Can AI really be used for orthodontic triage and screening?

Artificial intelligence appears to have many applications in dentistry. One of these may be to aid orthodontic screening. This is particularly relevant in countries where orthodontic treatment is government-funded for specific populations. In this context, AI-driven remote monitoring may help identify key factors that determine the need for orthodontic treatment. 

This system was examined in this study. 

A team from Sydney, Australia, did the study. The European Journal of Orthodontics published the paper.

What did they ask? 

They did this study too. 

“Compare the validity of a novel tele-orthodontics and artificial intelligence (AI) triaging method using dental monitoring with gold standard face-to-face consultation, using the index of orthodontic treatment need (dental health component) at a treatment threshold of 3.” 

What did they do? 

They conductedid a prospective single-centre 2-by-2 crossover randomised controlled trial. 

Participants 

Inclusion criteria were: 

Patients on the New South Wales waiting list aged more than 5 years who had been referred by dentists for orthodontic triage. They had access to a smartphone and email address

Intervention. 

TAI First.

They provided participants in the study with a Dental Monitoring Scan Box Pro and written instructions. The participants then submitted two scans of sufficient qualitforby the dental monitoring algorit to accepthm. They also completed a history form that included questions about traumatic deep bite and any speech and masticatory issues. 

Then one of the two investigators, who were senior orthodontic residents trained in the use of IOTN, assigned the IOTN dental health component grade. 

Control 

This group was examined face-to-face first. This involved the clinical examination of the patients and a patient-based medical history form. This enabled allocation of the IOTN grade. 

After a two-month gap, the same patients were screened again using the method not used in their first examination. 

Outcomes 

The primary outcome was the validity of the TAI triage in accepting or rejecting referrals, based on a minimum IOTN DHC threshold of IOTN ≥ 3. The study team also examined referral acceptance thresholds of IOTN ≥ 4 (this is the cut-off point for treatment in the UK).

They conducted a clear sample size calculation, which showed that they needed to enrol 136 participants in the study. 

They randomised participants to the test or control allocation using a permuted randomised block design. This was done by administrative staff, who allocated participants via telephone. 

t was not possible to blind the investigators and participants to the allocation sequence; however, the investigators were blinded to all other outcomes.

Standard exploratory statistics were carried out. The diagnostic validity of the triage for IOTN was evaluated by calculating the sensitivity, specificity, positive predictive value, and negative predictive value of the test. 

What did they find? 

At the end of the study, 178 participants completed the investigation. There were no differences between the two groups at the start of the study. 

They provided a reasonable amount of data on the effectiveness of the screening; however, willam onlo look closely at the screening at the IOTN thresholds of 3 and 4. At the IOTN ≥ 3 threshold, the TAI triage has a sensitivity of 1 and a specificity of 0.67 when compared with face-to-face consultations. At the IOTN ≥ 4 threshold, the sensitivity was 0.93, and the specificity was 0.73. 

Sensitivity and specificity are measures used to evaluate a diagnostic test against a gold standard. Sensitivity is the true-positive rate: the proportion of people with the disease who are correctly identified by the test. Specificity is the true-negative rate: the proportion of people without the disease who are correctly identified by the test. Ideally, both sensitivity and specificity equal 1.

The figures from this study show that the sensitivity and specificity at both IOTN levels examined were satisfactory for most screenings. Importantly, the sensitivity values were high. 

The authors also found that screening duration is shorter with TAI than with face-to-face consultation, although I could not find any data on this. 

Their overall conclusion was;

“Triage was a valid and reliable tool for screening orthodontic patients in a public system. It effectively categorised patients with malocclusions ranging from severe to mild based on malocclusion severity and IOTN. However, they emphasised the need for careful classification of borderline malocclusions due to the limitations of certain assessment traits”. 

What did I think? 

I have done a fair amount of research on referral patterns, screening, and the use of IOTN in a public health service. As a result, I found this paper very interesting and highly relevant. 

Overall, their methodology was sound, and the study was well conducted and clearly written up. 

When I reviewed other studies examining the same question in orthodontic screening, I was particularly interested in the effectiveness of the TAI method. This suggests that, with further development to enhance measurement accuracy and to identify other features such as crossbites, AI has a potential role in orthodontic screening. 

This would be a highly useful development in countries where orthodontic treatment is provided by a national health service following a screening programme. 

This is another example of the exciting world of artificial intelligence that we are entering. The pace of change is remarkable, and studies like this are vital to our understanding of this new technology.