Limits...
Optimizing parameters of an open-source airway segmentation algorithm using different CT images.

Nardelli P, Khan KA, Corvò A, Moore N, Murphy MJ, Twomey M, O'Connor OJ, Kennedy MP, Estépar RS, Maher MM, Cantillon-Murphy P - Biomed Eng Online (2015)

Bottom Line: All the considered cases have been segmented successfully with good results in terms of leakage presence.Results on clinical data are comparable to other teams' methods, as obtained by evaluation against the EXACT09 challenge, whereas results obtained from the phantom prove the reliability of the method across multiple CT platforms and acquisition parameters.Results suggest that the algorithm can be considered stable across multiple CT platforms and acquisition parameters and can be considered as a starting point for the development of a complete airway segmentation algorithm.

View Article: PubMed Central - PubMed

Affiliation: School of Engineering , University College Cork, College Road, Cork, Ireland. p.nardelli@umail.ucc.ie.

ABSTRACT

Background: Computed tomography (CT) helps physicians locate and diagnose pathological conditions. In some conditions, having an airway segmentation method which facilitates reconstruction of the airway from chest CT images can help hugely in the assessment of lung diseases. Many efforts have been made to develop airway segmentation algorithms, but methods are usually not optimized to be reliable across different CT scan parameters.

Methods: In this paper, we present a simple and reliable semi-automatic algorithm which can segment tracheal and bronchial anatomy using the open-source 3D Slicer platform. The method is based on a region growing approach where trachea, right and left bronchi are cropped and segmented independently using three different thresholds. The algorithm and its parameters have been optimized to be efficient across different CT scan acquisition parameters. The performance of the proposed method has been evaluated on EXACT'09 cases and local clinical cases as well as on a breathing pig lung phantom using multiple scans and changing parameters. In particular, to investigate multiple scan parameters reconstruction kernel, radiation dose and slice thickness have been considered. Volume, branch count, branch length and leakage presence have been evaluated. A new method for leakage evaluation has been developed and correlation between segmentation metrics and CT acquisition parameters has been considered.

Results: All the considered cases have been segmented successfully with good results in terms of leakage presence. Results on clinical data are comparable to other teams' methods, as obtained by evaluation against the EXACT09 challenge, whereas results obtained from the phantom prove the reliability of the method across multiple CT platforms and acquisition parameters. As expected, slice thickness is the parameter affecting the results the most, whereas reconstruction kernel and radiation dose seem not to particularly affect airway segmentation.

Conclusion: The system represents the first open-source airway segmentation platform. The quantitative evaluation approach presented represents the first repeatable system evaluation tool for like-for-like comparison between different airway segmentation platforms. Results suggest that the algorithm can be considered stable across multiple CT platforms and acquisition parameters and can be considered as a starting point for the development of a complete airway segmentation algorithm.

No MeSH data available.


Related in: MedlinePlus

Representation of length, width and height in a 3D volume as utilised in this approach.
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC4482101&req=5

Fig1: Representation of length, width and height in a 3D volume as utilised in this approach.

Mentions: The algorithm’s first step involves cropping the whole volume in order to extract the trachea. To this end, an average trachea length, the whole volume width, and a height given by the whole volume minus a small portion of volume itself are considered (for definition of depth, width and length in a 3D volume see Figure 1). An example of the first cropped volume as extracted by the algorithm is given in Figure 2 where an axial, a sagittal, and a coronal slice view are shown. Using this cropped volume the initial segmentation of the trachea starting from the placed fiducial point is accomplished. Details of how trachea segmentation is performed are reported in the first sub-section. Once this first segmentation is completed, the second step involves using the obtained trachea label to improve the cropping of the trachea volume. To achieve this, the carina position is computed automatically, by scanning from the fiducial position and moving slice by slice towards the carina along the axial slices. The algorithm recognises the carina as the point in which the segmented label splits into two different parts, representing the two main bronchi. As an example, Figure 3 shows the carina position as found on a CT image after the first trachea segmentation. The algorithm’s third step uses the carina position to compute the maximum height of the trachea, and the volume is cropped accordingly. Simultaneously with this third step, the cropping size is also updated in length, to take into account possible bends in the trachea. This is achieved by moving slice by slice along the z axis and identifying points in which the label touches the side borders of the previously cropped volume, in which cases the cropping is extended in length. The fourth step involves a second and final trachea segmentation of the new cropped volume. Once the trachea label is finalised, the carina position within the trachea label is used to automatically define the seed points for the segmentation of the right and left lungs. Details of right and left lungs segmentation are reported in the second sub-section.Figure 1


Optimizing parameters of an open-source airway segmentation algorithm using different CT images.

