Fig 1.
Pipeline of pulmonary phantom generation.
The first stage (orange) includes processes to generate the arterial, venous, and airway tree pathways. The second stage (purple) comprises different steps to provide realistic features to the final phantom. An additional previous stage (cyan) is introduced when generating anthropomorphic pulmonary CT phantoms, but not in bronchopulmonary segment phantoms, to extract information from the real CT image that will be used to adapt some of the parameters. Blue boxes represent input and output images; pink boxes represent input objects and parameters; green boxes are intermediate images for a single structure; and red boxes represent images with the three subtrees corresponding to the pulmonary air and blood flow systems together.
Fig 2.
Workflow for the generation of a synthetic bronchopulmonary structure for arteries (red arrows), airways (green), and veins (blue).
Numbers indicate the order in the workflow.
Fig 3.
Example of oxygenation maps used in the synthesis of a bronchopulmonary segment phantom.
The first row displays the central slices of the 3D oxygenation maps for arteries (a), airways (b), and veins (c); bright/dark areas indicate high/low levels of oxygenation demand, and therefore, the probability of growing structures there. The second row shows the 3D structures generated from the corresponding oxygenation maps, which are used to define subsequent oxygenation maps for the following structures. Null oxygen demands in places already occupied by previous structures can be observed in the top of (b) and (c).
Fig 4.
Example of the application of a deformation algorithm to a tubular tree structure (a). The resulting image is shown in (b) using a value of dmax = 3.
Fig 5.
Oxygenation maps in anthropomorphic pulmonary phantoms.
Sagittal views of arterial (a), bronchial (b), and venous (c) oxygenation maps.
Fig 6.
Effect of randomness and FA factor in bronchopulmonary segment phantoms, displaying arteries in red, veins in blue, and airways in green, with Nf = 50.
(a) FA = 1 ⇒ Nf(air) = 50, (b) FA = 3 ⇒ Nf(air) = 17. Although both examples share the same parameters for the synthesis of arteries, the generated arterial flow systems have different structures because of the change in random seed Rseed.
Fig 7.
Effect of the numbers of terminal nodes in bronchopulmonary segment phantoms.
From left to right, up to down, arterial (red), venous (blue), and airway (green) structures with Nf = {30, 40, 50, 60, 70, 80, 90, 100} and FA = 3.
Fig 8.
Results of anthropomorphic final phantoms, showing two CT images with synthetic right lungs (top row) and left lung (bottom row) obtained from a real CT case (left).
HU window: [-1000, 100].
Fig 9.
Synthetic arteries (red), veins (blue), and airways (green) conforming the reference standard for the anthropomorphic pulmonary CT phantoms.
The first column corresponds to synthetic air and blood flow systems of two phantoms generated from the left lung of one real case. Similarly, the second and third columns correspond to phantoms from the left and right lungs, respectively, from another real case.
Fig 10.
Histograms of evaluation measures showing mean probability density functions (lines) with 95% confidence intervals (shadows) for real (blue) and synthetic (red) lungs.
(a) Intensity distribution. (b) Dispersion of structures. (c) Relation between arteries and airways.
Table 1.
Statistics comparing intraclass similarity measures in real cases (real vs real), and interclass similarity measures (real vs phantom).
P-values (p), Cohen’s d effect size (cD) and AUC values are reported using Kolmogorov–Smirnov distance (dKS) and match distance (dM) for the three evaluation measurements computed. Mean and standard deviations of intraclass (in real cases) and interclass distance values are also displayed on the right.