Artificial Intelligence for Tooth Numbering and Full-Mouth Series Placement

Dentists rely on a specific set of X-ray images to see the full picture of a patient's mouth. This collection, known as a full-mouth series, consists of dozens of individual photographs, each capturing a small section of teeth and bone. To make sense of these images, a dentist must know exactly which tooth appears in each frame and where that frame belongs in the final sequence. Traditionally, this has been a manual task. A human must look at every image, identify the teeth, assign them a number based on a universal system, and then arrange the photos into the correct order. This process is time-consuming and prone to human error, especially when images are blurry, teeth are missing, or the X-ray sensor was held at a slight angle.

In recent years, computers have become powerful enough to learn how to see and interpret medical images, a field known as artificial intelligence. These systems use deep learning, a method where a computer analyzes thousands of examples to learn patterns without being explicitly programmed with rules for every scenario. While computers have already shown they can spot cavities or broken bones, arranging a full set of dental X-rays automatically has remained a difficult challenge. The complexity lies not just in identifying a single tooth, but in understanding how a group of images fits together like pieces of a larger puzzle, even when some pieces are missing or rotated.

A team of researchers at The University of Texas Health Science Center at Houston School of Dentistry, along with collaborators from National Yang Ming Chiao Tung University, has developed a new way to solve this problem. They created an automated system that can detect teeth, number them correctly, and arrange a full set of X-rays into the proper order with remarkable speed and accuracy. Their work, published in a study involving over 5,000 radiographs from 354 patients, demonstrates that a computer can now handle the entire workflow of organizing dental images, a task that previously required significant human attention.

The researchers began by gathering a massive dataset of dental X-rays. They collected 5,431 images from 354 different patients, covering both the upper and lower jaws. Every single tooth in these images was carefully marked by dental professionals who drew a box around each one and assigned it its correct number. This created a "ground truth," a perfect reference that the computer could use to learn. The team then trained their artificial intelligence models on this data, teaching the system to recognize the shape, position, and number of teeth in various conditions.

The system operates in three main stages. First, it uses a detection tool to find every tooth in an X-ray. Unlike older methods that draw simple square boxes around objects, this system uses angled boxes that can tilt and rotate to match the exact position of a tooth. This is crucial because teeth in an X-ray often appear at odd angles depending on how the sensor was placed in the mouth. Once the teeth are located, the system identifies which tooth it is, assigning it a number from one to thirty-two, following the standard universal numbering system used by dentists.

The most innovative part of the study is how the system arranges the images. Instead of just looking at one tooth at a time, the computer looks at the relationship between multiple images. The researchers designed the system to think about the X-rays like pieces of a jigsaw puzzle. Just as a person might look at the edge of one puzzle piece and the curve of another to see if they fit together, the computer analyzes the spatial relationship between different X-ray images. It learns that an image showing the front teeth belongs next to an image showing the side teeth, and that a picture of the upper jaw fits in a different spot than one of the lower jaw. This approach allows the system to place the images correctly even if some teeth are missing or if the anatomy is unclear, a situation where older methods often failed.

The results of this automated process were striking. When the researchers tested their system, it correctly identified and numbered the teeth in 99 percent of cases. More importantly, when it came to arranging the full set of X-rays into the correct order, the system was right 98 percent of the time. This high level of accuracy held true across different types of images, including those of the upper jaw, the lower jaw, and the bitewing images that show the crowns of the teeth. The system performed particularly well with the front teeth and the back teeth on both sides, with only minor confusion occurring between images that were very similar to each other, such as the two images on the far right of the lower jaw.

One of the most significant achievements of this study is the system's ability to handle bitewing images. These specific X-rays, which show the tops of the teeth and the bone between them, have historically been difficult for computers to place automatically because they show less of the tooth root and often lack clear landmarks. Previous attempts to automate dental image arrangement often struggled with these specific photos. The new system, however, successfully placed bitewing images with 96 percent accuracy, proving that it can manage the full complexity of a dental exam without needing a human to step in and fix the order.

The researchers tested their system rigorously to ensure the results were reliable. They split their data into five separate groups and trained the computer on four groups while testing it on the fifth, repeating this process five times so that every image was tested exactly once. This method ensured that the system was not just memorizing the answers but was actually learning to recognize patterns that apply to new, unseen patients. The consistency of the results across all five tests suggests that the system is robust and ready for real-world use, provided it is validated in a clinical setting.

This work represents a shift in how dental imaging could be handled in the future. By automating the tedious tasks of counting teeth and sorting images, the system frees up time for dental professionals to focus on diagnosis and treatment. The study shows that artificial intelligence can learn the subtle spatial relationships in dental X-rays, moving beyond simple detection to understand the context of the entire mouth. While the system is currently a research tool, its high accuracy and ability to handle difficult cases like bitewing images suggest it could soon become a standard part of the dental workflow, streamlining the process from the moment an X-ray is taken to the moment it is ready for a dentist to review.