Pechin Lo

Extraction of Airways From CT (EXACT'09)

This paper describes a framework for establishing a reference airway tree segmentation, which was used to quantitatively evaluate fifteen different airway tree extraction algorithms in a standardized manner. Because of the sheer difficulty involved in manually constructing a complete reference standard from scratch, we propose to construct the reference using results from all algorithms that are to be evaluated. We start by subdividing each segmented airway tree into its individual branch segments. Each branch segment is then visually scored by trained observers to determine whether or not it is a correctly segmented part of the airway tree. Finally, the reference airway trees are constructed by taking the union of all correctly extracted branch segments. Fifteen airway tree extraction algorithms from different research groups are evaluated on a diverse set of twenty chest computed tomography (CT) scans of subjects ranging from healthy volunteers to patients with severe pathologies, scanned at different sites, with different CT scanner brands, models, and scanning protocols. Three performance measures covering different aspects of segmentation quality were computed for all participating algorithms. Results from the evaluation showed that no single algorithm could extract more than an average of 74% of the total length of all branches in the reference standard, indicating substantial differences between the algorithms. A fusion scheme that obtained superior results is presented, demonstrating that there is complementary information provided by the different algorithms and there is still room for further improvements in airway segmentation algorithms.

Vessel-guided airway tree segmentation: A voxel classification approach

This paper presents a method for airway tree segmentation that uses a combination of a trained airway appearance model, vessel and airway orientation information, and region growing. We propose a voxel classification approach for the appearance model, which uses a classifier that is trained to differentiate between airway and non-airway voxels. This is in contrast to previous works that use either intensity alone or hand crafted models of airway appearance. We show that the appearance model can be trained with a set of easily acquired, incomplete, airway tree segmentations. A vessel orientation similarity measure is introduced, which indicates how similar the orientation of an airway candidate is to the orientation of the neighboring vessel. We use this vessel orientation similarity measure to overcome regions in the airway tree that have a low response from the appearance model. The proposed method is evaluated on 250 low dose computed tomography images from a lung cancer screening trial. Our experiments showed that applying the region growing algorithm on the airway appearance model produces more complete airway segmentations, leading to on average 20% longer trees, and 50% less leakage. When combining the airway appearance model with vessel orientation similarity, the improvement is even more significant (p<0.01) than only using the airway appearance model, with on average 7% increase in the total length of branches extracted correctly.

Texture-Based Analysis of COPD: A Data-Driven Approach

This study presents a fully automatic, data-driven approach for texture-based quantitative analysis of chronic obstructive pulmonary disease (COPD) in pulmonary computed tomography (CT) images. The approach uses supervised learning where the class labels are, in contrast to previous work, based on measured lung function instead of on manually annotated regions of interest (ROIs). A quantitative measure of COPD is obtained by fusing COPD probabilities computed in ROIs within the lung fields where the individual ROI probabilities are computed using a k nearest neighbor (kNN ) classifier. The distance between two ROIs in the kNN classifier is computed as the textural dissimilarity between the ROIs, where the ROI texture is described by histograms of filter responses from a multi-scale, rotation invariant Gaussian filter bank. The method was trained on 400 images from a lung cancer screening trial and subsequently applied to classify 200 independent images from the same screening trial. The texture-based measure was significantly better at discriminating between subjects with and without COPD than were the two most common quantitative measures of COPD in the literature, which are based on density. The proposed measure achieved an area under the receiver operating characteristic curve (AUC) of 0.713 whereas the best performing density measure achieved an AUC of 0.598. Further, the proposed measure is as reproducible as the density measures, and there were indications that it correlates better with lung function and is less influenced by inspiration level.

Mass preserving image registration for lung CT

This paper presents a mass preserving image registration algorithm for lung CT images. To account for the local change in lung tissue intensity during the breathing cycle, a tissue appearance model based on the principle of preservation of total lung mass is proposed. This model is incorporated into a standard image registration framework with a composition of a global affine and several free-form B-Spline transformations with increasing grid resolution. The proposed mass preserving registration method is compared to registration using the sum of squared intensity differences as a similarity function on four groups of data: 44 pairs of longitudinal inspiratory chest CT scans with small difference in lung volume; 44 pairs of longitudinal inspiratory chest CT scans with large difference in lung volume; 16 pairs of expiratory and inspiratory CT scans; and 5 pairs of images extracted at end exhale and end inhale phases of 4D-CT images. Registration errors, measured as the average distance between vessel tree centerlines in the matched images, are significantly lower for the proposed mass preserving image registration method in the second, third and fourth group, while there is no statistically significant difference between the two methods in the first group. Target registration error, assessed via a set of manually annotated landmarks in the last group, was significantly smaller for the proposed registration method.

Weight Preserving Image Registration for Monitoring Disease Progression in Lung CT

We present a new image registration based method for monitoring regional disease progression in longitudinal image studies of lung disease. A free-form image registration technique is used to match a baseline 3D CT lung scan onto a following scan. Areas with lower intensity in the following scan compared with intensities in the deformed baseline image indicate local loss of lung tissue that is associated with progression of emphysema. To account for differences in lung intensity owing to differences in the inspiration level in the two scans rather than disease progression, we propose to adjust the density of lung tissue with respect to local expansion or compression such that the total weight of the lungs is preserved during deformation. Our method provides a good estimation of regional destruction of lung tissue for subjects with a significant difference in inspiration level between CT scans and may result in a more sensitive measure of disease progression than standard quantitative CT measures.

