Integrated lung field segmentation of injured region with anatomical structure analysis by failure–recovery algorithm from chest CT images

• Automatic segmentation of lung anatomical structures: airways, lung vessels and lung lobes.
• Combination of segmented lung lobes and diffuse lung disease classification.
• Airway segmentation by failure tracking and recovery algorithm.

This work proposes a functionality for computerized tomography (CT) based investigation of diffuse lung diseases diagnosis that enables the evaluation of the disease from lung anatomical structures. Automated methods for segmenting several anatomy structures in chest CT are proposed: namely the lobe lungs, airway tree and pulmonary vessel tree. The airway and pulmonary vessel trees are segmented using a failure tracking and recovery algorithm. The algorithm checks intermediary results consistence, backtrack to a history position if a failure is detected. The quality of the result is improved while reducing the processing time even for subjects with lung diseases. The pulmonary vessels are segmented through the same algorithm with different seed points. The seed for the airway tree segmentation is within the tracheal tube, and the seed for the pulmonary vessels segmentation is within the heart. The algorithm is tested with CT images acquired from four distinct types of subjects: healthy, idiopathic interstitial pneumonias (IIPs), usual interstitial pneumonia (UIP) and chronic obstructive pulmonary disease (COPD). The main bronchi are found in the segmented airway and the associated lung lobes are determined. Combining the segmented lung lobes and the diffuse lung diseases classification, it is possible to quantify how much and where each lobe is injured. The results were compared with a conventional 3D region growing algorithm and commercial systems. Several results were compared to medical doctor evaluations: inter-lobe fissure, percentage of lung lobe that is injured and lung and lobe volumes. The algorithm proposed was evaluated to be robust enough to segment the cases studied.