Institute for
Robotics and Process Control

Deutsch   English

Model-based segmentation of the paranasal sinuses

Problem Description

In a robot-assisted endoscopic sinus surgery, one needs to know the outer border of the paranasal sinuses. This border is mandatory for the definition of the workspace of the robot in order to avoid patient injuries. Currently, this border has to be manually determined by a surgeon. This manual segmentation takes far too much time (a few hours) to be feasible in the daily surgical workflow.
To speed up the segmentation process, we search for new approaches to relieve the surgeon from this tedious and time-consuming work. The difficulty thereby is that the border of the paranasal sinuses to surrounding organs can be made up of bone, mucosa, or air. Also, the inter-patient variability of the paranasal sinuses is very high. To make it short: Expert knowledge is needed in order to select the right boundary.


Our approach is to use a virtual, deformable model of the paranasal sinuses in order to replace the expert knowledge. For the generation of such a model, a set of training shapes is analyzed, and a set of parameters is extracted. These parameters can be used to defom the model.
By automatically adjusting the set of parameters, we obtain an initial global solution for the outer border of the paranasal sinuses. However, due to the high inter-patient variability, not all anatomical details can be represented by the deformable model. So, a further local adaptation of the initial solution is needed.
This local adaptation is realized by a new approach that has been developed at the Institute for Robotics and Process Control. Instead of using one set of parameters to deform the whole model, the new approach involves many parameter sets that deform only small parts of the model. By skillfully combining these deformations, we obtain a locally adapted solution that is able to extract the paranasal sinus boundary more accurately than the initial global solution.

Flow Chart

Flow Chart for the model-based segmentation of the paranasal sinuses.


Location of the paranasal sinuses in the human head.

Deformation of the model by modification of the first parameter.

Global fitting of the deformable model, followed by local adaptation.

Further Information

A list of publications for this project can be found here: Publications of Carsten Last
Involved persons: Carsten Last, Dr.-Ing. Simon Winkelbach, Prof. Dr.-Ing. Friedrich M. Wahl
Project partners at the ENT Clinic in Bonn, Germany: Dr. Klaus Eichhorn(External), Prof. Dr. med. Dr. h.c. Friedrich Bootz(External)

It took 0.23s to generate this page.