Scholarlink

Purpose <p>Acute abdominal aortic dissection (AD) is a serious disease. Early detection based on ultrasound (US) can improve the prognosis of AD, especially in emergency settings. We explored the ability of deep learning (DL) to diagnose abdominal AD in US images, which may help the diagnosis of AD by novice radiologists or non-professionals.</p> Methods <p>There were 374 US images from patients treated before June 30, 2022. The images were classified as AD-positive and AD-negative images. Among them, 90% of images were used as the training set, and 10% of images were used as the test set. A Densenet-169 model and a VGG-16 model were used in this study and compared with two human readers.</p> Results <p>DL models demonstrated high sensitivity and AUC for diagnosing abdominal AD in US images, and DL models showed generally better performance than human readers.</p> Conclusion <p>Our findings demonstrated the efficacy of DL-aided diagnosis of abdominal AD in US images, which can be helpful in emergency settings.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deep learning-aided diagnosis of acute abdominal aortic dissection by ultrasound images

  • 期刊论文

  • 发表日期：

    2025年1月17日

查看全文
  • Zhanye Lin,
  • Jian Zheng,
  • Yaohong Deng,
  • Lingyue Du,
  • Fan Liu,
  • Zhengyi Li

摘要

Purpose

Acute abdominal aortic dissection (AD) is a serious disease. Early detection based on ultrasound (US) can improve the prognosis of AD, especially in emergency settings. We explored the ability of deep learning (DL) to diagnose abdominal AD in US images, which may help the diagnosis of AD by novice radiologists or non-professionals.

Methods

There were 374 US images from patients treated before June 30, 2022. The images were classified as AD-positive and AD-negative images. Among them, 90% of images were used as the training set, and 10% of images were used as the test set. A Densenet-169 model and a VGG-16 model were used in this study and compared with two human readers.

Results

DL models demonstrated high sensitivity and AUC for diagnosing abdominal AD in US images, and DL models showed generally better performance than human readers.

Conclusion

Our findings demonstrated the efficacy of DL-aided diagnosis of abdominal AD in US images, which can be helpful in emergency settings.