跳到正文
原文
Nature 健康护理主题·· 5 天前AI 评分41

单元件超声传感器信号深度学习分析用于乳腺肿块检测:概念验证研究

Deep learning analysis of echo signals from a single piezoelectric element for breast mass detection: a proof-of-concept study

AI 导读

该研究利用深度学习分析单元件超声传感器信号,旨在区分含有肿块的乳腺区域与正常组织。在200名患者中,通过自由扫查获取信号并重建二维图像,结果显示,与ImageNet初始化相比,使用公共B模式图像预训练的模型在额外验证中表现更佳,尤其对于ViT-B/16模型,优势随训练数据减少而增大。该方法需进一步验证无已知乳腺病变女性群体,以考虑用于筛查。

正文

We’re sharing this article early to provide faster access to peer-reviewed, accepted research. It is citable and carries a permanent DOI. This version is subject to further edits and will be replaced automatically by the final Version of Record. All legal disclaimers apply.

Abstract

A palpable mass is a common presenting symptom of breast cancer, but neither breast self examination nor clinical breast examination is sensitive enough to detect breast masses early. We hypothesized that deep learning analysis of signals from a single-element (A-mode) ultrasound sensor can differentiate breast regions containing a mass from normal breast tissue. A-mode signals were prospectively acquired from 200 patients (192 malignant and 8 benign or borderline lesions). Signals collected during a freehand sweep were stacked into two-dimensional images (reconstructed A-mode images), whose appearance varies with probe motion. Self-supervised pre-training on public B-mode images was compared with ImageNet initialization, and an acquisition-specific augmentation (Variable Speed and Pressure Augmentation, VSPA) with conventional augmentation. The first 131 patients were used for model development and the next 69 for additional validation. On additional validation, ultrasound pre-training gave higher region-level AUCs than ImageNet initialization (ResNet-50, 93.6% versus 91.7%; ViT-B/16, 96.3% versus 93.2%) and more mass-focused Grad-CAM heatmaps. For ViT-B/16, the advantage grew as training data decreased. Conventional spatial augmentation gave most of the gain (ViT-B/16, 96.8% versus 93.2%), and VSPA added little. Further validation in women without known breast lesions is needed before this approach can be considered for screening.

Acknowledgments

This study was supported by a grant from the Technology Development Program (RS-2024-00468475), funded by the Ministry of SMEs and Startups (MSS), Republic of Korea. We thank all the participants who contributed to this study. The authors take full responsibility for the content and integrity of the work.

Funding

This study was funded by a grant from the Technology Development Program (RS-2024-00468475), funded by the Ministry of SMEs and Startups (MSS), Republic of Korea.

Author information

Author notes

  1. Yunna Lee and Jinyoung Byeon contributed equally to this work.

  2. Hyeong-Gon Moon and Young-Gon Kim are co-senior authors.

Authors and Affiliations

  1. Department of Transdisciplinary Medicine, Seoul National University Hospital, #101, Daehak-ro, Jongno-gu, Seoul, 03080, Republic of Korea

    Yunna Lee, Youmin Shin & Young-Gon Kim

  2. Interdisciplinary Program in Medical Informatics, College of Medicine, Seoul National University, Seoul, Korea

    Yunna Lee

  3. Department of Surgery, Chungbuk National University Hospital, Cheongju, Korea

    Jinyoung Byeon

  4. College of Bioengineering, Interdisciplinary Program in Bioengineering, Seoul National University, Seoul, Korea

    Youmin Shin

  5. Seoul AI School, a Seoul School of Integrated Sciences and, Technologies University, Seoul, Korea

    Youmin Shin

  6. Metropolitan Government–Seoul National University Boramae Medical Center, Seoul, Korea

    Jong-Ho Cheun

  7. Department of Medicine, Seoul National University College of Medicine, Seoul, Korea

    Jong-Ho Cheun, Hyeong-Gon Moon & Young-Gon Kim

  8. Department of Surgery, Seoul National University College of Medicine, #101, Daehak-ro, Jongno-gu, Seoul, 03080, Republic of Korea

    Changjin Lim, Changhoon Lee & Hyeong-Gon Moon

  9. L’imagin Inc, Seoul, Korea

    Kyoung Min Kim, Hyeong-Gon Moon & Young-Gon Kim

  10. Medical Research Center, Genomic Medicine Institute, Seoul National University College of Medicine, Seoul, Korea

    Hyeong-Gon Moon

  11. Cancer Research Institute, Seoul National University College of Medicine, Seoul, Korea

    Hyeong-Gon Moon

Authors

  1. Yunna Lee
  2. Jinyoung Byeon
  3. Youmin Shin
  4. Jong-Ho Cheun
  5. Changjin Lim
  6. Changhoon Lee
  7. Kyoung Min Kim
  8. Hyeong-Gon Moon
  9. Young-Gon Kim

Corresponding authors

Correspondence to Hyeong-Gon Moon or Young-Gon Kim.

Ethics declarations

Competing interests

YGK, HGM, and KMK are affiliated with L’imagin Inc., a company developing portable ultrasound technologies that may benefit from the commercial application of this research. All other authors declare no competing interests.

Ethics statement

Between October 2022 and January 2024, we enrolled a total of 200 patients scheduled for breast surgery for our study. All participants provided written informed consent for the study. This prospective study was approved by the Institutional Review Board of the participating institutions (IRB no. 2209-039-1357 for Seoul National University Hospital and IRB no. 20-2023-45 for SMG-SNU Boramae Medical Center) and was conducted according to the Declaration of Helsinki. This trial study was retrospectively registered with ClinicalTrials.gov (identifier: NCT06255808).

Additional information

Publisher's note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Lee, Y., Byeon, J., Shin, Y. et al. Deep learning analysis of echo signals from a single piezoelectric element for breast mass detection: a proof-of-concept study. Sci Rep (2026). https://doi.org/10.1038/s41598-026-78663-3

Download citation

  • Received: 16 March 2026

  • Accepted: 27 October 2026

  • Published: 15 November 2026

  • DOI: https://doi.org/10.1038/s41598-026-78663-3

Keywords

来源:Nature 健康护理主题 · nature.com