AI for automatic classification of dental x-rays

Purpose and goal

The aim of the project was to develop and test AI-based models for automatic classification of unmarked dental x-ray plates, for use as decision support for the diagnosis of both caries and other ailments. The project has, based on dental x-ray plates with associated diagnosis, managed to train an AI model that automatically identifies caries in dental x-ray plates with higher precision and reliability than dentists and shows where the caries are most likely to be located. The work also shows that the same approach should be applicable to other ailments.

Expected results and effects

An evaluation of the caries detection model showed that the majority of discrepancies between the model´s and the dentist´s assessments were due to (1) that the dentists had more information available than only the pictures, for example when caries appeared on chewing surfaces; (2) the dentists´ assessment of incorrectness. The conclusion is that the model can save a lot of time and increase the reliability of dentists´ assessments. The work has attracted a great deal of interest from both players in the dental equipment area and the media, including the Tandläkartidningen and SVT.

Planned approach and implementation

The work has largely followed plan, with pre-processing, training of models for caries detection, evaluation of these, testing of models for osteoporosis and models for implementation and dissemination. As the work on the training has taken more calendar time than planned, due to onboarding of a new project partner with a new set of dental plates, the model for caries detection has been prepared for implementation within Folktandvården VGR, but this has not yet been started. However, dialogues with several equipment suppliers that have shown interest in the result have been initiated.