On the use of Bayesian statistics in the application of adaptive setup protocols in radiotherapy

To propose adaptive setup protocols using Bayesian statistics that facilitate, based on a prediction of coverage probability, making a decision on which patients should follow daily imaging prior to treatment delivery.

The suitability of the treatment margins was assessed combining interfraction variability measurements of the first days of treatment with previous data gathered from our patient population. From this information, we decided if a patient needs an online imaging protocol to perform daily isocenter correction before each treatment fraction. We applied our method to five different datasets. Protocol parameters were selected from each dataset based on coverage probability, the expected imaging workload of the treatment unit, and the accuracy of patient classification. Time trends were assessed and included in the proposed protocols. To validate the accuracy of the protocols, they were applied to a validation dataset of prostate cancer patients.

Adaptive setup protocols lead expected population coverage >97% in all datasets analyzed when time trends were considered. The reduction in imaging workload ranged from 40% in lung treatments to 28.5% in prostate treatments. Results of the protocol on the validation dataset were very similar to those previously predicted.

The adaptive setup protocols based on Bayesian statistics presented in this study enable the optimization of imaging workload in the treatment unit ensuring that appropriate dose coverage remains unchanged.