High-fidelity three-dimensional reconstruction of musculoskeletal tissues via diffusion based ultrasonic computed...

Compared with computed tomography (CT) and magnetic resonance imaging (MRI), traditional ultrasound imaging suffers from limited spatial resolution, poor contrast and restricted field of view. While offering significant improvements in imaging quality, full waveform inversion (FWI) remains ill-posed, is highly dependent on the initial model, and requires prohibitively long reconstruction times for iterative inversions. Considering these challenges, we propose a diffusion-based framework for ultrasonic computed tomography (DiffUCT) that enables rapid and high-fidelity reconstruction of musculoskeletal tissues. DiffUCT employs a two-stage architecture integrating attention-enhanced feature extraction with progressive diffusion refinement. Extensive validation on lower extremity datasets shows that DiffUCT could reconstruct high-fidelity ultrasonic parametric imaging, with accurate visualization of the anatomical structures for musculoskeletal tissues (including the skin, muscular, adipose, skeletal, blood vessel tissues, etc.). The DiffUCT significantly outperformed conventional B-mode ultrasound, FWI and the leading deep-learning approaches in quality metrics (PSNR = 37.78 dB, SSIM = 0.9734, LPIPS = 0.0097). Critically, DiffUCT achieves this quality with 10-step accelerated sampling in only 0.126 s per slice (approximately 4000-fold faster than FWI based methods). Through rapid and precise slice-wise reconstruction, DiffUCT enables high-quality three-dimensional (3D) results with arbitrary cross-sectional visualization of tissue structures. The DiffUCT framework represents a significant advancement in ultrasound imaging for musculoskeletal tissues, delivering fast 3D reconstruction with visually detailed anatomical structures reminiscent of X-ray based CT, while preserving the inherent safety and accessibility advantages of ultrasound.