is collected. In one example, probes are applied to the other three rows of molars to calculate multiplanar images of all molar elements in the patient.

[0005] The autonomous AI has a multi-aspect DPOCT volume as input and outputs either a likelihood of cavities in the volume or a diagnostic dichotomous or multi-level presence / absence or severity of cavities in the volume. Other diagnostically relevant outputs may also be produced. The AI ​​can be constructed as a hybrid partially independent biomarker multi-detector AI with a fusion stage, or as a multi-image based convolutional neural network using deep learning, shallow learning, recurrent networks, or any combination of such AI designs, as are known to those skilled in the art. An exemplary AI of this approach is disclosed in further detail in commonly owned U.S. Patent No. 10,115,194, issued October 30, 2018, the disclosure of which is hereby incorporated by reference herein in its entirety. Augmentation and transfer learning approaches, including unsupervised transfer learning, can be used as well. DPOCT training images labeled with the presence or location of cavities or likelihood or probability can be used to train such an AI system, or in the case of a multi-detector approach, labeled image samples can be used to train the detector, or a mathematical description of the biomarkers can be used to design the detector. DPOCT is merely exemplary, and whenever DPOCT or any other imaging technique is mentioned, any other form of imaging that captures the functionality described herein with respect to DPOCT (e.g., using low-coherence light, sound, vibration, or any other waveform-generating technique using interferometric techniques) equally applies. Furthermore, any low-coherence tomography (LCT) technique also equally applies when DPOCT is mentioned herein, including ultraviolet, infrared, dental penetration or deep penetration optical coherence tomography, microwave, visible light LCT, ultrasound imaging techniques, and the like. The present invention provides, for example, the following. (Item 1) 1. A method for diagnosing a dental condition, the method comprising: capturing image data indicative of a patient's teeth based on data obtained from a hardware device scanning the teeth; inputting the image data into a first supervised machine learning model; receiving a plurality of biomarkers as output from the first supervised machine learning model, each biomarker corresponding to a different location on the tooth; inputting the plurality of biomarkers into a second supervised machine learning model; receiving as output from the second supervised machine learning model a diagnosis of the dental condition; A method comprising: (Item 2) 2. The method of claim 1, wherein the data obtained from the hardware device scanning the teeth comprises deep penetration optical coherence tomography (DPOCT) data, and the image data comprises an intensity map reflecting one or more optical properties of the teeth based on the DPOCT data. (Item 3) 3. The method of claim 2, wherein the first supervised machine learning model is trained to detect intensity variations between regions in the intensity map and to output a respective classification for each different location of the tooth based on the intensity variations, each classification forming a biomarker. (Item 4) Item 4. The method of item 3, wherein the first supervised machine learning model is additionally trained to take a color image of the tooth as input, and to output biomarkers based on both the image and the intensity map, and the output of the first supervised machine learning model excludes color data from the color image. (Item 5) 10. The method of claim 1, further comprising accessing a historical biomarker for the tooth, wherein the historical biomarker is input into the second supervised machine learning model along with the plurality of biomarkers, and wherein the second supervised machine learning model outputs the diagnosis based on both the historical biomarker and the plurality of biomarkers. (Item 6) 6. The method of claim 5, wherein inputting the historical biomarker along with the plurality of biomarkers into the second supervised machine learning model comprises calculating an intensity difference for each biomarker of the plurality of biomarkers relative to the historical biomarker, and inputting each intensity difference into the second supervised machine learning model. (Item 7) 2. The method of claim 1, wherein receiving a diagnosis of the dental condition from the second supervised machine learning model as output comprises receiving a plurality of diagnoses, each diagnosis of the plurality of diagnoses corresponding to a different one of the plurality of biomarkers. (Item 8) 1. A non-transitory computer-readable medium comprising a memory having instructions encoded therein that, when executed, cause one or more processors to perform operations for diagnosing a dental condition, the instructions comprising: capturing image data indicative of a patient's teeth based on data obtained from a hardware device scanning the teeth; inputting the image data into a first supervised machine learning model; receiving a plurality of biomarkers as output from the first supervised machine learning model, each biomarker corresponding to a different location on the tooth; inputting the plurality of biomarkers into a second supervised machine learning model; receiving as output from the second supervised machine learning model a diagnosis of the dental condition; A non-transitory computer-readable medium comprising instructions for performing (Item 9) 9. The non-transitory computer-readable medium of claim 8, wherein the data obtained from the hardware device scanning the teeth comprises deep penetration optical coherence tomography (DPOCT) data, and the image data comprises an intensity map reflecting one or more optical properties of the teeth based on the DPOCT data. (Item 10) Item 10. The non-transitory computer-readable medium of item 9, wherein the first supervised machine learning model is trained to detect intensity variations between regions in the intensity map and to output a respective classification for each different location of the tooth based on the intensity variations, wherein each respective classification forms a biomarker. (Item 11) Item 11. The non-transitory computer-readable medium of item 10, wherein the first supervised machine learning model is additionally trained to take a color image of the tooth as an input, the first supervised machine learning model being additionally trained to output biomarkers based on both the image and the intensity map, and the output of the first supervised machine learning model excluding color data from the color image. (Item 12) 9. The non-transitory computer-readable medium of claim 8, wherein the method further comprises accessing historical biomarkers for the teeth, wherein the historical biomarkers are input into the second supervised machine learning model along with the plurality of biomarkers, and wherein the second supervised machine learning model outputs the diagnosis based on both the historical biomarkers and the plurality of biomarkers. (Item 13) 13. The non-transitory computer-readable medium of claim 12, wherein the instructions for inputting the historical biomarker along with the plurality of biomarkers into the second supervised machine learning model comprise instructions for calculating an intensity difference for each biomarker in the plurality of biomarkers relative to the historical biomarker and inputting each intensity difference into the second supervised machine learning model. (Item 14) 9. The non-transitory computer-readable medium of claim 8, wherein the instructions to receive as output the diagnosis of the dental condition from the second supervised machine learning model comprise instructions to receive a plurality of diagnoses, each diagnosis of the plurality of diagnoses corresponding to a different one of the plurality of biomarkers. (Item 15) 1. A system comprising: a memory having instructions encoded therein for diagnosing a dental condition; with one or more processors wherein the one or more processors, when executing the instructions, perform operations, the operations including: capturing image data indicative of a patient's teeth based on data obtained from a hardware device scanning the teeth; inputting the image data into a first supervised machine learning model; receiving a plurality of biomarkers as output from the first supervised machine learning model, each biomarker corresponding to a different location on the tooth; inputting the plurality of biomarkers into a second supervised machine learning model; receiving as output from the second supervised machine learning model a diagnosis of the dental condition; A system comprising: (Item 16) Item 16. The system of item 15, wherein the data obtained from the hardware device scanning the tooth comprises deep penetration optical coherence tomography (DPOCT) data, and the image data comprises an intensity map reflecting one or more optical properties of the tooth based on the DPOCT data. (Item 17) Item 17. The system of item 16, wherein the first supervised machine learning model is trained to detect intensity variations between regions in the intensity map and to output a respective classification for each different location of the tooth based on the intensity variations, each classification forming a biomarker. (Item 18) Item 18. The system of item 17, wherein the first supervised machine learning model is additionally trained to take a color image of the tooth as input, the first supervised machine learning model being additionally trained to output biomarkers based on both the image and the intensity map, and the output of the first supervised machine learning model excluding color data from the color image. (Item 19) 16. The system of claim 15, wherein the operations further comprise accessing historical biomarkers for the teeth, the historical biomarkers being input into the second supervised machine learning model along with the plurality of biomarkers, and the second supervised machine learning model outputting the diagnosis based on both the historical biomarkers and the plurality of biomarkers. (Item 20) 20. The system of claim 19, wherein inputting a historical biomarker along with the plurality of biomarkers into the second supervised machine learning model comprises calculating an intensity difference for each biomarker of the plurality of biomarkers relative to the historical biomarker and inputting each intensity difference into the second supervised machine learning model. [Brief explanation of the drawings]

