Automation to facilitate optimisation of breast radiotherapy treatments using EPID-based in vivo dosimetry

Abstract

Objective:
To use automation to facilitate the monitoring of each treatment fraction using an EPID based in-vivo dosimetry (IVD) system, allowing optimisation of breast radiotherapy delivery for individual patients and cohorts.

Approach:
A suite of in-house software was developed to reduce the number of manual interactions with the commercial IVD system, Dosimetry Check. An EPID specific pixel sensitivity map facilitated use of the EPID panel away from the central axis. Point dose difference and the change in standard deviation in dose were identified as useful dose metrics, with standard deviation used in preference to gamma in the presence of a systematic dose offset. Automated IVD was completed for 3261 fractions across 704 patients receiving breast radiotherapy. 

Main results:
Multiple opportunities for treatment optimisation were identified for individual patients and across patient cohorts as a result of successful implementation of automated IVD. 5.1% of analysed fractions were out of tolerance with 27.1% of these considered true positives. True positive results were obtained on any fraction of treatment and if IVD had only been completed on the first fraction, 84.4% of true positive results would have been missed. This was made possible due to the automation that saved over 800 hours of manual intervention and stored data in an accessible database.

Significance:
An improved EPID calibration to allow off-axis measurement maximises the number of patients eligible for IVD (36.8% of patients in this study). We also demonstrate the importance in selecting context-specific assessment metrics and how these can lead to a managable false positive rate. We have shown that the use of fully automated IVD facilitates use on every fraction of treatment. This leads to identification of areas for treatment improvement for both individuals and across a patient cohort, expanding the uses of IVD from simply gross error detection towards treatment optimisation.