文献速递技术与安全US

一种用于稀疏有限视角光声断层成像的可学习物理深度学习模型

A Learnable Physical Deep Learning Model for Sparse Limited View Photoacoustic Tomography.
PubMed 文献 · Ultrasound Med Biol · 2025-08-23

研究提出一种可学习物理深度学习模型,用于从稀疏有限视角光声阵列信号中重建高质量光声断层图像。

要点

  • 可学习物理深度学习模型可从稀疏有限视角光声信号重建高质量、无伪影光声断层图像。
  • 稀疏重建算法研究旨在缓解光声断层系统对大量阵元制造的需求。

核心结论

  • 可学习物理深度学习模型能够从稀疏有限视角光声阵列信号中实现高质量、无伪影的光声断层重建。

研究设计与样本

  • 研究提出可学习物理深度学习模型,通过数值模拟和在体实验验证稀疏有限视角光声信号的重建效果。

关键影像表现/征象

  • 极稀疏在体光声信号(155度内16个探测器)重建结果的峰值信噪比为20.700±2.404 dB。

诊断标准与关键数值

  • 在体极稀疏条件(155度内16个探测器)下,重建图像峰值信噪比为20.700±2.404 dB。

临床应用与阅片要点

  • 稀疏重建算法研究可能有助于缓解光声断层系统对大量阵元制造的需求。

局限性

  • 原文未报告局限性。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

The photoacoustic tomography system requires custom-built and extensive array elements to achieve high-sensitivity and high-contrast imaging in clinical practice. However, the manufacture of photoacoustic transducers with extensive elements is costly and limits their flexible clinical applications. In this work, we focused on the research of sparse reconstruction algorithms to alleviate the demand for manufacturing. A learnable physical deep learning model was proposed to achieve high-quality photoacoustic image reconstruction from signals collected by sparse and limited view photoacoustic array elements. Numerical simulation and in vivo experiments showed that the learnable physical deep learning model was able to achieve high-quality and artifact-free photoacoustic tomography reconstruction from sparse photoacoustic signals. Even with the extremely sparse in vivo photoacoustic signals (16 detectors in 155 degrees), considerable reconstruction results with a structural similarity of 0.728 ± 0. 029 and peak signal-to-noise ratio of 20.700 ± 2.404 dB could be achieved.

来源:Ultrasound Med Biol,查看原文。版权归原作者所有。

引用

  1. Lu M, Li B, Wang J, et al. A Learnable Physical Deep Learning Model for Sparse Limited View Photoacoustic Tomography. Ultrasound Med Biol. 2025. PMID: 40850901. doi:10.1016/j.ultrasmedbio.2025.07.014

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。