Related Experiment Video
Updated: Jun 8, 2025
10:41
Analysis of Cerebral Vasospasm in a Murine Model of Subarachnoid Hemorrhage with High Frequency Transcranial Duplex Ultrasound
Published on: June 3, 2021
3.6K
LSMD: Long-Short Memory-Based Detection Network for Carotid Artery Detection in B-Mode Ultrasound Video Streams
Summary
A new deep learning model, the long-short memory-based detection (LSMD) network, accurately detects carotid artery anatomy and atherosclerotic plaques in ultrasound videos. This tool improves diagnostic accuracy for conditions like type II diabetes with real-time performance.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Cardiovascular Disease
Background:
- Carotid atherosclerotic plaques are a significant complication of type II diabetes.
- Carotid ultrasound is crucial for diagnosing carotid vascular disease.
- Inconsistent image capture and plaque identification by less experienced physicians pose challenges in primary hospitals.
Purpose of the Study:
- To develop a novel deep learning approach for automated carotid artery detection and plaque localization in ultrasound video streams.
- To enhance the efficiency and accuracy of identifying critical anatomical structures and atherosclerotic plaques.
Main Methods:
- Proposed a long-short memory-based detection (LSMD) network utilizing short-term temporal aggregation (STA) and long-term temporal aggregation (LTA) modules.
- Implemented memory buffers with a dynamic updating strategy to optimize temporal receptive field coverage and computational efficiency.
- Trained and evaluated the LSMD model on carotid ultrasound videos, comparing its performance against a single shot multibox detector (SSD) baseline.
Main Results:
- The LSMD network demonstrated significant improvements over the SSD baseline, with precision, recall, average precision (AP), and mean AP (mAP) increases of 6.83%, 12.29%, 11.23%, and 13.21%, respectively.
- Achieved real-time inference speeds: 6.97 ms on a high-end GPU and 29.69 ms on an edge device.
- The model accurately localized carotid anatomy and atherosclerotic plaques.
Conclusions:
- The proposed LSMD network offers a robust and efficient solution for automated carotid artery and plaque detection in ultrasound imaging.
- Its real-time inference capabilities and improved accuracy suggest significant potential for enhancing clinical diagnostic accuracy, particularly in settings with varying physician experience levels.
