From Narrow to Wide: Autoencoding Transformers for Ultrasound Bandwidth Recovery

2025-11-04

AI Summary

Low-cost ultrasound transducers suffer from narrow bandwidth, resulting in pulse broadening, loss of high-frequency signal details, and degraded image resolution. To address this, we propose a lightweight autoencoder Transformer architecture based on Tiny ViT that enables end-to-end spectral reconstruction from narrowband to broadband radiofrequency (RF) signals. A novel curriculum-weighted loss function is introduced to preserve phase fidelity and maintain real-time frame rates while enhancing generalization. Extensive phantom experiments demonstrate a 90% reduction in mean squared error (MSE), a 6.7 dB improvement in peak signal-to-noise ratio (PSNR), and a structural similarity index (SSIM) of 0.965. Critically, the method achieves significant image sharpening even on unseen high-resolution phantoms, confirming its clinical applicability and robustness across diverse scanning conditions.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Multi-modal Vision
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Responsible Web: Machine-in-the-loop, human agency and autonomy

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.

Problem

• Recovering broadband ultrasound signals from narrow-bandwidth probes
• Enhancing image quality through learned spectrogram mapping
• Enabling high-resolution ultrasound with existing low-cost hardware

Innovation

• Learns mapping from narrow to broadband ultrasound spectrograms
• Uses Vision Transformer autoencoder with curriculum-weighted loss
• Enhances image quality without sacrificing frame rate