Nardelli P, Khan KA, Corvò A, Moore N, Murphy MJ, Twomey M, O'Connor OJ, Kennedy MP, Estépar RS, Maher MM, Cantillon-Murphy P - Biomed Eng Online (2015)

Representation of length, width and height in a 3D volume as utilised in this approach.
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC4482101&req=5

Fig1: Representation of length, width and height in a 3D volume as utilised in this approach.
Mentions: The algorithm’s first step involves cropping the whole volume in order to extract the trachea. To this end, an average trachea length, the whole volume width, and a height given by the whole volume minus a small portion of volume itself are considered (for definition of depth, width and length in a 3D volume see Figure 1). An example of the first cropped volume as extracted by the algorithm is given in Figure 2 where an axial, a sagittal, and a coronal slice view are shown. Using this cropped volume the initial segmentation of the trachea starting from the placed fiducial point is accomplished. Details of how trachea segmentation is performed are reported in the first sub-section. Once this first segmentation is completed, the second step involves using the obtained trachea label to improve the cropping of the trachea volume. To achieve this, the carina position is computed automatically, by scanning from the fiducial position and moving slice by slice towards the carina along the axial slices. The algorithm recognises the carina as the point in which the segmented label splits into two different parts, representing the two main bronchi. As an example, Figure 3 shows the carina position as found on a CT image after the first trachea segmentation. The algorithm’s third step uses the carina position to compute the maximum height of the trachea, and the volume is cropped accordingly. Simultaneously with this third step, the cropping size is also updated in length, to take into account possible bends in the trachea. This is achieved by moving slice by slice along the z axis and identifying points in which the label touches the side borders of the previously cropped volume, in which cases the cropping is extended in length. The fourth step involves a second and final trachea segmentation of the new cropped volume. Once the trachea label is finalised, the carina position within the trachea label is used to automatically define the seed points for the segmentation of the right and left lungs. Details of right and left lungs segmentation are reported in the second sub-section.Figure 1

Bottom Line: All the considered cases have been segmented successfully with good results in terms of leakage presence.Results on clinical data are comparable to other teams' methods, as obtained by evaluation against the EXACT09 challenge, whereas results obtained from the phantom prove the reliability of the method across multiple CT platforms and acquisition parameters.Results suggest that the algorithm can be considered stable across multiple CT platforms and acquisition parameters and can be considered as a starting point for the development of a complete airway segmentation algorithm.

View Article: PubMed Central - PubMed

Affiliation: School of Engineering , University College Cork, College Road, Cork, Ireland. p.nardelli@umail.ucc.ie.

ABSTRACT

Background: Computed tomography (CT) helps physicians locate and diagnose pathological conditions. In some conditions, having an airway segmentation method which facilitates reconstruction of the airway from chest CT images can help hugely in the assessment of lung diseases. Many efforts have been made to develop airway segmentation algorithms, but methods are usually not optimized to be reliable across different CT scan parameters.

Methods: In this paper, we present a simple and reliable semi-automatic algorithm which can segment tracheal and bronchial anatomy using the open-source 3D Slicer platform. The method is based on a region growing approach where trachea, right and left bronchi are cropped and segmented independently using three different thresholds. The algorithm and its parameters have been optimized to be efficient across different CT scan acquisition parameters. The performance of the proposed method has been evaluated on EXACT'09 cases and local clinical cases as well as on a breathing pig lung phantom using multiple scans and changing parameters. In particular, to investigate multiple scan parameters reconstruction kernel, radiation dose and slice thickness have been considered. Volume, branch count, branch length and leakage presence have been evaluated. A new method for leakage evaluation has been developed and correlation between segmentation metrics and CT acquisition parameters has been considered.

Results: All the considered cases have been segmented successfully with good results in terms of leakage presence. Results on clinical data are comparable to other teams' methods, as obtained by evaluation against the EXACT09 challenge, whereas results obtained from the phantom prove the reliability of the method across multiple CT platforms and acquisition parameters. As expected, slice thickness is the parameter affecting the results the most, whereas reconstruction kernel and radiation dose seem not to particularly affect airway segmentation.

Conclusion: The system represents the first open-source airway segmentation platform. The quantitative evaluation approach presented represents the first repeatable system evaluation tool for like-for-like comparison between different airway segmentation platforms. Results suggest that the algorithm can be considered stable across multiple CT platforms and acquisition parameters and can be considered as a starting point for the development of a complete airway segmentation algorithm.

No MeSH data available.


Related in: MedlinePlus