A Pilot Comparative Study of Dental Students' Ability to Detect Enamel-only Proximal Caries in Bitewing Radiographs With and Without the use of AssistDent (R) Deep Learning Software

Devlin, H.; Ashley, M.; Williams, T. G.; Purvis, B.

Enamel-only proximal caries, if detected, can be reversed by non-invasive treatments. Dental bitewing radiograph analysis is central to diagnosis and treatment planning and when used to detect enamel-only proximal caries it is an important tool in minimum intervention and preventive dentistry. However, the subtle patterns of enamel-only proximal caries visible in a bitewing radiographs are difficult to detect and often missed by dental practitioners. This pilot study measures the ability of a cohort of third-year dental students to detect enamel-only proximal caries in bitewing radiographs with and without the use of a deep learning assistive software AssistDent(R). We demonstrate an increased ability in the detection of enamel-only proximal caries by the students using AssistDent, showing a mean sensitivity level of 0.80 (95%CI {+/-} 0.04), increased from 0.50 (95%CI {+/-} 0.13) p<0.01 shown by students not using AssistDent. This improvement in ability was achieved without an increase in false positives. Mean false positives per bitewing radiograph recorded by students when using AssistDent was 2.64 (95%CI {+/-} 0.57), and by students without using AssistDent was 2.46 (95%CI {+/-} 1.51). Based on these results we conclude that the AI-based software AssistDent significantly improves third-year dental students' ability to detect enamel-only proximal caries and could be considered as a tool to support minimum intervention and preventive dentistry in teaching hospitals and general practice. We also discuss how the experience of conducting this pilot study can be used to inform the design and methodology of a follow-on study of AssistDent in dental practice use.