Caries is one of the most common chronic diseases in the world, but its detection has so far been manual and time-consuming. Dentists or their assistants must carefully examine X-ray images to determine the presence and type of caries. This process is not only repetitive, but also subject to high variability in image quality and interpretation between professionals.

Although experts are trained to recognize subtle patterns in the scans, time pressure and the increasing number of diagnostic tasks reduce their ability to focus on complex or rare cases. The aim was therefore to develop a system that enables automatic detection of different types of caries, freeing up time for high-quality patient care.

We developed a computer vision system that is able to analyze bitewing X-ray images and detect and classify nine different types of caries. Using a deep learning model trained on annotated scan data from the customer, the system learned to localize patterns and assign them to specific caries classes.

Initial experiments showed promising results, particularly in the model's ability to generalize to previously unknown X-ray images. This indicates that the model can become a reliable decision support system for dentists.

The system showed great potential as a proof of concept, as it was able to accurately identify multiple types of caries and reduced the reliance on manual evaluation. These initial results provide a solid foundation for further development, with higher quality data sets and continuous validation.

Such a system offers the opportunity to increase diagnostic efficiency in dental practices, reduce diagnostic variance and ultimately improve patient care through earlier and more consistent caries detection.