Encrypted login | home

Program Information

Comparison of Knowledge Based Planning Models Populated From Plans Created by Trial-And-Error Optimization Versus Prioritized Optimization Methodologies

no image available
Y Lin

Y Lin1*, S Berry2 , L Hong2 , M Hunt2 , (1) Hiroshima University, Hiroshima, Hiroshima, (2) Memorial Sloan Kettering Cancer Center, New York, NY


MO-RPM-GePD-TT-2 (Monday, July 31, 2017) 3:45 PM - 4:15 PM Room: Therapy ePoster Theater

Purpose: To investigate whether using plans constructed with prioritized optimization (PO) as input to a knowledge based planning (KBP) model would result in more efficient model creation with more consistent outputs than a model created using plans from a traditional, trial-and-error based, optimization (TO) technique.

Methods: Using a commercially available KBP package, three KBP models (M₁, M₂, and M₃) were constructed from various subsets of 60 post-prostatectomy IMRT patient plans. M₁ was created from 56 TO plans, selected to represent typical clinical variations in target and organ-at-risk sizes and shapes. M₂ and M₃ shared the same 30 patients but were populated with PO and TO plans, respectively. The plans comprising M₁ and M₃ were those that had been used for patient treatment while the M₂ plans were created for research purposes only. The three models were each applied to a new set of 18 patient scans and dose volume histogram estimates (DVHE’s) were generated for rectal and bladder walls and compared for each patient.

Results: On average, for the 18 evaluation patients, M₂ resulted in a lower estimated dose with a significantly tighter range in DVHE’s (p < 0.01) for both the rectal and bladder walls compared with M₁ and M₃. A tighter range in estimates indicates less uncertainty in the dose prediction. The average DVHE trended lower for M₂, consistent with the data input into each respective model, but did not reach significance.

Conclusion: Populating a KBP model with PO data resulted in a high quality model. Since PO plans can be generated automatically offline, a KBP-PO model can quickly adapt to clinical changes without having to wait for the accrual of sufficient numbers of clinical TO patient plans. This may facilitate the use of KBP approaches for initial or ongoing quality assurance procedures and plan audits.

Funding Support, Disclosures, and Conflict of Interest: S Berry and M Hunt hold research grants, unrelated to this project, from Varian Medical Systems. The KBP software used in this project is manufactured by Varian Medical Systems.

Contact Email: