UC Irvine

Restricted transducer placement in clinical environments limits radiacoustic imaging (RAI) to incomplete angular measurements, producing characteristic artifacts—including spurious tails and missing features—that compromise dose reconstruction accuracy during radiation therapy monitoring. We address this limited-view challenge through two-stage reconstruction: parametric level-set optimization first identifies approximate dose region boundaries, then model-based inversion reconstructs detailed distributions within this spatially-constrained domain. Shape representation using radial basis functions reduces dimensionality from pixel-wise unknowns to geometric parameters describing piecewise-constant regions, while domain restriction concentrates measurement information onto fewer reconstruction variables. Validation studies encompass computational simulations and clinical LINAC experiments under limited as well as sparse angular coverage. In the simulated study, an elliptical source reconstructed from 120° detector aperture demonstrates improvements in ground-truth correlation from 0.86 (standard model-based) to 0.97 (shape-optimized). Experimental validation with clinical LINAC shows consistent gains: correlations versus full-view reconstructions increase from 0.87 to 0.90 for sparse four-detector arrays (144° coverage) and from 0.82 to 0.91 for seventeen-detector configurations ( with 96° coverage). Shape-based spatial priors substantially reduce limited-view artifacts while improving quantitative accuracy and can be a crucial tool for solving the limited-view problem in RAI.