Shape-optimized model-based reconstruction algorithm for radiacoustic imaging

We present a shape-optimized model-based reconstruction framework that addresses the limited-view problem in radiacoustic imaging arising from restricted detector placement, which creates severely ill-posed inverse problems that manifest as characteristic artifacts in reconstructed images. We first reconstruct an approximate dose-region boundary using a parametric level set–based shape optimization. Pointwise model-based dose reconstruction is then performed only within this shape-constrained region. By restricting reconstruction from the full imaging domain to only the pixels within the identified region, the number of unknowns is significantly reduced, transforming a severely ill-posed inverse problem into a much better-constrained one. We validate this methodology through computational studies and experiments using clinical x-ray and proton radiation sources under limited angular coverage data acquisition settings. Results demonstrate substantial artifact reduction and improved accuracy compared to standard model-based reconstructions, with a 3%–25% improvement in correlation coefficients across different imaging settings, with the more pronounced improvements occurring under stronger limited-view conditions. This shape-optimized approach provides a viable pathway for accurate radiation therapy monitoring under clinically realistic detector configurations.