Room: AAPM ePoster Library
To evaluate and validate the use of a commercial Knowledge-Based Planning (KBP) program for predicting the highest achievable prescription dose (HAPD) for post-pleurectomy mesothelioma patients treated with hemithoracic IMRT.
Delivered IMRT/VMAT plans (1.8 Gy/fraction, 23-28 fractions) for 55 right and 42 left-sided patients were used to create a KBP model for each side. To validate these models, the differences between predicted and clinically planned mean organ doses (MODs) for lung (MLD), liver and heart were evaluated for an additional 7 right and 8 left-sided cases. For new cases, the planning scan, contours and proposed prescription are the inputs: HAPD is the highest prescription for which predicted MLD, adjusted to account for results of the validation cases, is below 20.5 Gy. The MLD and other objectives predicted by the KBP provide initial patient-specific optimization objectives for the new case. This approach was used to retrospectively replan 9 test cases (6 right, 3 left) for which the clinical plan disagreed with HAPD predicted by a different published method (6/9 cases) or where target coverage could be improved.
KBP reports objectives in approximately 5 seconds. The mean and standard deviations of the difference between predicted and clinical MODs for lungs, heart and liver for the right and left-sided validation plans are (159.2±73.8; 113.8±185.5; 339.6±271.4 cGy) and (174.5±51.0; 256.4±118.1; 155.3±144.0 cGy), respectively. 8/9 replanned test cases met all our clinical planning constraints at predicted HAPD; one met lung but not other normal tissue constraints at HAPD. With MLD approximately 20 Gy, the median difference between predicted and replanned MLD was -0.018 Gy (range -0.31 to 0.92 Gy).
KBP models built from previously treated cases efficiently predict the HAPD for hemithoracic IMRT plans to treat mesothelioma and provide starting optimization objectives for plans to achieve this prescription.
Funding Support, Disclosures, and Conflict of Interest: This research was funded in part through the NIH/NCI Cancer Center Support Grant P30 CA008748.
Not Applicable / None Entered.