My thesis work focuses on automatic guidewire segmentation on 2D fluoroscopic X-ray images.
There are several methodological difficulties: the limited amount of imaging data, the low quality of the images and the lack of an annotated database. To overcome these challenges, several contributions are proposed in this thesis. First we created a data base.
Then our approach is to compare different loss functions for guidewire segmentation. We obtain good segmentation of numerically important portions.
Finally, we also segment the endoscope that carries the guidewire at the entrance to the bile ducts, and assess whether knowledge of its position in the image can influence guidewire segmentation.
In conclusion, this work represents a first approach to guidewire segmentation to improve the ERCP procedure.