Fully-Automatic Pulmonary Lobe Segmentation in CT Images

Nicola Giuliani

Publikation: StudienabschlussarbeitMasterarbeit


Pulmonary lobe segmentation is of importance for the localisation and quantification of lung diseases. However, fully-automatic lobe segmentation is still a challenging task, especially in pathological lungs. A new algorithm based on refining an approximate lobe segmentation by minimizing an energy equation using graph cuts is presented. The energy equation combines anatomical information such as the airways, vessels and fissures with prior knowledge on lobar shape. In a first version of the algorithm, the lobe-based labelled airway is used to compute the approximate lobe segmentation. A second, refined version
is relying on the lobe-based labelled vasculature. For this purpose an algorithm for lobe-based vessel labelling has been developed. Both versions of the algorithm were evaluated on two different datasets, including the LObe and Lung Analysis 2011 (LOLA11) dataset. The first version of the algorithm achieved a score of 86.6% at the LOLA11 challenge. The second version, including the labelled vasculature, achieved 90.1%, which is currently the highest score in this challenge for a fully automatic method. The benefit of pulmonary lobe segmentation for artery/vein separation is evaluated in further consequence. The clinical use of the algorithm is demonstrated by computing and evaluating quantitative readouts on each lobe.
QualifikationMaster of Science
Gradverleihende Hochschule
  • Technische Universität Graz (90000)
Betreuer/-in / Berater/-in
  • Stollberger, Rudolf, Betreuer
  • Urschler, Martin, Betreuer
PublikationsstatusVeröffentlicht - 2018


Untersuchen Sie die Forschungsthemen von „Fully-Automatic Pulmonary Lobe Segmentation in CT Images“. Zusammen bilden sie einen einzigartigen Fingerprint.

Dieses zitieren