Session Item

The aim of this study was to clinically implement a multivariate risk-adjusted Hotelling’s T2 control chart for monitoring the brainstem dose after adjusting for inter-patient variations and variations in patient anatomy. Out-of-control plans were investigated if re-optimization could lower the brainstem dose and improve the plan quality.

For the clinical implementation of the control chart, we acquired 80 head-and-neck VMAT plans previously treated at our institution. The control chart signaled four patients as out-of-control (P34, P43, P67, and P68). P34 and P43 were signaled as out-of-control since they were on the lower end of the brainstem DVH distribution across all the patients, and as a result judged as high-quality plans. For out-of-control patients with high brainstem dose, P67’s and P68’s plans were re-optimized to reduce the dose without compromising PTV coverage and worsening other treatment objectives. To reduce the brainstem dose in P67, three new planning structures were created. Structure 1 accounts for the brainstem volume inside the 42 Gy isodose, structure 2 accounts for the brainstem volume inside the 35 Gy isodose, and structure 3 accounts for the brainstem volume inside the 28.5 Gy isodose. Three new objectives for optimization were added: (structure 1: maximum dose = 41 Gy, structure 2: maximum dose = 33 Gy, and structure 3: uniform dose = 27.5 Gy). To reduce the brainstem dose in P68, five new objectives were added: (brainstem: maximum D2 = 35 Gy and maximum D20 = 25 Gy, PTV 6996: minimum dose = 69.96 Gy, PTV 6000: minimum dose = 60 Gy, and PTV 5700: minimum dose = 57 Gy). The prescription percentage was reduced from 96.5% to 95.9% to maintain the coverage to all three PTVs.

For P67, the maximum, mean, and minimum brainstem dose was reduced from 57.23 Gy to 47.04 Gy, 18.63 Gy to 16.26 Gy, and 3.57 Gy to 3.52 Gy, respectively. For P68, the maximum, mean, and minimum brainstem dose was reduced from 41.20 Gy to 38.79 Gy, 15.77 Gy to 14.92 Gy, and 3.02 Gy to 2.98 Gy, respectively.

The clinical implementation of the multivariate risk-adjusted Hotelling’s T2 control chart on our head-and-neck VMAT dataset shows that the control chart can monitor and detect treatment plans with unusual DVH points. Our results show that the out-of-control plans can be re-optimized to improve the plan quality prior to treatment delivery.