Direct beamforming for heterogeneous photoacoustic computed tomography: a two-dimensional benchmark

Direct beamforming methods in photoacoustic computed tomography are often compared under idealized conditions, making it difficult to distinguish artifact suppression from the preservation of weak structures. We benchmark seven direct beamforming methods and four limited-view adaptive minimum-variance variants in heterogeneous two-dimensional photoacoustic computed tomography simulations that incorporate fluence variation, full-wave acoustic propagation, finite-aperture detection, measurement noise, and incomplete angular coverage. Eleven phantoms were evaluated over multiple signal-to-noise ratios using phantom-level statistics. The delay-and-sum method most consistently preserved extended low-amplitude structures, achieved the highest peak signal-to-noise ratio among the direct methods, and showed the lowest mean globally scaled amplitude error. Coherence-, variance-, and multiplication-based weighting reduced clutter and often narrowed point responses, but also increasingly attenuated weak targets. Under severe full-view noise, delay-multiply-and-sum and delay-and-sum-to-delay-standard-deviation-factor weighting improved structural similarity while reducing peak signal-to-noise ratio and the Dice coefficient. In limited-view imaging, spatially smoothed minimum-variance beamforming produced small gains in structural similarity at 0 and −5 dB but lower peak signal-to-noise ratio, Dice coefficient, and generalized contrast-to-noise ratio. Sound-speed perturbations did not alter the leading ranking of the direct methods, while spatial point-target analysis revealed strong anisotropy. Cleaner or narrower reconstructions did not necessarily provide better preservation of structure or amplitude.