How AI can benefit periodontology and implant dentistry

In broad terms, how is AI going to impact the clinical practice of periodontology and implant dentistry?

Artificial intelligence is going to be (and is already) transformative in healthcare and it has many broad ramifications as it relates to personalized or precision care. One of the very big areas where AI is being leveraged is in biomedical imaging, in radiographic interpretation.

Many AI companies in dentistry are really focused on radiographic imaging and then adding a variety of tools to identify dental caries, periodontal disease, and so on. Looking at bone levels around implants and teeth and then using this concept of computer vision from the radiographs, you can identify a variety of different structures within the oral cavity for those disease states. But you can also identify things like tumours within the jaws and then be able to discriminate different types of tumours. Through machine learning, once these radiographs are annotated, the systems continue to get better and better.

More dental radiographs have been annotated in this way than medical radiographs. And this is where dentistry is leading in the whole field of healthcare, because a patient who goes to the dentist will oftentimes get a full-mouth series of individual radiographs.

There are so many different items on a radiograph that dentists use daily: bone, tooth, implant, soft tissue. We can also see where the root canal is located, if a root canal filling has been done, if there is a certain type of filling or crown or potential pathology. So many things that can be identified and where artificial intelligence can be leveraged. Looking at a radiograph, the naked eye can discriminate, I believe, 64 shares of grey, but it can now be infinite numbers depending on the quality of the AI.

The other big area in dentistry and medicine that is rapidly being improved by AI is pathology. Pathological specimens, histological slides derived from a biopsy, are now being read by AI tools. And this means being able to identify the needle in the haystack in terms of those cancer cells that were maybe somehow missed by a clinician. Or the AI imaging software picks up these different histological or pathological changes and brings them to the attention of the clinician. And it might be something that can be monitored versus something that needs further biopsy, or maybe we're going to have to deliver a specific chemotherapeutic regimen, maybe surgery.

Clinicians might be able refine clinical protocols with this, and it can potentially be used in the development of clinical guidelines. The consensus report of the EFP’s workshop on novel technologies in the diagnosis of periodontal diseases has just been published (see bibliography).

In what other ways will AI help periodontists in their clinical work?

 In terms of the staging and grading of periodontal disease under the 2018 classification, we may be able to be more automated. Diabetes and smoking are part of that grading scheme, and we now have new genetic markers, as we also have different biomarkers of disease. Those can be built into our grading formulas, where we can improve and customize and build off the 2018 guidelines. And this is where AI is going to be able to take it to another level, in my opinion.

How is AI being deployed in terms of research in periodontology and implant dentistry?

We are now seeing AI methods being incorporated into so many different aspects. Back in 2021, we published a study[1] which was the first use of machine learning for risk-stratifying patients who are receiving treatment for peri-implantitis and reconstructive therapy. As a part of that study, we looked not only at the clinical presentation, but also at radiographs, age, all those basic demographics of a patient. We also harvested biopsies from these patients because then they received surgical therapy, laser treatment, bone grafts. And we looked at the immune cells that were comprising those defects around the dental implants. Then we also looked at the microbial composition.

So, what were those bacteria contributing to the disease presence, and then what were the cell populations that were adjacent to that implant? We used research methods to develop a machine-learning algorithm to identify patients who were low risk, moderate risk, and high risk for responding to the intervention.

We’ve also harnessed AI to address oral-health disparities[2], and there’s a recent paper that used some AI models to identify workforce shortages and so-called “dental deserts”[3], really looking at where these patient populations are located.

What about the potential problems with AI in terms of replicating cultural biases and in terms of data protection?

From a regulatory standpoint and from a bioethics standpoint, we can see the two big aspects that affect the data. One is you have to examine the variability of any disease, of any specific situation. It could be racial, ethnic, gender differences—all those components will relate to the variance. And so that level of variance is going to be important as these models are developed to look at the individual patient level.

And that builds into bias, which is the other component. Variance and bias. In the models that are being developed, there are conscious and unconscious biases that clinicians may be in putting, indoctrinating into the models.

You hear this when it relates to precision or personalized health. So eventually you may want to incorporate specific attributes at the patient level—for example, female, 40 years old, has cardiovascular diseases, and this is the past medical history. Different rubrics may be developed that can hopefully reduce some of the biases.  

Data protection is always a challenge in all of medicine. For example, I did a test with 23andMe (the genomics and biotechnology company that filed for bankruptcy in March this year, raising privacy concerns), and when I took the test, I knew that I was contributing to a data warehouse where they're going to be able to better understand a variety of genetic characteristics associated with disease risk.

In the United States, we have a law for genetic protections—the Genetic Information Non-discrimination Act (GINA)—and we hope we are building trust in many of these systems that they are going to make an appropriate use of the data.