[0006] [Figure 1] FIG. 1 illustrates one embodiment of a system environment for implementing a caries determination tool.

[0007] [Figure 2] FIG. 2 illustrates one embodiment of exemplary modules and databases used by a treatment decision tool in using artificial intelligence to directly output a treatment.

[0008] [Figure 3] FIG. 3 shows one embodiment of further details of the sub-modules used by the caries decision module of the treatment decision tool.

[0009] [Figure 4] FIG. 4 shows one embodiment of a DPOCT intensity map.

[0010] [Figure 5]FIG. 5 illustrates an exemplary process for determining a diagnosis of a dental condition using a supervised machine learning approach.

[0011] The figures depict various embodiments of the present invention for illustrative purposes only. Those skilled in the art will readily recognize from the following discussion that alternative embodiments of the structures and methods illustrated herein may be used without departing from the inventive principles described herein. DETAILED DESCRIPTION OF THE INVENTION

[0012] (Detailed explanation) (a) Overview Figure 1 illustrates one embodiment of a system environment for implementing a caries determination tool. As depicted in Figure 1, environment 100 includes imaging equipment 110, network 120, caries determination tool 130, and patient data 140. The elements of environment 100 are merely exemplary, and fewer or more elements may be incorporated into environment 100 to achieve the functionality disclosed herein.

[0013] The imaging instrument 110 can be any instrument that captures images of a patient's teeth. The imaging instrument 110 can include a camera sensor, a DPOCT sensor, or any other long-wavelength sensor capable of capturing indications of dental problems and caries. The imaging instrument 110 can acquire credentials and other information about the patient. These can be entered by a technician through a user interface or directly by the patient. That is, the imaging instrument 110 can include a keyboard or touchscreen interface for entering data about the patient, as well as one or more imaging components that capture images. The imaging components can be any known imaging component (e.g., a slider that is slid over and around the teeth by the technician). The imaging instrument 110 can capture images and prompts for patient information based on running software controlled or otherwise distributed by the caries determination tool 130. Because the imaging device 110 uses non-X-ray technology, it can be operated by individuals who are not qualified to operate X-ray equipment (e.g., technicians with minimal training and a high school education), thus enabling scalability of caries detection.

[0014] The imaging instrument 110 may use multiplanar OCT technology in addition to a DPOCT sensor. For example, because DPOCT cannot fully penetrate some teeth, multiplanar OCT can be used to capture images showing caries that DPOCT may not be able to capture. The imaging instrument 110 may use multiplanar OCT to simultaneously or sequentially measure captured images (e.g., 2D images) and convert these images into 3D or other multi-volume measurements of any parameterization of the caries (or defect) at any given point on the tooth. The AI ​​disclosed herein can process multiple 3D or other multi-volume images captured together in this manner. For example, if multiple teeth or multiple sections of teeth are scanned together, the 3D or other multi-volume image may be a capture of those multiple teeth and / or multiple sections of teeth, and may be processed taking into account each other if the information in adjacent volumes constrains each other (e.g., by occlusion).

[0015] The network 120 may be any data network capable of transmitting data communications between the client device 110 and the treatment decision tool 130. The network 120 may be, for example, the Internet, a local area network, a wide area network, or any other network.

[0016] Cavity determination tool 130 receives and processes images from imaging instrument 110. Cavity determination tool 130 may also receive other patient information from imaging instrument 110 and may retrieve information about the patient from patient data 140. Based on some or all of the received information, caries determination tool 130 may output a patient status. Further details of the operation of caries determination tool 130 are discussed below with reference to FIG. 2. The operation of treatment decision tool 130 may be instantiated in whole or in part on imaging instrument 110 (e.g., through an application installed on imaging instrument 110 or accessed by imaging instrument 110 through a browser).

[0017] Patient data 140 is a database that stores records of data for one or more patients. Patient data 140 can be hospital records, patient personal records, doctor's notes, and the like. Patient data 140 can be co-located with imaging instrument 110 and / or caries determination tool 130. In an exemplary embodiment, patient data 140 includes profile information about the patient (e.g., age, gender, race, etc.), previous images (if any) of the patient's teeth captured by imaging instrument 110, and previous conditions (if any) detected by caries determination tool 130. Patient data 140 can be integrated with a dental clinic, where a dentist can indicate treatment (e.g., cavity filling) or a decision that no treatment was necessary (e.g., a false positive) in a consultation prompted by a condition identified by caries determination tool 130. Patient data 140 can be used as a baseline to detect caries progression as images of the patient's teeth are captured over time. The patient data 140 may be used to retrain the machine learning model used by the caries determination tool 130 (e.g., relabeling the training data as indicating that a false positive does not correspond to a caries condition if indicated by a dentist). In some embodiments, the use of the patient data 140 is optional (e.g., if the patient data 140 may contain noise that reduces the accuracy or other performance of the AI ​​model).

