Click here to

Session: Therapy: External Beam: Automatic Treatment Planning [Return to Session]

Robust Lung SBRT Knowledge-Based Planning with Missing Data Management

A Kroshko1,2*, L Archambault1,2, O Morin3, (1)Universite Laval, Quebec, Canada (2) CHUQ Pavillon Hotel-Dieu de Quebec, Quebec, Canada(3) University of California San Francisco, San Francisco, CA

Presentations

MO-IePD-TRACK 5-7 (Monday, 7/26/2021) 12:30 PM - 1:00 PM [Eastern Time (GMT-4)]

Purpose: Lung SBRT is a complex site and treatment technique. A knowledge-based planning (KBP) technique is harder to implement because not all organs-at-risk in the thorax are delineated based on the location of the tumor and its size, which often results in a heterogenous plan database. We developed a novel technique to predict dosimetric parameters for the lung SBRT effective even in the presence of missing data.

Methods: A retrospective study was made of 399 lung SBRT patients (219 training, 30 validation and 50 test) treated with VMAT. Prescribed dose was 48 or 52 Gy in 4 fractions and 50 Gy in 5 fractions. Bayesian Stochastic Frontier Analysis (BSFA) with Markov Chain Monte Carlo (MCMC) algorithm were used to design a knowledge-based technique to predict achievable dose sparing to organs-at-risk. 95% Highest Density Interval (HDI) is also calculated within the MCMC framework and give us credible interval of dose prediction given a patient specific anatomy. Missing data for non-delineated OARs were estimated following a truncated normal distribution specific to each OAR.

Results: In total, 16 dosimetric parameters were predicted for the spinal cord PRV, esophagus, heart, main bronchus, great vessels and chest wall. Mean difference between the observed and predicted values for the test group ranges between 1.5 (1.9) Gy for the D0.35cc of the spinal cord PRV and 4.9 (5.3) Gy for the D0.035cc of the main bronchus. Investigation of the missing data model shows that it is robust in our predictive model.

Conclusion: BSFA for lung SBRT with missing data management shows convincing results in predicting dosimetric parameters for lung SBRT.

ePosters

    Keywords

    Not Applicable / None Entered.

    Taxonomy

    Not Applicable / None Entered.

    Contact Email

    Share: