Evaluation of Algorithms for Lung Fissure Segmentation in CT Images

Automatic detection of the interlobular lung fissures is a crucial task in computer aided diagnostics and intervention planning, and required for example for determination of disease spreading or pulmonary parenchyma quantification. Moreover, it is usually the first step of a subsequent segmentation of the five lung lobes. Due to the clinical relevance, several approaches for fissure detection have been proposed. They aim at finding plane-like structures in the images by analyzing the eigenvalues of the Hessian matrix. Furthermore, these values can be used as features for supervised fissure detection. In this work, two approaches for supervised an three for unsupervised fissure detection are evaluated and compared to each other. The evaluation is based on thoracic CT images acquired with different radiation doses and different resolutions. The experiments show that each approach has advantages and the choice should be made depending on the specific requirements of following algorithm steps.