AI may help clinicians choose best path for patients missing permanent teeth

BUFFALO, N.Y. — In most individuals, permanent second premolars begin to erupt around age 11, pushing out the baby, or primary, teeth.

However, in anywhere between 2 and 11% of the population, these permanent second premolars do not exist.

“Tooth agenesis, a congenital condition characterized by the absence of one or more teeth, is among the most common and clinically challenging dental anomalies encountered today,” says Thikriat Al-Jewair, DDS, L.B Badgero Endowed Chair and associate professor in the Department of Orthodontics in the University at Buffalo School of Dental Medicine.

She explains that it presents a conundrum for orthodontists who usually must choose among three main approaches: retain the primary molars that have no permanent successors, extract them and close the space orthodontically, or extract them and maintain the space for future prosthetic replacement, such as an implant or bridge.

“Managing patients who present with missing teeth can be challenging not only for the patient, but also for the orthodontist, because it introduces another layer of complexity,” she says. “We must carefully evaluate the degree of crowding, the initial position of the teeth and the patient’s facial aesthetics, while also factoring in patient treatment preferences and cost considerations. In some cases, the retained primary tooth remains healthy and functional. In others, it does not.”

Such decision-making may become easier in the future using artificial intelligence, according to an exploratory study Al-Jewair recently completed with William Tanberg, statistician with the Biostatistics Center for Collaborative Research and Data Coordination in the Department of Biostatistics in the School of Public Health and Health Professions.

The study was published in the January 2026 issue of Orthodontics and Craniofacial Research.

The two other researchers on the study are Ozge Colak, a former resident in UB’s Department of Orthodontics who has a private practice in Katy, Texas, and Mohammed Elnagar, a faculty member in the Department of Orthodontics at the University of Illinois, Chicago, College of Dentistry who is also an expert in AI.

Tanberg, who has been collaborating with the orthodontics department for more than four years as a biostatistician, also has formal training in AI.

“My main role was to build the machine learning or AI models using the data Dr. Colak collected and to assist in the interpretation of the results,” Tanberg says. “Many people do not realize that much of the foundations of machine learning and AI come from the field of statistics.”

Top AI algorithms show high accuracy

Third molars, or wisdom teeth, are the most common to undergo agenesis. After that, what’s often missing are the second mandibular premolars, which are the teeth that the team focused on in this study.

Long-term clinical studies indicate that more than 80% of the primary second molars can remain functional for decades. However, in some patients, the teeth either fall out or start to become submerged below the gum level, a condition called ankylosis.

“This becomes problematic because as the tooth submerges, the surrounding bone begins to recede, altering the bone’s height and thickness,” Al-Jewair says. “Over time, this can complicate future restorative treatment, especially if the patient plans to receive dental implants.”

The team developed and applied an AI algorithm using dental models, radiographs, clinical photographs, and the dental and medical history of 100 patients who were undergoing treatment in UB’s Orthodontic Clinic and two private practices in Western New York between 2010 and 2024. The goal was to predict the most appropriate treatment approach in each case.

The mean age of the patients was 14, and all of them had their primary second molars still intact.

Out of four machine-learning models the team evaluated, Random Forest classifier showed the highest accuracy — 96.4% — making it particularly promising, Al-Jewair says.

Features such as patient preference for restoration, the amount of mandibular arch crowding and ankylosis were the strongest predicators of treatment decision accuracy in the Random Forest model.

“Errors in orthodontic decision-making can have long-lasting consequences, ranging from compromised facial aesthetics to periodontal deterioration and patient dissatisfaction,” Al-Jewair explains. “Enhancing diagnostic accuracy and improving treatment predictability are critical and is where AI may provide meaningful support.”

AI not new but improving in dental field

She points out that AI is not new to orthodontics. In fact, various forms of AI have been used in the field for nearly two decades, particularly in diagnostic imaging and digital treatment planning.

What has changed in the last five years is the sophistication of these systems. Advancements in machine learning and deep learning have made AI increasingly capable of supporting complex treatment decision-making.

“Our research showed that even with a limited set of data in terms of sample size highly accurate predictive models could be trained to act as an expert system to assist a clinician,” Tanberg says. “I think the future of such expert systems to assist orthodontists or other clinicians in treatment decision-making is very promising and will likely be commonplace within a decade.”

Both Tanberg and Al-Jewair emphasize that the AI work they are exploring is not meant to replace the decisions of licensed orthodontists but rather to support them.

“Less experienced orthodontists, such as new graduates, struggle with these type of malocclusions, and AI could be a helpful tool,” Al-Jewair says. “If we can integrate these AI algorithms into the technology we already use in clinical practice, I think they have the potential to enhance the accuracy of our clinical decisions and improve treatment outcomes.”