Online Dose Verification & Anomaly Detection Using Radiation Acoustic Imaging In Proton and Electron Radiotherapy
Abstract
Purpose
Ensuring precise delivery of radiotherapy is paramount, especially in high dose rate settings where small deviations can create large, unplanned doses, jeopardizing patient safety and treatment outcome. Current dosimetry and QA methods operate on signals analyzed pre or post treatment and lack the ability to verify intrafractional dose during treatment. We propose an automated, in-vivo dose monitoring and anomaly detection algorithm, using real-time signals from ionizing Radiation Acoustic Imaging (iRAI) under conventional and ultra-high dose rate (UHDR>40Gy/s) proton and electron beam delivery.
Methods
Two setups were designed to record acoustic signals in oil phantoms from incoming beams delivered at conventional (CONV~0.1Gy/s) and UHDR. In setup one, a phantom was coupled with a 2D planar matrix array (32x32 elements,0.35MHz). iRAI signal was acquired from a CONV proton beam (Synchrocylcotron,230MeV), and an experimental pseudo-continuous UHDR proton beam (40μs,106 MHz micro-bunches, >100Gy/s). In setup two, phantoms were coupled with a P4-1 transducer (1x96 elements,2.5MHz). One phantom was inserted with a Styrofoam ball as an air-equivalent treatment anomaly. Signal was acquired from a UHDR electron beam (Linac,9MeV). Two statistical anomaly detection (AD) algorithms were developed to detect deviation (“error”) from expected signal morphology: 1D timeseries optimal change point detection (OCD) and 2D matrix comparison using single value decomposition (SVD) and Shannon entropy metrics.
Results
Both AD algorithms showed automatic dose localization with 93% error detection accuracy in setup one across 27 paired experiments. Pairs differed in transducer distance to the Bragg peak and proton beam energy and current. In setup two, AD showed 100% error detection accuracy across 48 paired experiments. These pairs varied in field size and air bubble quadrant in the phantom.
Conclusion
We have developed and tested a method for detecting incidental errors in radiation dose delivery using AD techniques, enabling online dose verification and error detection in radiotherapy.