Related Experiment Video
Updated: Jul 1, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
533
Automated Spontaneous Echo Contrast Detection Using a Multisequence Attention Convolutional Neural Network.
Ouwen Huang1, Zewei Shi1, Naveen Garg2
1Department of Biomedical Engineering, Duke University, Durham, NC, USA.
Ultrasound in Medicine & Biology
|March 9, 2024
Summary
A deep learning model automatically identifies spontaneous echo contrast (SEC) in vascular ultrasounds, a finding linked to thromboembolism risk. This AI tool enhances SEC detection without needing expert input or extra clinician time.
Area of Science:
- Medical imaging analysis
- Artificial intelligence in healthcare
- Vascular ultrasound interpretation
Background:
- Spontaneous echo contrast (SEC) is a vascular ultrasound finding associated with increased thromboembolism risk.
- Current SEC identification requires expert determination and significant clinician time, posing barriers to its widespread adoption.
- Automated detection of SEC could streamline clinical practice and facilitate large-scale data analysis.
Purpose of the Study:
- To develop and evaluate a deep learning model for the automatic identification of spontaneous echo contrast (SEC) in femoral vein ultrasound images.
- To assess the model's performance using standard metrics and compare its efficacy with and without attention mechanisms.
- To provide a publicly available model and dataset to encourage further research and application in clinical practice.
Main Methods:
- A dataset of 801 archival femoral vein ultrasound acquisitions from 201 patients was curated.
- A multisequence convolutional neural network (CNN) with a ResNetv2 backbone was employed, incorporating soft attention for keyframe importance visualization.
- Model performance was evaluated using an 80/20 train/test split, reporting ROC-AUC, sensitivity, specificity, and F1 score.
Main Results:
- The deep learning model achieved an Area Under the Curve (AUC) of 0.74 with soft attention, demonstrating a sensitivity of 0.73 and specificity of 0.68.
- The model without soft attention yielded a lower AUC of 0.69, with sensitivity of 0.71 and specificity of 0.60.
- Attention visualizations indicated that the model assigns higher importance to ultrasound frames with greater vessel lumen, correlating with SEC presence.
Conclusions:
- A multisequence CNN model effectively identifies spontaneous echo contrast (SEC) in ultrasound keyframes with an AUC of 0.74.
- This automated approach can enable SEC screening applications and facilitate the discovery of more SEC-related data.
- The model eliminates the need for expert intervention and additional clinician reporting time, overcoming current barriers to SEC adoption.
