Devlin, H., Williams, T., Graham, J. et al. The ADEPT study: a comparative study of dentists’ ability to detect enamel-only proximal caries in bitewing radiographs with and without the use of AssistDent artificial intelligence software. Br Dent J 231, 481–485 (2021).
Background
Picking up enamel-only caries using bitewing radiography is difficult and often missed by dental practitioners. This research summary covers the use of AssistDent (an artificial intelligence software) and whether is positively affects the ability to pick up enamel-only caries.
Radiographs can increase chances of picking up caries compared to by clinical examination alone. Regarding this, many studies have revealed poor diagnostic sensitivity for radiographic detection of demineralisation by dentists. The introduction of AI and this software programme uses algorithms to search for evidence of enamel-only proximal caries on bitewing radiographs.
Methods
The participants were chosen from two sources: dentists practising as GDPs who also tutor at Manchester Dental School and practising dentists who are doing postgraduate training at University of Manchester. 23 dentists were randomly divided into a control group- without use of AI, and an experimental group- using AI with on-screen prompts showing potential enamel-only proximal caries. They analysed 24 bitewings, an example of which is shown in Fig 1.
Fig 1: An example of the AI screen prompts showing enamel-only caries on a bitewing radiogram. The orange arrows show presence of enamel-only proximal caries.
The experimental group were provided with indicators made by the AI software , showing enamel-only caries, while the control group had none.
Results
The control group found 44.3% of caries, compared to the experimental group finding 75.8%. However, the experimental group incorrectly identified caries in 14.6% of the healthy surfaces compared to 3.7% in the control group. Although there shows a positive result of use of AI from the review, using programmes with increased sensitivity also increases the number of false positives. This results in unnecessary preventative treatment and use of limited healthcare resources. This is something to keep in mind when using more specific programmes in preventative care.
Table 1 illustrates that the use of AssistDent increased the ability to detect enamel-only proximal caries by 71%. It is also indicated that the experimental group were 11% less likely to correctly identify healthy proximal surfaces as non-caries.
Table 1: Odds ratio of true positive and true negative rates of the experimental group compared to the control group with 95% confidence intervals
Conclusion
Overall, it was found that the AI programme, AssistDent greatly improved the dentists’ ability to detect enamel-only proximal caries and is considered as a gateway tool into more use of AI in dentistry.
Research Summary Written By: Tamara Al-Sabah