Expanding Objective CT-based Phenotyping to Lungs with Enhanced Radiodensities

项目摘要 自动生成的、客观的和可重复的基于CT成像的生物标志物对乳腺癌至关重要。 对肺部疾病的亚类进行鉴定，以更好地将表型与基因型联系起来， 慢性阻塞性肺疾病（COPD）、间质性肺疾病等肺部疾病的新治疗方法 纤维化、结节病或哮喘。虽然先前关于开发基于图像的COPD生物标志物的研究， 哮喘已经显示出有希望的结果，已经开发的疾病实体的CT生物标志物的翻译， 具有由显著炎症、实变、肺泡泛液或肺泡炎引起的高放射密度病理学， 由于缺乏自动化的鲁棒的肺图像分析技术，SIS到目前为止还不成功。这 这一缺点不仅阻碍了现有CT生物标志物的利用， 这些肺部疾病的基于图像的生物标志物。本提案的目的是解决这一瓶颈问题 在定量肺部图像分析中，通过开发用于单次CT扫描的稳健和快速图像分析技术 以及高/低容量CT扫描对。具体而言，用于肺、肺叶和气道分割的新方法将 开发和验证，它可以耐受高密度肺部病理，并且是所需的关键组件 用于计算肺部病理解剖学的CT生物标志物。肺和肺叶分割方法的鲁棒性将 通过利用基于模型的图像分析方法来实现。到目前为止， 因为由于大的器官尺寸，对于肺图像分析计算要求太高。我们将解决 这个问题通过利用通用计算图形处理硬件技术，使我们能够 如初步结果所示，显著减少计算时间。通过提供有效的手段 客观地识别肺结构，这些结构可以被量化并用于亚表型 患者人群，根据需要促进20，000例或更多受试者的大型多中心研究。

Project Summary Automatically generated, objective, and reproducible CT imaging-based biomarkers are critical for the differen- tiation of sub-classes of pulmonary disease to better link phenotypes to genotypes, enabling the development of new treatments for lung diseases like chronic obstructive pulmonary disease (COPD), interstitial pulmonary fibrosis, sarcoidosis, or asthma. While previous research on developing image-based biomarkers for COPD and asthma has shown promising results, the translation of already developed CT biomarkers for disease entities with high radiodensity pathology caused by significant inflammation, consolidation, alveolar flooding or fibro- sis was not successful up to now due to the lack of automated robust lung image analysis techniques. This shortcoming not only hinders the utilization of existing CT biomarkers, it is also problematic for developing new image-based biomarkers for these lung diseases. The objective of this proposal is to address this bottleneck in quantitative lung image analysis by developing robust and fast image analysis techniques for single CT scan and high/low volume CT scan pairs. Specifically, novel methods for lung, lung lobe, and airway segmentation will be developed and validated, which can tolerate high density lung pathology and are a key component required for calculating CT biomarkers for lung pathoanatomy. Robustness of lung and lobe segmentation methods will be achieved by utilizing model-based image analysis methods. Up to now, such approaches were considered as too computationally demanding for lung image analysis because of the large organ size. We will address this issue by utilizing general-purpose computation on graphics processing hardware techniques, allowing us to reduce computation times significantly, as demonstrated by preliminary results. By providing an efficient means of objectively identifying lung structures, these structures can be quantified and utilized for sub-phenotyping patient populations, as required to facilitate large multi-center studies with 20,000 or more subjects.