Clinical dentistry requires the effort of highly trained professionals who allocate their precious time among a multitude of activities contributing to patient care. The throughput of a single dental professional can’t be cheated, if the quality of care is to be preserved. However, some of the regular activities, such as analysis of dental RTGs, can be strengthened by automated assistants. Besides faster retrieval of information, they can serve as providers of second opinions, which is very valuable in dental RTG diagnostics.
In Hopero, we worked with AI:Dental, a company that spearheads the automated analysis of dental RTGs to support dental care professionals (clinical use) and apprentices (students of dentistry). In particular, they experimentally explored the state of the art in dental RTG image retrieval (both with text and image queries) and created a prototype model tailored to AI:Dental’s data. It was a challenging task because in dentistry diagnostics, obtaining ground truth is harder than in other comparable domains due to relatively low annotator consensus, even among expert annotators – experienced dentists. Due to this, as another avenue of our consultations, we explored the theoretical models of consensus building in crowdsourcing scenarios, which will guide future annotation schemes of AI:Dental.
The collaboration with AI:Dental was an example of synergy between domain knowledge, data and prior knowledge about machine learning from AI:Dental, and more general research experience and experimentation rigour provided by Hopero.
