Click here to


Are you sure ?

Yes, do it No, cancel

An Algorithm for Fully Automated Catheter Digitization of the Tandem and Ovoid Applicators in HDR Brachytherapy

B Yan1*, C Deufel2 , I Petersen3 , M Haddock4 , (1) ,Rochester, MN, (2) Mayo Clinic, Rochester, MN, (3) Mayo Clinic, Rochester, MN, (4) Mayo Clinic, Rochester, MN


(Sunday, 7/14/2019) 4:30 PM - 5:00 PM

Room: Exhibit Hall | Forum 4

Purpose: To develop a fully automated catheter digitization algorithm for HDR brachytherapy that does not require manual interaction with the patient images, such as the identification of a seed voxel, threshold selection, or ‘clicking’ onto the image. The automation of catheter digitization may reduce the duration of treatment planning, staff resources, and potential for human error.

Methods: The algorithm segments the applicator (Titanium Fletcher-Suit-Delclos-style Applicator, #AL13030001,Varian,Inc.,Palo Alto,CA) through sequential image processing of the CT images (GE Discovery RT590,General Electric Inc.,Boston,MA). The voxels with the greatest densities were identified using a threshold of 2250 Hounsfield Units(HU). The threshold voxels were refined through morphological erosion and dilation to remove small artifacts. HDBSCAN, a density-based linkage clustering algorithm, was used to assign voxels to catheter and non-catheter structures. For each catheter, the digitized points written to the DICOM structure were calculated from the centroid of the identified voxels on individual slices. The location of the most distal digitized point was refined by the difference between the inner applicator lumen and external applicator tip. The algorithm was retrospectively tested on CT scans from10 cervix cancer patients, and the results were compared with manual digitization by the treatment planner.

Results: The automatic and expert-defined manual reconstructions had a mean (standard deviation, maximum) Euclidean distance of 0.7mm (0.3mm, 1.8mm). The mean (standard deviation, minimum, maximum) difference for the most distal digitized points (applicator tips) was -0.4mm (0.8mm, -1.5mm, 1mm), where a negative number denotes that the automated segmentation was less distal than the manual segmentation. The disagreements were comparable to the resolution of the CT image set (X and Y=1mm, Z=1.25mm).

Conclusion: This work demonstrates the viability of fully automated catheter digitization in HDR brachytherapy treatment planning. Automation of the catheter reconstruction process may lead to safer radiation delivery and reduce the duration of treatment planning.


Segmentation, Treatment Planning, Computer Software


TH- Brachytherapy: Dose optimization and planning

Contact Email