Click here to

Session: Machine Intelligence Efficacy and Quality I [Return to Session]

An Automated Evaluation Tool for Quantitative Assessment of Knowledge-Based Planning Models

J Harms*, D Stanley, R Popple, University of Alabama at Birmingham, Birmingham, AL


MO-A-BRC-1 (Monday, 7/11/2022) 7:30 AM - 8:30 AM [Eastern Time (GMT-4)]

Ballroom C

Purpose: Knowledge-based planning (KBP) offers the ability to predict dose-volume and plan quality metrics based on information extracted from previous plans. As the prevalence and integration of KBP continues to increase, the need for rapid, quantitative evaluation of KBP models is paramount due to the time and labor-intensive nature of this process. An automated tool was created for automated plan generation and subsequent model analysis.

Methods: The input to the evaluation tool is a spreadsheet containing model ID, calculation parameters, and a list of reference plans. The tool copies the reference plan, applies the KBP model, optimizes the KBP plan, and generates quantitative plots that allow for evaluation of dose-volume metrics. A NSCLC KBP model including plans treated with IMRT or VMAT from 2018-2021 was created and used for testing the evaluation tool. The patient cohort was divided into training (50 cases), testing (10), and validation (25) cohorts. The KBP model was paired with four different optimization templates to compare various planning strategies. The planning approaches involved using line objectives, fixed volume constraints, fixed dose constraints, or gEUD constraints.

Results: Averaged over the testing cohort, all KBP models reduced mean lung and esophageal dose, V45 and V40 to the heart, and max spinal cord dose cord while maintaining target coverage. The line objective model was chosen as it favored a balance between target coverage and OAR sparing. Upon final validation, the model showed decreases in esophageal mean dose (19.5 vs 18.2 Gy), Lung V20 (25.9% vs 25.1%), spinal cord max dose (34.9 vs 29.8 Gy).

Conclusion: Automated evaluation allowed for exploration of multiple optimization templates in a larger validation cohort than would have been feasible using a manual approach. A final KBP model was validated and found to be suitable for clinical use with no human intervention for most cases.


Not Applicable / None Entered.


TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation

Contact Email