Encrypted login | home

Program Information

A Hyper-Parameter Optimization Approach to Automated Radiotherapy Treatment Planning

no image available
S Haaf

S Haaf*, V Kearney , Y Interian , G Valdes , T Solberg , A Perez-Andujar , UCSF Comprehensive Cancer Center, San Francisco, CA


SU-I-GPD-T-318 (Sunday, July 30, 2017) 3:00 PM - 6:00 PM Room: Exhibit Hall

Purpose: Intensity-modulated-radiation-therapy (IMRT) and volumetrically-modulated-arc-therapy (VMAT) have become standard of care for many cancerous disease sites. Reaching clinical standard dosimetric criteria for these techniques, can be exceedingly time consuming, which can potentially compromise the dosimetric quality of the plan, and limit patient throughput. Most planning systems use some form of local optimization, which require several replanning events to meet dosimetric criteria. The intermediate replanning steps are often necessary to get to the final dosimetric outcome since local optimization is susceptible local minima. If the set of steps taken by planning personnel are synthesized then the clinical strain of IMRT and VMAT techniques could be drastically reduced. This study aims to automate the planning process by implementing a novel hyper-parameter based optimization algorithm.

Methods: This program interacts with an optimizer by taking a set of constraints, and priority pairs to generate a VMAT plan that minimizes the error in each voxel with respect to these constraints. The objective is to get the minimum prescribed dose to as much of the target as possible, while minimizing the percentage of the target and healthy tissues that receive more than the maximum prescribed dose. This program interacts with Nimble’s treatment planning system (TPS) to obtain dose and update constraints at every iteration, until all the dosimetric parameters are met and the system has reached convergence with respect to the penalty function.

Results: Hyper-parameter based optimization was able to achieve the lowest possible penalty function while simultaneously meeting all dosimetric criteria, within an average of 15 cycles.

Conclusion: Hyper-parameter based optimization is able to synthesize the treatment planning process for prostate plans without the need of clinician oversight. Although more disease sites are needed to fully test this system, Hyper-parameter based optimization represents an important step towards practical automated treatment planning.

Contact Email: