Physics guided deep learning improves ultrasound imaging from sparse data
Physics-Guided Multi-Objective Deep Learning for Ultrasound RF Data Interpolation in Resource-Constrained Imaging
Machine Learning
Summary
Ultrasound machines often collect fewer data points to save power and hardware resources, but this can make images blurry or noisy. The authors developed a deep learning method that uses physical knowledge of how ultrasound works combined with data-driven training to fill in the missing data. Their approach trains the system to improve both the raw signals and the image quality, handling different ways of missing data automatically. Tests showed the method can reconstruct detailed ultrasound images even when using very sparse data, helping portable and wearable ultrasound devices produce better images.
What this means in practice
- •For medical device engineers: Build portable ultrasound machines that reconstruct high-quality images from fewer data points, saving power and hardware costs without losing image detail.$Commercial implications: Enables development of cost-effective ultrasound devices for hospitals and clinics by improving image quality with sparse data acquisition.
- •For wearable technology developers: Create wearable ultrasound sensors that use sparse scanning and the proposed deep learning model to produce clear images despite low data availability.
Authors
Luoyuan Zhang, Yiyang You, Ananya Tandri, Yinan Feng, Hyunwoo Song, Jeeun Kang, Youzuo Lin
Abstract
Ultrasound imaging increasingly targets portable, point-of-care, and wearable settings where constraints on power, bandwidth, and hardware complexity often necessitate sparse data acquisition in spatiotemporal scanning. However, image reconstruction using the sparse data can introduce insufficient phase information in coherent beamforming process, resulting in grating-lobe artifacts that degrade imaging contrast resolution. We present a physics-guided, data-driven framework for sparse-to-dense radio-frequency (RF) reconstruction that aligns training with downstream image formation. Our approach trains an end-to-end interpolation network using a hybrid supervision scheme that combines an RF-domain and a beamforming-domain loss with exponential moving average (EMA) to stabilize the multi-objective training. To improve generalization under variable acquisition layouts, we also introduce a random-skip masking strategy that varies sparsity patterns during training so a single model can handle diverse decimation factors and irregular channel configurations. We evaluate the framework on a held-out test set using the mean structural similarity index measure (SSIM) between reconstructed and ground-truth beamformed images. Across decimation factors $\times 2$ to $\times 13$, the best-performing configuration maintains mean SSIM around 0.95. Overall, the results show consistent gains in RF reconstruction and post-beamforming image quality across diverse acquisition conditions. This approach enables robust, high-quality ultrasound imaging at resource-constrained settings by allowing more sparse scanning in spatiotemporal domain.