Dental Imaging Project – Caries Region Detection


This is Ashish from the Dental Imaging Team. I, along with Ashray have been working on developing a system that can automatically detect caries in intraoral images. Here is an example image captured by our camera.


Given such image, which potentially has caries, we have predominantly three regions of interest:

  1. Teeth : White, Yellowish white
  2. Gums : Pink to red
  3. Caries : Coffee Brown to Black

Thus it seems plausible that color information is sufficient to distinguish these regions.

Hue captures nearness to red. Thus, we decided to threshold on hue to remove gums. White color corresponds to high intensity, so that can be used to remove teeth, so that we are left only with cavity regions.

Although the model is simple, it is not easy to set the thresholds. This happens because

  1. Some regions can be dark due to lack of light and can be confused as caries.
  2. There are different types and stages of caries with different color distributions
  3. Specular reflections cause artifacts

We made an interactive GUI so that expert dentists can set the thresholds to give best results. Here is the picture of the GUI


Pratima (3rd year Dental student), helped us set these thresholds using the GUI. This approach gave promising results in the images captured by our intraoral camera. We use the values that we obtain from the above GUI in our current algorithm. Here you can see an example input and output:


As can be seen, many times the noise can end up creating thin connections and creating holes in coherent regions (example as in image above). We use the following two operations to combat this

  1. Morphological closing: Dilation followed by erosion
  2. Hole filling: Fill regions that are not reachable from the edges.

Here is the output after these two operations and the corresponding segmented output


In the above diagram we can see that the caries region is nicely segmented from the rest of the image, although there are some other non-caries segments as well. Next step would be to develop algorithms to automatically declare whether the selected regions contain caries or not.

Stay tuned for more updates!

The Dental Imaging Team: Hyunsung, Avi, Ujjwal, Raghu, Meghasyam, Ashish, Ashray, Pratima, Samriddh