Establishing a Normative Atlas of the Human Lung: Computing the Average Transformation and Atlas Construction

Rationale and Objectives: To establish the range of normal for quantitative computed tomography (CT)-based measures of lung structure and function, we seek to develop methods for matching pulmonary structures across individuals and establishing a normative human lung atlas. Materials and Methods: In our previous work, we have presented a three-dimensional (3D) image registration method suitable for pulmonary atlas construction based on CT datasets. The method has been applied to a population of normative lungs in multiple experiments and, in each instance, has resulted in significant reductions in registration errors. This study is a continuation to our previous work by presenting a method for synthesizing a computerized human lung atlas from previously registered and matched 3D pulmonary CT datasets from a population of normative subjects. Our method consists of defining the origin of the atlas coordinate system; defining the nomenclature and labels for anatomical structures within the atlas system; computing the average transformation based on the displacement fields to register individual subject to the common template subject; constructing the atlas by deforming the template with the average transformation; and calculating shape variations within the population. Results: The feasibility of pulmonary atlas construction was evaluated using CT datasets from 20 normal volunteers. Substantial reductions in shape variability were demonstrated. In addition, the constructed atlas depends only slightly on a specific subject being selected as the template. These results indicate the framework is a robust and valid method for pulmonary atlas construction based on CT scans. The atlas consists of a grayscale CT dataset of the template, a labeled mask dataset of the template (ie, lungs, lobes, and lobar fissures are labeled with different gray levels), a data set representing the population's average shape, datasets representing the population's shape variations (ie, the magnitude of standard deviation), a data structure to contain the labels and coordinates of major airway branchpoints, and the labels of the mask dataset, and a reference coordinate system for each lung. Conclusion: A computerized human lung atlas representing by the average shape of a population of twenty normal subjects was constructed and visualized. The atlas provides a basis for establishing regional ranges of normative values for structural and functional measures of the human lung. In the future, we plan to use the computerized human lung atlas to help detect and quantify early signs of lung pathology.