[0018] 2 illustrates one embodiment of exemplary modules and databases used by a caries determination tool in using artificial intelligence to output tooth conditions. As depicted in FIG. 2, caries determination tool 130 includes an image capture module 231, a caries determination module 232, and an image pre-processing module 233, as well as a machine learning module 241. The modules and databases depicted in FIG. 2 are merely exemplary, and caries determination tool 130 may include more or fewer modules and / or databases and still achieve the functionality described herein. Furthermore, the modules and / or databases of treatment decision tool 130 may be instantiated in whole or in part on client device 110 and / or one or more servers.

[0019] Image capture module 231 performs activities related to causing imaging instrument 110 to capture information and transmit that information to caries determination tool 130 and / or patient data database 140. This activity may include instructing imaging instrument 110 (e.g., via a locally installed application or via remote command) to capture images of the patient's teeth and transmit them to image capture module 231 for processing. In response to receiving the images, image capture module 231 optionally performs pre-processing on the images (e.g., with image pre-processing module 233). Pre-processing may include optimizing the images, removing artifacts, verifying that the images comply with one or more protocols (e.g., that the images show certain teeth that they are supposed to show), and the like.

[0020] In some embodiments, the image capture module 231 prompts the technician to perform certain activities. For example, the technician may be prompted by the image capture module 231 to capture images of certain teeth, gum areas, slide the instrument to capture images from certain angles, or perform any other manipulation of the image capture device. Images captured during actions corresponding to these prompts are validated through a pre-processing protocol that validates images that comply with the protocol, and in response to them not complying with the protocol, the image capture module 231 may prompt the technician to re-capture such images.

[0021] The caries determination module 232 determines one or more conditions of the patient based on the captured images. In one embodiment, the caries determination module 232 determines a condition that is a binary decision of whether the patient may have caries or not. That is, the output of the caries determination module 232 is either that the patient may have caries and should see a dentist, or that the patient does not have caries and does not need to see a dentist. In other embodiments, the caries determination module 232 may output a diagnosis (e.g., the upper right incisor has caries through the enamel). Additionally or alternatively, the caries determination module 232 may output a treatment (e.g., a filling is needed on the upper right incisor). Additionally or alternatively, the caries determination module 232 may output a risk urgency (e.g., a binary value of "urgent" or "not urgent," a 1 to 10 scale that increases in proportion to urgency, or any other manner of indication) based on how likely the caries is to become serious within a threshold time period (e.g., needing a filling now, but needing root canal treatment in a month). Additionally or alternatively, the caries determination module 232 may output an indication of any risks posed by the caries (e.g., developing periodontal disease if the caries is not treated).

[0022] The caries determination module 232 may make these determinations by using machine learning. In one embodiment, one or more long-wavelength images (e.g., DPOCT images) are obtained by the image capture module 231 and input into a machine learning model by the caries determination module 232. The machine learning model is trained using training data that labels images with one or more conditions (e.g., the training images are labeled to indicate whether a dental visit is necessary, whether a specific diagnosis exists, urgency, whether a certain treatment is required, risk, and the like). The machine learning model may be a neural network (e.g., a convolutional neural network), a deep learning network, or any other type of machine learning model. The machine learning model may be trained using other auxiliary data, such as patient profile information paired with the patient image, and may accept the given patient profile data in addition to the long-wavelength images as input. Optical images showing tooth surfaces, gum surfaces, and the like may additionally be used as input to the machine learning model or portions thereof (e.g., tooth boundary polygons).

[0023] A single-model approach may be accurate but may tolerate bias. For example, if a captured image contains gum pigment, the gum pigment may form the basis for training a model to determine a condition. For this purpose, two machine learning models may be used. The first machine learning model may accept an image as input and, for each region of a tooth, output a likelihood that the tooth contains a certain property (e.g., caries, staining, previous fillings or sealants, or other dental treatments, and the like), or some other indication of the image (e.g., intensity at each location in the image). The second machine learning model may be trained to take a vector of likelihoods (or other indications) for a given tooth or set of teeth as input and output a condition based on that vector. This eliminates the risk of unnecessary factors biasing the condition determination. In some embodiments, the likelihood vector may be supplemented with auxiliary information, such as a visible-light image of the periodontal tissue.

[0024] In some embodiments, patient data 140 can be combined with the long-wavelength image and input into the model. For example, a baseline image from a previous image can be input, or image processing can be used to identify differences between the baseline image and a new image. The differences can be input into a machine learning model(s) to output a disease state. The machine learning model can be trained using historical data of the same tooth at different times, with differences in biomarker intensity over time labeled to indicate the severity of the condition (e.g., worsening or improvement of cavities, worsening staining, etc.).

[0025] 3 shows one embodiment of further details of the sub-modules used by the caries determination module of the treatment decision tool. As depicted in FIG. 3, the sub-modules of the caries determination module 232 may include a DPOCT abstraction module 315, an intensity map module 325, a biomarker determination module 335, and a caries diagnosis module 345. Fewer or more modules may be used to achieve the functionality described herein.

[0026] As described above, in some embodiments, the caries determination module 232 inputs the image into the machine learning model. However, in addition to or intentionally, one or more abstractions of the DPOCT image may be input into the machine learning model. Abstractions as described herein may refer to a representation of the image. For example, a DPOCT image may include pixels of varying intensities. The abstractions may include intensity differences between adjacent pixels in respective regions of the image, and the intensity differences may be input into the machine learning model. In some embodiments, the DPOCT abstraction module 315 may determine regions that have at least a threshold intensity difference (e.g., an average or median intensity difference) relative to their neighboring pixels and use a map of those regions as input into the machine learning model. Ground truth training data may exist for any form of abstraction labeling. In embodiments where a single machine learning model is used, the training data may include a label for the caries diagnosis corresponding to the abstraction. In embodiments where the image is input into a first machine learning model, the abstractions may be labeled with a ground truth for any of the aforementioned output classifications.

[0027] The intensity map 325 may generate one or more intensity maps for an image (e.g., a DPOCT image). The intensity map may indicate the intensity at each pixel of the image for a given property (e.g., reflectance, refractive index, brightness). These intensity maps may be used for any purpose (e.g., to abstract intensity differences, for direct input into a machine learning model, and the like). Generally, healthy dental materials do not absorb much of the DPOCT scan. However, cavities or other artifacts (e.g., cavity fillings) may absorb relatively more and have higher intensity. Therefore, ground truth data labeling a classification (e.g., cavity, stain, previous filling) for the intensity map may enable the machine learning model to output a classification based on the intensity map.

[0028] In one embodiment, the intensity map module 325 may determine the intensity map using the intensity difference between the intensity of the most recent image and the intensity of a previous image from the patient data 140 (e.g., the intensity difference of the current image relative to a previous image of the same tooth).

[0029] The caries determination module 232 may input the full-color image, the DPOCT image, and any mentioned abstractions, either alone or together, into a machine learning model trained to either directly output a diagnosis or to output a likelihood for a region of the tooth containing a certain property. When a machine learning model is trained to incorporate two or more of these inputs together, the ground truth for each type of input is utilized to optimally determine the associated classification.

[0030] In some embodiments, the biomarker determination module 335 applies heuristics to the intensity map and / or abstractions representing intensity differences, where the heuristics identify parameters for the size of a region and / or the intensity change of a region that are classified as a biomarker. These biomarkers can be input into a single machine learning model that outputs a diagnosis based on the biomarkers. In embodiments where two machine learning models are used, a first machine learning model can be trained to detect intensity changes from the intensity map and output a classification for each location on the tooth based on the intensity change. From the classification and optionally other heuristics (e.g., location size), the biomarker determination module 335 can determine the likelihood that each region with a given intensity change forms a particular type of biomarker (e.g., a cavity, a previous filling, etc.). The biomarker determination module 335 can compare these likelihoods to previous likelihoods from the patient data 140 (e.g., from images captured in a previous scan of the same tooth during an earlier examination) and use the comparison as input to the second machine learning model. The biomarker determination module 335 may additionally or alternatively annotate biomarkers using intensity differences of biomarkers from earlier scans of the patient 140 for input into a second machine learning model.

