FRDD-Net: Automated Carotid Plaque Ultrasound Images Segmentation Using Feature Remapping and Dense Decoding.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease Research

Background:

• Automated segmentation of carotid plaques in ultrasound images is crucial for diagnosing cardiovascular and cerebrovascular diseases.
• Low ultrasound image quality and plaque heterogeneity present significant challenges for current segmentation methods.

Purpose of the Study:

• To develop a novel deep convolutional neural network, FRDD-Net, for accurate and robust automated carotid plaque segmentation.
• To address the limitations of existing methods by improving feature extraction and utilization.

Main Methods:

• Proposed FRDD-Net, an encoder-decoder deep convolutional neural network incorporating Feature Remapping Modules (FRMs).
• Introduced a dense decoding mechanism to enhance encoded feature utilization.
• Developed a compound loss function to improve network robustness.

Main Results:

• FRDD-Net achieved superior performance in segmenting carotid plaques across multiple ultrasound datasets compared to state-of-the-art methods.
• Ablation studies confirmed the effectiveness of the proposed architectural components.

Conclusions:

• FRDD-Net offers a promising solution for automated carotid plaque segmentation, enhancing diagnostic capabilities for cardiovascular and cerebrovascular diseases.
• The proposed FRMs and dense decoding mechanism significantly contribute to the model's performance and robustness.