Purpose
Artificial Intelligence (AI) tools provide rapid analysis of complex datasets, at the cost of flexibility in the data that is fed to them. To have the best performance, AI tools require training on data similar to the data that will be encountered in their clinical use. For Computed Tomography (CT) scans, a wide variety in scanning protocols imposes challenges on the use of pre-trained AI tools.
In the case of airway segmentation, the required training data are complete airway segmentations. However, complete and high quality...
Methods and materials
The initial airway segmentation 3D-Unet [2] was trained on datasets from the Danish Lung Cancer Screening Trial [3] and from Erasmus MC, Rotterdam, NL. The resulting incomplete airway segmentations of 13 ImaLife scans [4] were imported into Slicer 3D 4.0 [5] and viewed with a window setting of Width: 800 and Level: -625.
The corrections are performed using the segment editor tool in Slicer3D. First, the incomplete airway segmentation is obtained from the segmentation volume. An airway is followed with the cursor in the 3D...
Results
Corrections were performed on 13 CT scans and required 2-4 hours per segmentation. The total count of the distal airways (6th-10th generation) increased by 6 branches (2-8) (median IQR).[Fig 3]Length per distal airway was improved by 29.6% (19.2%-40.8%).[Fig 4]The distal airway lumen diameters were 2.47mm (2.32mm-2.63mm).
Conclusion
Manual correction of incomplete airway segmentations on CT scans provides an avenue for the creation of rich and detailed airway trees for application in artificial intelligence. This method utilises freely available software for manual correction, opening up its use to a wide research base.
Personal information and conflict of interest
I. Dudurych:
Nothing to disclose
A. Garcia-Uceda Juarez:
Nothing to disclose
R. Vliegenthart:
Nothing to disclose
M. de Bruijne:
Nothing to disclose
References
[1] P. Lo et al., ‘Extraction of Airways From CT (EXACT’09)’, IEEE Trans. Med. Imaging, vol. 31, no. 11, pp. 2093–2107, Nov. 2012, doi: 10.1109/TMI.2012.2209674.
[2] A. G.-U. Juarez, H. A. W. M. Tiddens, and M. de Bruijne, ‘Automatic Airway Segmentation in Chest CT Using Convolutional Neural Networks’, in Image Analysis for Moving Organ, Breast, and Thoracic Images, 2018, pp. 238–250.
[3] J. H. Pedersen et al., ‘The Danish Randomized Lung Cancer CT Screening Trial—Overall Design and Results of the Prevalence Round’, Journal of Thoracic...