Short-term effect of changes in smoking behaviour on emphysema quantification by CT

Background The effect of smoking cessation and smoking relapse on lung density was studied using low-dose CT. Methods Spiral, multidetector, low-dose CT was performed on 726 current and former smokers (>20 pack-years) recruited from a cancer screening trial. Lung density was quantified by calculating the 15th percentile density (PD15), which was adjusted to predicted total lung capacity. Data were analysed by linear regression models. Results At baseline mean PD15 was 45 g/l in former smokers (n=178) and 55 g/l in current smokers (n=548), representing a difference of 10 g/l (p<0.001). After smoking cessation (n=77) PD15 decreased by 6.2 g/l (p<0.001) in the first year, and by a further 3.6 g/l (p<0.001) in the second year, after which no further change could be detected. Moreover, the first year after relapse to smoking (n=18) PD15 increased by 3.7 g/l (p=0.02). Conclusions Current smoking status has a major influence on lung density assessed by CT, and the difference in lung density between current and former smokers observed in cross-sectional studies corresponds closely to the change in lung density seen in the years after smoking cessation. Current smoking status, and time since cessation or relapse, should be taken into account when assessing the severity of diseases such as emphysema by CT lung density.

Toward a Theory of Statistical Tree-Shape Analysis

To develop statistical methods for shapes with a tree-structure, we construct a shape space framework for tree-shapes and study metrics on the shape space. This shape space has singularities which correspond to topological transitions in the represented trees. We study two closely related metrics on the shape space, TED and QED. QED is a quotient euclidean distance arising naturally from the shape space formulation, while TED is the classical tree edit distance. Using Gromovs metric geometry, we gain new insight into the geometries defined by TED and QED. We show that the new metric QED has nice geometric properties that are needed for statistical analysis: Geodesics always exist and are generically locally unique. Following this, we can also show the existence and generic local uniqueness of average trees for QED. TED, while having some algorithmic advantages, does not share these advantages. Along with the theoretical framework we provide experimental proof-of-concept results on synthetic data trees as well as small airway trees from pulmonary CT scans. This way, we illustrate that our framework has promising theoretical and qualitative properties necessary to build a theory of statistical tree-shape analysis.

Toward automatic regional analysis of pulmonary function using inspiration and expiration thoracic CT

PURPOSE To analyze pulmonary function using a fully automatic technique which processes pairs of thoracic CT scans acquired at breath-hold inspiration and expiration, respectively. The following research objectives are identified to: (a) describe and systematically analyze the processing pipeline and its results; (b) verify that the quantitative, regional ventilation measurements acquired through CT are meaningful for pulmonary function analysis; (c) identify the most effective of the calculated measurements in predicting pulmonary function; and (d) demonstrate the potential of the system to deliver clinically important information not available through conventional spirometry. METHODS A pipeline of automatic segmentation and registration techniques is presented and demonstrated on a database of 216 subjects well distributed over the various stages of COPD (chronic obstructive pulmonary disorder). Lungs, fissures, airways, lobes, and vessels are automatically segmented in both scans and the expiration scan is registered with the inspiration scan using a fully automatic nonrigid registration algorithm. Segmentations and registrations are examined and scored by expert observers to analyze the accuracy of the automatic methods. Quantitative measures representing ventilation are computed at every image voxel and analyzed to provide information about pulmonary function, both globally and on a regional basis. These CT derived measurements are correlated with results from spirometry tests and used as features in a kNN classifier to assign COPD global initiative for obstructive lung disease (GOLD) stage. RESULTS The steps of anatomical segmentation (of lungs, lobes, and vessels) and registration in the workflow were shown to perform very well on an individual basis. All CT-derived measures were found to have good correlation with spirometry results, with several having correlation coefficients, r, in the range of 0.85-0.90. The best performing kNN classifier succeeded in classifying 67% of subjects into the correct COPD GOLD stage, with a further 29% assigned to a class neighboring the correct one. CONCLUSIONS Pulmonary function information can be obtained from thoracic CT scans using the automatic pipeline described in this work. This preliminary demonstration of the system already highlights a number of points of clinical importance such as the fact that an inspiration scan alone is not optimal for predicting pulmonary function. It also permits measurement of ventilation on a per lobe basis which reveals, for example, that the condition of the lower lobes contributes most to the pulmonary function of the subject. It is expected that this type of regional analysis will be instrumental in advancing the understanding of multiple pulmonary diseases in the future.

Optimal graph based segmentation using flow lines with application to airway wall segmentation

This paper introduces a novel optimal graph construction method that is applicable to multi-dimensional, multi-surface segmentation problems. Such problems are often solved by refining an initial coarse surface within the space given by graph columns. Conventional columns are not well suited for surfaces with high curvature or complex shapes but the proposed columns, based on properly generated flow lines, which are non-intersecting, guarantee solutions that do not self-intersect and are better able to handle such surfaces. The method is applied to segment human airway walls in computed tomography images. Comparison with manual annotations on 649 cross-sectional images from 15 different subjects shows significantly smaller contour distances and larger area of overlap than are obtained with recently published graph based methods. Airway abnormality measurements obtained with the method on 480 scan pairs from a lung cancer screening trial are reproducible and correlate significantly with lung function.

Geometries on spaces of treelike shapes

In order to develop statistical methods for shapes with a tree-structure, we construct a shape space framework for treelike shapes and study metrics on the shape space. The shape space has singularities, which correspond to topological transitions in the represented trees. We study two closely related metrics, TED and QED. The QED is a quotient euclidean distance arising from the new shape space formulation, while TED is essentially the classical tree edit distance. Using Gromovs metric geometry we gain new insight into the geometries defined by TED and QED. In particular, we show that the new metric QED has nice geometric properties which facilitate statistical analysis, such as existence and local uniqueness of geodesics and averages. TED, on the other hand, has algorithmic advantages, while it does not share the geometric strongpoints of QED. We provide a theoretical framework as well as computational results such as matching of airway trees from pulmonary CT scans and geodesics between synthetic data trees illustrating the dynamic and geometric properties of the QED metric.