I have a bunch of dental X-rays that I need to annotate for an object detection/semantic segmentation project. One of the features we hope to detect is bone level. Bone level is measured as the distance between the gum line and the bone surface, illustrated by the two pair of points in the picture below. The top point is the gum line, and the bottom point the bone surface.
Example of dental X-ray found on http://allsmilesdentspa.com/features/x-ray/
I'm planning on custom training YOLOv8, and I am looking for ideas to annotate those two features.
The points were actually placed using the annotation platform CVAT. **Would the points be a good way to annotate those two features? ** My concern:
Points represent a very specific location, and the model might miss the target if the detection isn't accurate enough.YOLOv8 might struggle with small, precise targets, leading to potential errors in measurement.
Points don't provide contextual information about the surrounding area, which might make it harder for the model to understand the overall structure.
Alternatives:
Using bounding boxes or polygons, but then measuring the distance between the two feature become inaccurate. Maybe I could use some good old Canny Edge Detection post-processing?
Thank you!
I would suggest annotating the white part of the teeths as well as the bone with polygons. You can then take the smallest distance between each tooth and the bone to get your desired measure.
You could also try your approach with using bounding boxes, but I doubt it will works as well.
