Use this URL to cite or link to this record in EThOS:
Title: Lung image segmentation and vision-based navigation for interventional bronchoscopy
Author: Shen, Mali
ISNI:       0000 0004 9356 7400
Awarding Body: Imperial College London
Current Institution: Imperial College London
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
Interventional bronchoscopy provides a minimally invasive approach for diagnosis and treatment of pulmonary diseases such as emphysema and pulmonary carcinoma. Abnormalities identified on chest X-ray computed tomography (CT) scans by using computer-aided diagnosis are often targetted in bronchoscopic intervention. Recent advances in lung volume reduction (LVR) offer new therapies with much lower complication rates than open surgery for patients with emphysema. Image segmentation of the pulmonary lobes from CT scans to quantify lobar volumes, and emphysema severity plays a pivotal role in treatment planning and post-operative assessment. The existing automatic lung lobe segmentation strategies are still vulnerable to anatomical variability due to pathologies. Therefore, to assist lobar density analysis for emphysema patients, an interactive lung lobe segmentation program is developed. A validation study on a cohort of emphysema patients shows the reliability of the program by comparing with commercial software. Furthermore, automatic fi ssure segmentation is explored by adopting a surface growing approach based on Bayesian network and fully convolutional neural networks. The pulmonary lobe with the highest emphysema severity identifi ed in the lobar density analysis is usually targeted in the bronchoscopic LVR procedure. Navigation systems have been studied to assist camera localisation either with video-CT registration or electromagnetic tracking. Video-CT registration is favoured due to easy setup and low cost. Nevertheless, the effect of illumination variations, image artefacts and airway deformation hinders its applicability in real clinical scenarios. Therefore, a video-CT registration based on depth information is proposed to achieve more robustness of camera localisation to illumination changes. A structure-preserving depth recovery method based on generative adversarial learning is developed to address partial occlusion due to image artefacts. Moreover, the tissue deformation is tackled by introducing an airway descriptor based on shape context. The proposed navigation approaches have been validated on phantom data and in vivo data, and results show improved camera localisation accuracy, thus better clinical feasibility for bronchoscopic navigation.
Supervisor: Yang, Guang-Zhong ; Shah, Pallav Sponsor: Imperial College London
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral