Answers are grounded in this paper and cite their sources.
As a promising imaging modality, ultrasound computed tomography (USCT) holds significant clinical value by providing quantitative parametric maps, such as speed of sound (SOS), in addition to high-resolution anatomical structures. While full-waveform inversion (FWI) is optimal for high-fidelity SOS reconstruction, its severe non-convexity frequently causes traditional discrete grid-based methods to trap in local minima, yielding anatomical artifacts that hinder clinical translation. Here, we propose POLARIS, an unsupervised continuous neural field inversion framework that utilizes spatial continuity as implicit regularization to stabilize the complex FWI optimization. Tailored to the circular acquisition geometry of USCT, we design a scale-aware radially-modulated polar encoding that effectively mitigates radial-angular sampling non-uniformity. We leverage the deterministic nature of the coupling background velocity to extract anatomical boundaries directly from acoustic signals, serving as a structural prior to preserve tissue integrity. Optimization is facilitated by a stochastic shot-batched optimization strategy, leveraging global parameter coupling to update the SOS field from subset emissions and accelerate convergence. Extensive validations on numerical and in vivo breast and musculoskeletal data demonstrate the POLARIS's higher accuracy and structural fidelity. Comparisons with MRI confirm the high anatomical consistency of the reconstructed SOS images. Notably, the framework maintains robust reconstruction performance under sparse-view configurations, highlighting its potential for accelerated USCT acquisition with reduced measurement requirements.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: