Automatic multi-plaque tracking and segmentation in ultrasonic videos.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Ultrasound

Background:

• Accurate carotid plaque tracking and segmentation are crucial for evaluating plaque properties and guiding treatment.
• Existing methods face challenges due to poor image quality, plaque variability, and multiple plaque detection.

Purpose of the Study:

• To develop an automatic multi-plaque tracking and segmentation (AMPTS) framework to overcome current limitations in carotid ultrasound analysis.
• To enhance the accuracy and robustness of carotid plaque detection and tracking in ultrasound videos.

Main Methods:

• Proposed a novel AMPTS framework with three modules: a Dual Attention U-Net for multi-object detection, single-object trackers with attention mechanisms, and a parallel tracking module.
• Employed a simplified 'tracking-by-detection' mechanism to handle variations in tracked objects.
• Compared the AMPTS framework against state-of-the-art deep learning methods.

Main Results:

• Achieved a Dice similarity coefficient of 0.83, outperforming existing methods by 0.16 to 0.27.
• Demonstrated significant improvements across seven other performance indicators.
• Attained a Dice score of 0.80 and high accuracy (0.79) on an additional testing set.

Conclusions:

• The proposed AMPTS framework offers high accuracy and generalizability for carotid plaque tracking and segmentation.
• Ablation studies confirmed the effectiveness of individual components, highlighting the framework's potential for clinical application.
• The method shows great promise for improving clinical practice in managing carotid artery disease.