Scholarlink

Objectives <p>To develop an AI model using ultrasound features of carotid plaque for predicting the risk of acute ischemic stroke (AIS) and assess its efficacy in comparison with conventional regression prediction models.</p> Materials and methods <p>This study retrospectively included 923 patients who underwent US and CEUS examinations of carotid plaque at our institution. They were randomly divided into training + validation and test set in an 8:2 ratio. Additionally, 143 prospectively collected patients from three other centers were included as an external test set. Two expert radiologists described and documented the ultrasound images. Logistic regression analysis was used to analyze plaque ultrasound characteristics, leading to the establishment of statistical predictive models for AIS risk based on US alone and US combined with CEUS. AI models were developed using ResNet34 architecture trained on ultrasound images. ROC curves were generated, and AUC values were computed to compare the performance of the statistical models with the AI models.</p> Results <p>During a median follow-up of 5.3 years, 523 patients experienced AIS, while 543 had no history of stroke. The AUC was 0.719 for the model using US alone, 0.819 for the model combining US with CEUS, and 0.917 for the model incorporating AI, with all pairwise comparisons being statistically significant (<i>p</i> &lt; 0.05). Furthermore, the AUC of the AI model in the external test set was 0.866, indicating good generalization ability and stability.</p> Conclusion <p>US and CEUS characteristics of carotid plaque were strongly associated with AIS. Deep learning enhances AIS prediction in carotid plaque assessment using ultrasound.</p> Key Points <p><Emphasis Type="BoldItalic">Question</Emphasis> <i>Can AI enhance the prediction of AIS risk by analyzing the characteristics of carotid artery plaques using US and CEUS compared to conventional statistical models?</i></p> <p><Emphasis Type="BoldItalic">Findings</Emphasis> <i>The AI model integrating US and CEUS plaque features significantly outperformed traditional methods in AIS risk prediction, with CEUS providing substantial value beyond US alone.</i></p> <p><Emphasis Type="BoldItalic">Clinical relevance</Emphasis> <i>This AI-driven approach offers an automated and standardized method for stratifying stroke risk directly from ultrasound images of carotid plaques. This advancement facilitates early identification of high-risk individuals, eliminating the need for additional testing or labor-intensive manual analysis.</i></p> Graphical Abstract <p></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

Features of B-mode ultrasound and contrast-enhanced ultrasound of carotid plaque based on deep learning enhance the prediction of vulnerable plaques associated with acute ischemic stroke

  • 期刊论文

  • 发表日期：

    2026年2月21日

查看全文
  • Zhi-Ying Jin,
  • Chen-Yun Zhou,
  • Shi Tan,
  • Xiao-Zhou Fan,
  • Jun-Kang Li,
  • Shi-Yu Li,
  • Si-Si Huang,
  • Xiao-Meng Zou,
  • Rui-Lan Niu,
  • Nai-Qin Fu,
  • Yue-Ming Li,
  • Yu-Jiao Deng,
  • Zhi-Li Wang

摘要

Objectives

To develop an AI model using ultrasound features of carotid plaque for predicting the risk of acute ischemic stroke (AIS) and assess its efficacy in comparison with conventional regression prediction models.

Materials and methods

This study retrospectively included 923 patients who underwent US and CEUS examinations of carotid plaque at our institution. They were randomly divided into training + validation and test set in an 8:2 ratio. Additionally, 143 prospectively collected patients from three other centers were included as an external test set. Two expert radiologists described and documented the ultrasound images. Logistic regression analysis was used to analyze plaque ultrasound characteristics, leading to the establishment of statistical predictive models for AIS risk based on US alone and US combined with CEUS. AI models were developed using ResNet34 architecture trained on ultrasound images. ROC curves were generated, and AUC values were computed to compare the performance of the statistical models with the AI models.

Results

During a median follow-up of 5.3 years, 523 patients experienced AIS, while 543 had no history of stroke. The AUC was 0.719 for the model using US alone, 0.819 for the model combining US with CEUS, and 0.917 for the model incorporating AI, with all pairwise comparisons being statistically significant (p < 0.05). Furthermore, the AUC of the AI model in the external test set was 0.866, indicating good generalization ability and stability.

Conclusion

US and CEUS characteristics of carotid plaque were strongly associated with AIS. Deep learning enhances AIS prediction in carotid plaque assessment using ultrasound.

Key Points

Question Can AI enhance the prediction of AIS risk by analyzing the characteristics of carotid artery plaques using US and CEUS compared to conventional statistical models?

Findings The AI model integrating US and CEUS plaque features significantly outperformed traditional methods in AIS risk prediction, with CEUS providing substantial value beyond US alone.

Clinical relevance This AI-driven approach offers an automated and standardized method for stratifying stroke risk directly from ultrasound images of carotid plaques. This advancement facilitates early identification of high-risk individuals, eliminating the need for additional testing or labor-intensive manual analysis.

Graphical Abstract