Encrypted login | home

Program Information

A Knowledge Based Implementation of a Fully Automated Treatment Planning System

no image available
h wang

h wang*, L Xing, Stanford Univ School of Medicine, Stanford, CA

TH-C-137-9 Thursday 10:30AM - 12:30PM Room: 137

Purpose: Currently, there is no effective way to automate the treatment planning process. The most common approach is to use criteria from a selected plan from a database and apply those configurations to the plan you want to work with. We proposed a direct mapping between any two given plans. We found a way to cleverly use information from one plan to the next, regardless of how different or similar they are.

Methods: We have broken the entire automation process into two stages. In the first stage, we find a transformation between the two given patient. We find this transformation using the prescribed dose for the new plan and calculated dose from template plan. We apply this transformation and as a result provide a coarse tuning. In the second stage, we use a voxel-based penalty scheme to provide a finer tuning. The rationale behind this fine-tuning stage is no matter how similar two plans are, there are inevitable differences we cannot account for. We evaluate this novel knowledge based system against manual planning for assessment.

Results: We implemented a Visual Studio interface that automatically controls the Varian Eclipse's Beam Treatment Planning. No user intervention is needed for the entire planning duration. We compared the dose volume histogram between the automated plans and manual plan and observed nearly identical results. We also achieved up to 86x speed up between using knowledge from another plan versus starting completely from scratch.

Conclusion: Knowledge based system would transform the treatment planning process into a completely automated procedure. This work has the potential to be a promising platform for developing treatment plans and afford a powerful way to reliably automate the treatment planning process. It could have major predictive value to save clinical time in developing a cancer treatment plan.

Contact Email: