ASSET: Auto-Segmentation of the Seventeen SEgments for Ventricular Tachycardia Ablation in Radiation Therapy
E Morris1*, R Chin1, T Wu1, C Smith1, S Nejad-Davarani2, M Cao1, (1) Department of Radiation Oncology, UCLA Health, Los Angeles, CA, (2) Department of Radiation Oncology, University of Michigan, Ann Arbor, MI
Presentations
TU-A-202-2 (Tuesday, 7/12/2022) 7:30 AM - 8:30 AM [Eastern Time (GMT-4)]
Room 202
Purpose: Stereotactic radioablation has been shown to be a safe and promising treatment option for patients with refractory ventricular tachycardia (VT). However, target segmentation is complex and time consuming with large uncertainties. A pipeline for Auto-segmentation of the Seventeen SEgments for Tachycardia ablation (ASSET) was developed to aid in target delineation and image registration for radiotherapy planning.
Methods: ASSET was retrospectively applied to non-contrast CTs for 26 VT patients with pacemakers. Principle component analysis (PCA) was used to locate the parasternal long-axis (PLAX) of the LV and then compared to the physician defined apex and basal points using minimum distance to agreement (MDA) and angle of separation. The manually selected right ventricle insertion point (RVIP) and LV contours were used to apply the ASSET model to automatically generate the 17 segments of the LV myocardium (LVM). Lastly, morphological closing and smoothing was used for post-processing of the segments. The 17 segments were evaluated by three experts and were also compared to the originally treated gross tumor volume.
Results: The ASSET model takes less than 5 minutes to run and requires only one physician defined point, the RVIP. The physician defined PLAX differed from the 1st principle component by 1.4±0.3 mm MDA and 1.2±0.7 degrees over all patients. This translated to a difference between segments generated from a physician defined PLAX and the PLAX defined by PCA of <0.05 DSC and <0.5 mm. Manual segmentation of the 17 segments took >2 hours/patient across 3 experts. Compared across experts, ASSET segmentations were <2 mm MDA and >0.80 DSC and qualitatively deemed clinically useable. Analysis of inter-observer variability revealed a standard deviation between experts of 0.54 mm in MDA and 0.07 DSC.
Conclusion: ASSET offers efficient and reliable automatic segmentations for the 17 segments of the LVM on non-contrast CTs for radiotherapy planning.
Keywords
Segmentation, Heart, Target Localization
Taxonomy
TH- External Beam- Photons: extracranial stereotactic/SBRT
Contact Email