Purpose: We present updates on knowledge-based planning (KBP) driven linac-based total marrow irradiation (TMI). A KBP model was developed to handle multiple-isocenter volumetric modulated arc therapy (VMAT) covering a large target volume from head to mid femur.
Methods: Fifty-one patients previously treated in our clinic were selected for the model training, while 12 patients from another clinic were used for validation. All plans used a 3-isocenter, 13-arc VMAT technique to cover sub-target volumes of head and neck (HN), chest, and pelvis. Both the HN and pelvis plans used the chest plan as the base dose to reduce hot spots around the field junctions. This leads to a wide range in dose-volume histogram (DVH) plots of target volumes. To address this, re-optimized plans without the base-dose plan were added to the library to train the KBP model. The performance of KBP-TMI model was then tested using a cohort of 11 clinical plans that did not use KBP. Conformity index and homogeneity index were evaluated using V95%/planning target volume (PTV) and D0.03cc/Rx dose, respectively.
Results: KBP achieved our clinical goals (Rx dose = D95%) within 2-3 iterative optimizations, which used to take 4-6 days of effort without KBP. Both KBP and clinical plans had mean dose values in the range of 108%-112% of the prescribed dose for PTVs, 22%-30% for lenses and oral cavity, and 41%-64% for brain, heart, lungs, bowels, liver, kidneys, eyes. Statistically significant reductions with KBP were observed in the mean dose values to brain, lungs, eyes and lenses. KBP substantially improved conformity index (1.5 ± 0.1 vs 1.8 ± 0.1) and homogeneity index (1.25 ± 0.02 vs 1.33 ± 0.03).
Conclusion: Our study demonstrates that VMAT-TMI powered by KBP improves dosimetric performance with uniform quality. Moreover, it improves planning time considerably to allow its wide-spread clinical implementation.
Intensity Modulation, Radiation Therapy, Treatment Planning
TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation