272 - Real-Time Volumetric Radiation Dose Monitoring for Image-Guided Adaptive Radiotherapy Using Ionizing Radiation Acoustic Imaging
Presenter(s)
Y. Liu1, Y. Huang1, D. W. Litzenberg2, K. Cuneo3, I. El Naqa4, X. Wang1, L. Wei2, and W. Zhang1; 1University of Michigan, Ann Arbor, MI, 2Department of Radiation Oncology, University of Michigan, Ann Arbor, MI, 3University of Michigan, Ann Abor, MI, 4Machine Learning & Radiation Oncology, Moffitt Cancer Center, Tampa, FL
Purpose/Objective(s): Accurate delivery of the prescribed radiation dose to the target tumor volume is fundamental to the success of radiation therapy (RT), which has proven effective in cancer treatment and symptom palliation. However, intra-treatment uncertainties create substantial discrepancies between planned and delivered dose distributions, highlighting the need for real-time volumetric monitoring during dose delivery and for subsequent fraction guidance. This study presents an integrated, GPU-accelerated ionizing radiation acoustic imaging (iRAI) system for real-time volumetric radiation dose monitoring during treatment delivery, applicable for both proton and photon therapies.
Materials/Methods: We propose an integrated software–hardware system that combines the iRAI system, GPU acceleration techniques, and the delay-and-sum beamforming algorithm to enable real-time volumetric image reconstruction within a predefined region of interest. The integrated system comprises three primary components: (1) a programming environment-based data acquisition module, (2) a CPU-mediated data transfer pipeline, and (3) a GPU-based reconstruction engine. Radiation induced acoustic wave frames acquired at each radiation pulse are written via a C-MEX interface to shared host memory, read continuously by the transfer module, and processed by a voxel-wise parallel GPU kernel. Reconstructed volumes are accumulated in device memory and transferred to the host after acquisition, while raw frames are buffered in RAM and saved to disk for retrospective analysis. The system performance is evaluated with a customized C-shape treatment plan, which consists of 23 beam angles.
Results: The integrated system was deployed with full hardware-software synchronization and operated end-to-end when connected to the acquisition platform. For performance evaluation, pre-collected experimental data were replayed. On an NVIDIA RTX 3090 GPU, reconstruction of a 101×101×101 volume (0.5 mm isotropic voxels) achieved an average reconstruction time of 0.49 ms. The programming environment acquisition pipeline achieved an average frame publishing latency of 0.4 ms per 640×256 frame. The reconstructed C-shaped volume achieved a Dice coefficient of 0.828 at the 60% isodose level.
Conclusion: The proposed system supports pulse-by-pulse volumetric imaging at repetition rates higher than 1 kHz, which is compatible with both photon and proton accelerators. These results demonstrate the feasibility of real-time ionizing radiation acoustic imaging for intra-treatment verification and future image-guided adaptive radiation therapy.