Automatic trachea segmentation and evaluation from MRI data using intensity pre-clustering and graph cuts

A high amount of magnetic resonance imaging (MRI) data is processed in modern epidemiological studies. Reliable and fast automatic segmentation algorithms are required to assist in data analysis. In our project, tracheal dimensions in living patients are studied. We present a fully automated segmentation method for trachea extraction based on intensity pre-clustering and narrow band graph cuts, where the clustering results are used for initialization. The volume of the extracted trachea is evaluated and its size is analyzed using anterior-posterior and lateral diameters. The method was evaluated qualitatively using 10 data sets by measuring volume fraction, volume error, and the Dice's coefficient. It produced sufficiently good results and showed potential to be integrated in clinical routine for processing thousands of participants.