Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Narrow to Wide: Autoencoding Transformers for Ultrasound Bandwidth Recovery

Sepideh KhakzadGharamaleki, Hassan Rivaz, Brandon Helfield

TL;DR

This work tackles the limited axial resolution of narrow‑band ultrasound by learning a spectral super‑resolution mapping from bandlimited RF data to broadband content. A Tiny Vision Transformer–based auto‑encoder operates on STFT spectrograms of RF lines, trained on synthetic Field II data with a curriculum‑weighted loss to recover missing high‑frequency components while preserving phase. The model achieves substantial gains (e.g., PSNR gains up to +6.7 dB and SSIM near 0.996) on speckle phantoms and generalizes to unseen resolution phantoms, suggesting it can retrofit existing narrow‑band probes with broadband‑like performance without hardware changes. This software‑driven approach holds promise for higher‑resolution ultrasound in resource‑constrained settings, with caveats related to amplitude estimation and computational memory that guide future improvements.

Abstract

Conventional pulse-echo ultrasound suffers when low-cost probes deliver only narrow fractional bandwidths, elongating pulses and erasing high-frequency detail. We address this limitation by learning a data-driven mapping from band-limited to broadband spectrogram of radio-frequency (RF) lines. To this end, a variation of Tiny Vision Transform (ViT) auto-encoder is trained on simulation data using a curriculum-weighted loss. On heterogeneous speckle-cyst phantoms, the network reduces image-domain MSE by 90 percent, boosts PSNR by 6.7 dB, and raises SSIM to 0.965 compared with the narrow-band input. It also sharpens point-target rows in a completely unseen resolution phantom, demonstrating strong out-of-distribution generalisation without sacrificing frame rate or phase information. These results indicate that a purely software upgrade can endow installed narrow-band probes with broadband-like performance, potentially widening access to high-resolution ultrasound in resource-constrained settings.

From Narrow to Wide: Autoencoding Transformers for Ultrasound Bandwidth Recovery

TL;DR

This work tackles the limited axial resolution of narrow‑band ultrasound by learning a spectral super‑resolution mapping from bandlimited RF data to broadband content. A Tiny Vision Transformer–based auto‑encoder operates on STFT spectrograms of RF lines, trained on synthetic Field II data with a curriculum‑weighted loss to recover missing high‑frequency components while preserving phase. The model achieves substantial gains (e.g., PSNR gains up to +6.7 dB and SSIM near 0.996) on speckle phantoms and generalizes to unseen resolution phantoms, suggesting it can retrofit existing narrow‑band probes with broadband‑like performance without hardware changes. This software‑driven approach holds promise for higher‑resolution ultrasound in resource‑constrained settings, with caveats related to amplitude estimation and computational memory that guide future improvements.

Abstract

Conventional pulse-echo ultrasound suffers when low-cost probes deliver only narrow fractional bandwidths, elongating pulses and erasing high-frequency detail. We address this limitation by learning a data-driven mapping from band-limited to broadband spectrogram of radio-frequency (RF) lines. To this end, a variation of Tiny Vision Transform (ViT) auto-encoder is trained on simulation data using a curriculum-weighted loss. On heterogeneous speckle-cyst phantoms, the network reduces image-domain MSE by 90 percent, boosts PSNR by 6.7 dB, and raises SSIM to 0.965 compared with the narrow-band input. It also sharpens point-target rows in a completely unseen resolution phantom, demonstrating strong out-of-distribution generalisation without sacrificing frame rate or phase information. These results indicate that a purely software upgrade can endow installed narrow-band probes with broadband-like performance, potentially widening access to high-resolution ultrasound in resource-constrained settings.

Paper Structure

This paper contains 10 sections, 6 equations, 3 figures, 3 tables, 1 algorithm.

Table of Contents

  1. INTRODUCTION
  2. METHODS AND MATERIALS
  3. Data Acquisition
  4. Low vs. High Bandwidth Simulation
  5. Short‑time Fourier Transform And Spectrogram Construction
  6. Network Architecture
  7. Curriculum--Weighted Multi--Term Loss
  8. Per--component and coupled objectives.
  9. Results
  10. Discussion and Conclusion

Figures (3)

  • Figure 1: Network architecture visualizing a Tiny ViT-based encoder-decoder structure that reconstructs the phase and magnitude of the higher-bandwidth spectrograms.
  • Figure 2: Performance on the speckle–cyst phantom. Columns show (left) broadband ground truth, (centre) network prediction, and (right) Low-BW input. The proposed method restores speckle granularity and cyst contrast that are suppressed in the narrow-band acquisition.
  • Figure 3: Performance on the unseen point-target resolution phantom. The network sharply reconstructs the sparse scatterer rows that are severely blurred in the Low-BW input, evidencing strong out-of-distribution generalisation.