[0031] The caries diagnosis module 345 generates a diagnosis for one or more regions of the tooth. In some embodiments, the caries diagnosis module 345 determines a diagnosis for each biomarker of the tooth. That is, a machine learning model (e.g., a single model or a second machine learning model in a two-model approach) may output a probability that the tooth (or biomarkers) has one or more conditions (e.g., cavity, previous filling, stain, etc.) for either the tooth as a whole or different biomarkers of the tooth. The caries diagnosis module 345 may apply a threshold such that a probability higher than a minimum threshold leads to the caries diagnosis module 345 determining that the corresponding condition applies as a diagnosis. Thus, a tooth with three biomarkers (one for cavity, one for previous filling, and one for stain) may have each of those biomarkers individually diagnosed in conjunction with its location on the tooth. Alternatively, a tooth may be externally labeled as having each of those three conditions without labeling different regions of the tooth accordingly. Finally, the caries diagnosis module 345 may output an abstraction of the diagnosis. For example, heuristics may be applied to indicate whether the diagnosis should be classified as "see dentist" or "no problems detected." For example, classifications indicating staining and previous fillings on the tooth may lead to a "no problems detected" classification based on those heuristics, while any finding of a cavity or pre-cavity material may lead to a "see dentist" classification.

[0032] FIG. 4 illustrates one embodiment of a DPOCT intensity map. As shown in FIG. 4, image 400 shows a tooth color image and corresponding intensity map. These intensity maps can be used as training data, where experts indicate labels (e.g., enamel, dentin, secondary caries, fissures, and composite fillings). The labels can be for any type of tooth, such as a healthy molar versus a reconstructed molar. The patient data 140 can be used to determine the tooth type (e.g., the patient data 140 determines that the molar is reconstructed, and then the machine learning model matches the reconstructed molar to ground truth data). Different machine learning models can be trained for different tooth types, and the caries determination module 232 can select a machine learning model that matches the patient's tooth type as determined from the patient data 140 to diagnose the condition for a given tooth. This can improve efficiency and reduce noise in the output of the machine learning model by using a better target set of ground truth data.

[0033] 5 illustrates an exemplary process for determining a diagnosis of a dental condition using a supervised machine learning approach. As depicted in FIG. 5, process 500 begins with caries determination tool 130 capturing 502 (e.g., using image capture module 231) image data representations of a patient's teeth based on data obtained from a tooth-scanning hardware device. In one embodiment, the data obtained from the tooth-scanning hardware device may include deep penetration optical coherence tomography (DPOCT) data. The image data may include, or be, an intensity map reflecting one or more optical properties of the teeth based on the DPOCT data.

[0034] The caries determination module 130 inputs 504 the image data into a first supervised machine learning model (e.g., from machine learning model 241). The first supervised machine learning model, in some embodiments, may be trained to detect intensity variations between regions in the intensity map and output classifications for each different location of the tooth based on the intensity variations, each classification forming a biomarker. In some embodiments, the first supervised machine learning model may additionally take a tooth color image as input and output a biomarker based on both the image and the intensity map.

[0035] The caries determination module 130 receives 506 multiple biomarkers as output from the first supervised machine learning model, each biomarker corresponding to a different location on the tooth. If the input includes a color image, the output of the first machine learning model may exclude color data from the color image (e.g., to eliminate bias in the use of the second machine learning model). The caries determination module 130 inputs 508 the multiple biomarkers into a second supervised machine learning model and receives 510 a diagnosis of the tooth condition from the second supervised machine learning model as output. In some embodiments, historical biomarkers for the tooth (e.g., from the patient data 140 where previous images were captured and decisions were made from the first and / or second models) are retrieved and input into the second machine learning model. In some embodiments where intensity information is input into the model, intensity difference may be input into the second machine learning model of current intensity relative to previous intensity, as changes in intensity over time may be useful for certain classifications. In some embodiments, the diagnosis may include a classification for each biomarker of a given tooth.

[0036] Although the disclosure herein refers to human patients, the patient need not be human. Dogs, horses, etc. may have cavities using the systems and methods disclosed herein. For example, training data may be input for the animal's teeth to train an animal-specific machine learning model(s) to diagnose the condition. The machine learning model may, in some embodiments, be trained using cross-species training data. The correct answer in any of these training sets may be based on one or more of outcomes from treatment, pain, and clinician labeling.

[0037] (b) Summary The foregoing description of embodiments of the present invention has been presented for purposes of illustration. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. Those skilled in the art will recognize that numerous modifications and variations are possible in light of the above disclosure.

[0038] Some portions of this description describe embodiments of the invention in terms of algorithms and symbolic representations of operations on information. These algorithmic descriptions and representations are commonly used by those skilled in the data processing arts to effectively convey the substance of their work to others skilled in the art. These operations, while described functionally, computationally, or logically, are understood to be implemented by computer programs or equivalent electronic circuits, microcode, or the like. Further, without loss of generality, it has also proven convenient at times to refer to these arrangements of operations as modules. The described operations and their associated modules may be embodied in software, firmware, hardware, or any combination thereof.

[0039] Any of the steps, operations, or processes described herein may be performed or implemented using one or more hardware or software modules, alone or in combination with other devices. In an embodiment, software modules are implemented using a computer program product comprising a computer-readable medium containing computer program code, which can be executed by a computer processor to perform any or all of the steps, operations, or processes described.

[0040] Embodiments of the present invention may also relate to apparatus for performing the operations herein. This apparatus may be specially constructed for the required purposes and / or it may comprise a general-purpose computing device selectively activated or reconfigured by a computer program stored in the computer. Such a computer program may be stored in a non-transitory tangible computer-readable storage medium or any type of medium suitable for storing electronic instructions, which may be coupled to a computer system bus. Furthermore, any computing system referred to herein may include a single processor or may be an architecture employing a multiple processor design for increased computing power.

[0041] Embodiments of the present invention may also relate to products produced by the computational processes described herein. Such products may comprise information resulting from the computational processes, which information is stored on a non-transitory, tangible, computer-readable storage medium, and may include any embodiment of a computer program product or other data combination described herein.

[0042] Finally, the language used herein has been selected primarily for readability and instructional purposes, and not necessarily to delineate or limit the subject matter of the present invention. Therefore, it is intended that the scope of the present invention be limited not by this detailed description, but rather by any claims that issue in an application based thereon. Additionally, the disclosure of embodiments of the present invention is intended to be illustrative, but not limiting, of the scope of the invention, which is set forth in the following claims.

Claims

1. 1. A method of diagnosing a dental condition, the method being implemented on a computer having one or more processors, the method comprising: the one or more processors capturing image data indicative of the patient's teeth based on data obtained from a hardware device scanning the teeth, the image data obtained from the hardware device scanning the teeth comprising deep penetration optical coherence tomography (DPOCT) data, and the image data comprising an intensity map reflecting one or more optical properties of the teeth based on the DPOCT data; the one or more processors input the image data into a first supervised machine learning model, the first supervised machine learning model being trained to detect intensity variations between regions in the intensity map and to output a respective classification for each different location of the tooth based on the intensity variations, each classification forming a biomarker; the first supervised machine learning model additionally taking a color image of the tooth as input, the first supervised machine learning model being additionally trained to output a biomarker based on both the color image and the intensity map, the output of the first supervised machine learning model excluding color data from the color image; the one or more processors receiving as output a plurality of biomarkers from the first supervised machine learning model, each biomarker corresponding to a different location on the tooth; and the one or more processors inputting the plurality of biomarkers into a second supervised machine learning model; the one or more processors receiving as output a diagnosis of a dental condition from the second supervised machine learning model; A method comprising:

2. 10. The method of claim 1, further comprising: the one or more processors accessing historical biomarkers for the teeth; the historical biomarkers being input by the one or more processors into the second supervised machine learning model along with the plurality of biomarkers; and the second supervised machine learning model outputting the diagnosis based on both the historical biomarkers and the plurality of biomarkers.

3. 3. The method of claim 2, wherein inputting the historical biomarker along with the plurality of biomarkers into the second supervised machine learning model comprises the one or more processors calculating an intensity difference for each biomarker of the plurality of biomarkers relative to the historical biomarker and inputting each intensity difference into the second supervised machine learning model.

