Author: Yong Chen, Gilberto Gonzalez, Omprakash Gottam, Liangzhong Xiang 👨🔬
Affiliation: University of California, Irvine, Koneru Lakshmaiah Education Foundation, University of Texas at San Antonio., University of Oklahoma Health Science Center 🌍
Purpose: Clinics currently lack techniques for real-time, in vivo monitoring of radiation therapy (RT), which is desirable for precise treatment. Radiation-induced acoustic computed tomography (RACT) offers strong potential for this purpose and can integrate seamlessly into clinics. However, transducer placement in clinical settings is limited, and traditional RACT algorithms yield strong limited-view artifacts. Here, we address this challenge.
Methods: Temporally varying radiation (LINACs, synchrocyclotrons) generates radiation-induced ultrasound (RUS) through thermoelastic expansion. RUS is captured by transducers and processed by RACT algorithms to reconstruct dose maps. Our linear optimization-based model-based (MB) algorithms, reduce artifacts compared to traditional methods but face challenges in highly limited-view geometries. To overcome this, we employ shape-based optimization (SO) using level-set functions to define irradiated region boundaries. By reducing the search space, the SO enables more accurate reconstruction, allowing the MB algorithm to significantly improve dose map quality. We validated the principle using RayStation-simulated doses for 107.66MeV protons (Hyperscan S250i, Mevion), generating acoustic signals in k-Wave on a 16-element linear array (length: 5cm). These signals were reconstructed using the SO+MB algorithm, which was then applied to experimental proton-RUS data. To demonstrate versatility, it was also tested on X-ray-RUS datasets (120° and 90° views) acquired with a single-element-transducer in a circular scan.
Results: In proton-RACT simulation, the SO+MB algorithm achieved a correlation coefficient (CC) of 0.77, compared to 0.52 for the traditional MB algorithm. Experimental results showed corresponding CCs of 0.73 and 0.46, respectively. In X-ray RACT, the standard MB algorithm yielded CCs of 0.90 and 0.84 for 120° and 90° views, while SO+MB improved these to 0.93 for both. The SO+MB algorithm significantly reduced limited-view artifacts compared to the standard MB reconstructions.
Conclusion: The SO+MB algorithm addresses RACT's biggest challenge: "limited-view problem", enhancing image reconstruction and thus advancing its potential to transform radiation therapy and monitoring practices.