R Foster¹*, A Magliari², R Clark³, (1) Northwestern Medicine Chicago Proton Center, Warrenville, IL, (2) Varian Medical Systems Inc, (3) Varian Medical Systems Inc,

• 
  R Foster

MO-E115-IePD-F3-2 (Monday, 7/11/2022) 1:15 PM - 1:45 PM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 3

Purpose: Treatment planning for hippocampal sparing whole brain (HSWB) radiation therapy cases is complex and time consuming. Knowledge based planning (KBP) tools expedite and streamline treatment planning. In 2016, a HSWB model was made publicly available (HSWBv1). From this model, we developed a new HSWB KBP model (HSWBv2) using recursive process and newly developed scorecard to improve hippocampal sparing, homogeneity, and conformity. This research intends to assist clinicians develop and implement their own robust KBP models.

Methods: Anonymized whole brain DICOM data sets were obtained from patients treated per NRG-CC001 protocol. New plans were created with HSWBv1 and then manually reoptimized to improve homogeneity and conformity while reducing OAR dose. Dosimetric improvements in reoptimized plans were quantified using a new scorecard tool. Manually improved plans became the training set for the first iteration HSWBv2. Recursive model creation was employed ensuring final HSWBv2 training data consisted of plans generated from the initial HSWBv2 model. Case selection, including removing outliers, and retuning global optimization objectives resulted with the HSWBv2 training set consisting of 42 cases.

Results: The new scorecard integrated five years clinical experience using HSWBv1 helping guide the training process of HSWBv2. The refined scorecard allowed efficient, thorough evaluation of dosimetric plan quality. Recursive processing utilizing high quality plans generated a model with narrow, accurate dose volume histogram (DVH) prediction bands. Narrow estimation bands allow for aggressive objectives with improved OAR sparing. HSWBv2 and HSWBv1 models had an average hippocampal D100% of 5.75 Gy and 6.46 Gy and PTV V105% of 6.44% and 35.35% respectively.

Conclusion: HCSWBv2 model efficiently creates high quality plans providing low variability specific to scorecard NRG-CC001 metrics with improved homogeneity/conformity. KBP model creation emphasizes the effectiveness of utilizing recursive training while evaluating dosimetric quality using a scorecard.

Keywords

Not Applicable / None Entered.

Taxonomy

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

Contact Email