4. 2. The method of claim 1, wherein receiving the diagnosis of the dental condition from the second supervised machine learning model as output comprises the one or more processors receiving a plurality of diagnoses, each diagnosis of the plurality of diagnoses corresponding to a different one of the plurality of biomarkers.

5. 1. A non-transitory computer-readable medium comprising a memory having instructions encoded thereon that, when executed, cause one or more processors to perform operations for diagnosing a dental condition, the instructions comprising: capturing image data indicative of a patient's teeth based on data obtained from a hardware device scanning the teeth, the image data obtained from the hardware device scanning the teeth comprising deep penetration optical coherence tomography (DPOCT) data, and the image data comprising an intensity map reflecting one or more optical properties of the teeth based on the DPOCT data; inputting the image data into a first supervised machine learning model, the first supervised machine learning model being trained to detect intensity variations between regions in the intensity map and to output a respective classification for each different location of the tooth based on the intensity variations, each classification forming a biomarker; the first supervised machine learning model additionally taking a color image of the tooth as input, the first supervised machine learning model being additionally trained to output a biomarker based on both the color image and the intensity map, the output of the first supervised machine learning model excluding color data from the color image; receiving a plurality of biomarkers from the first supervised machine learning model as output, each biomarker corresponding to a different location on the tooth; inputting the plurality of biomarkers into a second supervised machine learning model; receiving as output from the second supervised machine learning model a diagnosis of the dental condition; A non-transitory computer-readable medium comprising instructions for performing

6. 6. The non-transitory computer-readable medium of claim 5, wherein the instructions further comprise instructions for accessing historical biomarkers for the teeth, the historical biomarkers being input along with the plurality of biomarkers into the second supervised machine learning model, and the second supervised machine learning model outputting the diagnosis based on both the historical biomarkers and the plurality of biomarkers.

7. 7. The non-transitory computer-readable medium of claim 6, wherein the instructions for inputting the historical biomarker along with the plurality of biomarkers into the second supervised machine learning model comprise instructions for calculating an intensity difference for each biomarker of the plurality of biomarkers relative to the historical biomarker, and inputting each intensity difference into the second supervised machine learning model.

8. 6. The non-transitory computer-readable medium of claim 5, wherein the instructions for receiving the diagnosis of the dental condition from the second supervised machine learning model as output comprise instructions for receiving a plurality of diagnoses, each diagnosis of the plurality of diagnoses corresponding to a different one of the plurality of biomarkers.

9. 1. A system comprising: a memory having instructions encoded thereon for diagnosing a dental condition; one or more processors Equipped with The one or more processors, when executing the instructions, perform actions, the actions including: capturing image data indicative of a patient's teeth based on data obtained from a hardware device scanning the teeth, the image data obtained from the hardware device scanning the teeth comprising deep penetration optical coherence tomography (DPOCT) data, and the image data comprising an intensity map reflecting one or more optical properties of the teeth based on the DPOCT data; inputting the image data into a first supervised machine learning model, the first supervised machine learning model being trained to detect intensity variations between regions in the intensity map and to output a respective classification for each different location of the tooth based on the intensity variations, each classification forming a biomarker; the first supervised machine learning model additionally taking a color image of the tooth as input, the first supervised machine learning model being additionally trained to output a biomarker based on both the color image and the intensity map, the output of the first supervised machine learning model excluding color data from the color image; receiving a plurality of biomarkers from the first supervised machine learning model as output, each biomarker corresponding to a different location on the tooth; inputting the plurality of biomarkers into a second supervised machine learning model; receiving as output from the second supervised machine learning model a diagnosis of the dental condition; Including, the system.

10. 10. The system of claim 9, wherein the operations further include accessing historical biomarkers for the tooth, the historical biomarkers being input along with the plurality of biomarkers into the second supervised machine learning model, and the second supervised machine learning model outputting the diagnosis based on both the historical biomarkers and the plurality of biomarkers.

11. 11. The system of claim 10, wherein inputting a historical biomarker along with the plurality of biomarkers into the second supervised machine learning model comprises calculating an intensity difference for each biomarker of the plurality of biomarkers relative to the historical biomarker and inputting each intensity difference into the second supervised machine learning model.

Images