# Real-time dose reconstruction and dose coverage forecasting using the magnetic resonance linear accelerator

## TL;DR

This paper presents a workflow for real-time dose reconstruction during MR-linac radiotherapy, enabling accurate dose monitoring and coverage forecasting.

## Contribution

The novel contribution is a real-time workflow for dose reconstruction and coverage forecasting during MR-linac treatments with low latency and high accuracy.

## Key findings

- The workflow achieved a latency of 150 ms and maintained real-time performance with 92% efficiency.
- Phantom measurements showed a 2.3% standard deviation from calculated dose, significantly better than the 13.1% for planned dose.
- Dose coverage forecasts were updated every 6.5 seconds, with increasing frequency as treatment progressed.

## Abstract

•Real-time dose reconstruction during magnetic resonance-linear accelerator treatment.•The latency of the presented workflow was 150 ms.•The phantom measurements were within a 2.3  % standard deviation of calculation.•The presented workflow updated accumulated dose and target coverage every 6.5 s.•The presented workflow supported rigid and deformable dose accumulation.

Real-time dose reconstruction during magnetic resonance-linear accelerator treatment.

The latency of the presented workflow was 150 ms.

The phantom measurements were within a 2.3  % standard deviation of calculation.

The presented workflow updated accumulated dose and target coverage every 6.5 s.

The presented workflow supported rigid and deformable dose accumulation.

Intrafraction patient motion during radiotherapy treatment affects the delivered dose, thereby, deviating from the planned dose. To monitor the deviation, we propose a workflow capable of performing and displaying real-time dose calculations during treatment with a magnetic resonance linear accelerator (MR-linac). Additionally, the workflow can forecast the accumulated dose and enables evaluation of dose guidance in real time for triggering automatic plan adaptation or intrafraction drift corrections (IDCs).

Information on the treated anatomy, rigid translations or three-dimensional (3D) volumes, and the linac state were collected to perform clinical grade dose calculations. Real-time dose reconstructions and the accumulated dose forecasts were performed for two dynamic scenarios. The first, rigid translations derived from two-dimensional orthogonal MR images where the calculated dose was compared to diode measurements. The second, deformable dose accumulation using deformable image registration, based on 3D MR imaging and compared to a dose film measurement.

The real-time dose reconstruction had a maximum standard deviation error of 2.3 % with the measured dose, compared to 13.1 % for the planned dose. The workflow maintained real-time performance using a single graphics processing unit up to 92 % of the time that the beam was on. Finally, dose coverage forecasts were updated every 6.5 s at the start of treatment, with the update frequency increasing towards the end of treatment.

We demonstrated the feasibility of displaying real-time delivered dose distributions and forecasted dose coverage using an MR-Linac. This enables triggering automatic plan adaptations or IDCs.