A team of UK researchers is building an AI system that can spot dental disease on X-rays, and now needs to prove it is safe and effective enough for the NHS. Dental disease is extremely common, but diagnosing it from radiographs relies on human experts who vary in skill and consistency. The researchers have already built a prototype that learns from noisy annotations by weighing the reliability of different annotators. The current project will collect high-quality expert annotations from NHS and private dentists, engage patients and the public through citizen science and workshops, and run the system through formal regulatory and health-economic evaluations. If successful, the AI could be integrated into clinical workflows as a Class I medical device, helping dentists detect disease earlier and more consistently. This would directly improve patient outcomes and could reduce the burden on NHS dental services. The project also includes human factors research and market analysis to understand barriers to commercial adoption, ensuring the technology is not just effective but practical for real-world use.
View original technical description
Background: Over the past two years, we have been developing an Artificial Intelligence (AI)-assisted dental disease detection prototype system capable of learning from noisy dental radiograph annotations by automatically differentiating between the reliability of different annotators. We have filed a UK patent application (Application Number 2212322.8), developed a Zooniverse project (zooniverse.org/projects/huhui/dental-disease-detection) that crowdsourced over 57,000 annotations of 4,000 dental radiographs from over 2,500 volunteer annotators. Based on the success of the Zooniverse project, we have been building a Dental Radiograph Collection, Annotation, and Prediction System (DR CAPS) for dentists to contribute expert annotations on dental radiographs. Research question: While our proof-of-concept results are promising, the route towards a clinical-ready AI system for dental disease detection with radiography at an NHS scale presents several important research questions: Q1: Can we utilise DR CAPS to streamline annotation and accelerate the collection of a large quantity of expert annotations from dentists in the NHS (including primary and secondary care settings) and private practices? Q2: How can we build on the momentum of having engaged with over 2,500 public members to further increase the impact of planned PPI activities and inform our study? Q3: What evidence must we compile to ensure that the system meets the regulatory requirements for UKCA mark Medical Devices? Q4: How can we model the health economics and lower barriers to entry for DR CAPS by integrating the system into the clinical workflow? Aims and Objectives: The project aims to provide the comprehensive clinical, technical, and health economic evidence necessary to ensure safe progression towards adoption within the clinical workflow, to achieve the following objectives: O1: To produce clinical-level annotations from dental clinicians for further model training and evaluation. O2: To engage with patients and members of the general public about dental radiography and dental diseases through the project s citizen science project, various media formats, and online/onsite workshops. O3: To implement a Quality Management System (QMS) and ensure that the developed AI system meets regulatory requirements to obtain UKCA mark as Class I Medical Devices. O4: To perform detailed analyses of system performance, human factors and modelling of health economics, including identification of failure case analysis, user experience research, and evaluation of the costs and benefits of the system to the NHS. Methods: M1: Expert annotation collection using retrospective datasets from three clinical sites. M2: Integration of a Medical Device QMS for regulation compliance and approval. M3: Evaluation of simulated usage of the system in NHS and private practices. M4: Co-production with a dentist advisory group and a PPI group; usability and health economics study for eventual NHS adoption. Timelines for delivery: M1: Start immediately, duration 1.5 years. M2: Start immediately, duration 2.5 years. M3: Start in late 2023, duration 2 years. M4: Start immediately, duration 2.5 years. Anticipated Impact and Dissemination: I1: Generate clinical, technical, and health economic evidence required for clinical adoption of the developed Medical Device. I2: Human factors and market analysis to understand challenges and mitigation for commercialisation. I3: High impact publications and public-facing materials.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know