Dynamic tracking using a respiratory-only trained correlation model was more effective at reducing maximum localization errors and 95% internal target volume margins than a cardiorespiratory model.
Does filtering out cardiac motion prior to training a respiratory correlation model improve dynamic tracking accuracy in stereotactic arrhythmia radioablation?
In a phantom model of stereotactic arrhythmia radioablation, training dynamic tracking models on respiratory motion alone (filtering out cardiac motion) reduces localization errors and treatment margins.
Stereotactic arrhythmia radioablation (STAR) is a novel treatment employing single high dose radiation beam treatment for ventricular tachycardia (VT). STAR is a non-invasive alternative to catheter ablation, both of which aim to ablate arrhythmogenic myocardial scar tissue driving reentry. Radiation treatment of moving targets necessitates utilization of motion management methods in order to minimize healthy tissue irradiation. For STAR, the radiation target is affected by both cardiac and respiratory motion, introducing errors in both treatment planning and delivery. Dynamic tracking is a motion management method which employs real-time radiation beam adjustment throughout treatment delivery, based on a respiratory correlation model between internal and external surrogate motion generated prior to treatment. This work investigates the efficacy of dynamic tracking using a correlation model trained on cardiorespiratory motion, and additionally determines the potential benefits of filtering out the cardiac component of motion prior to training the correlation model. A cardiac phantom was successfully designed and constructed. Respiratory motion was simulated using a commercial respiratory gating phantom. 24 cardiorespiratory and eight respiratory-only traces were used for motion simulation. Dynamic tracking delivery was tested using a BrainLAB ExacTrac Vero4DRT linear accelerator, with the internal-external surrogate correlation model trained on either the combined cardiorespiratory motion or only the respiratory motion. Absolute localization errors of radiation delivery location from the predicted target location were obtained from automated ExacTrac delivery logs. Similar distributions of position discrepancies for hypothetical treatments delivered with no tracking enabled were calculated by convolving the distributions of each cardiac and respiratory trace. Internal target volume (ITV) margins were obtained by calculating the 95th percentile of the localization errors. Both correlation model scenarios decreased mean and maximum localization errors compared to no tracking, as well as 95% ITV margins. The respiratory-only trained model was more effective at reducing maximum localization errors and 95% ITV margins than the cardiorespiratory trained model. Dynamic tracking has the potential to decrease treatment margins in STAR for VT, and filtering cardiac motion out during respiratory correlation model training can further reduce margins, reducing healthy tissue irradiation and improving patient outcomes.
Katja Nell (Thu,) conducted a other in Ventricular tachycardia. Dynamic tracking with respiratory-only trained correlation model vs. No tracking and cardiorespiratory trained model was evaluated on Absolute localization errors and 95% internal target volume (ITV) margins. Dynamic tracking using a respiratory-only trained correlation model was more effective at reducing maximum localization errors and 95% internal target volume margins than a cardiorespiratory model.