Abstract
The objective of this study is to develop a deep-learning based detection and diagnosis technique for carotid atherosclerosis using a portable freehand 3D ultrasound (US) imaging system. A total of 127 3D carotid artery scans were acquired using a portable 3D US system which consisted of a handheld US scanner and an electromagnetic tracking system. A U-Net segmentation network was firstly applied to extract the carotid artery on 2D transverse frame, then a novel 3D reconstruction algorithm using fast dot projection (FDP) method with position regularization was proposed to reconstruct the carotid artery volume. Furthermore, a convolutional neural network was used to classify healthy and diseased cases qualitatively. 3D volume analysis methods including longitudinal image acquisition and stenosis grade measurement were developed to obtain the clinical metrics quantitatively. The proposed system achieved sensitivity of 0.714, specificity of 0.851 and accuracy of 0.803 respectively for diagnosis of carotid atherosclerosis. The automatically measured stenosis grade illustrated good correlation (r=0.762) with the experienced expert measurement. The developed technique based on 3D US imaging can be applied to the automatic diagnosis of carotid atherosclerosis.
Introduction
Carotid atherosclerosis, characterized by plaque buildup in the neck's primary arteries, is a significant predictor of stroke. While 2D ultrasound is a standard screening tool, its effectiveness is often limited by the operator's skill and a lack of 3D spatial context necessary for accurate plaque volume and arterial geometry assessment. High-end 3D imaging modalities like MRI or CT provide this context but are expensive and stationary. This study introduces an automated, portable freehand 3D ultrasound system designed to bridge this gap, providing high-quality reconstructions and reliable diagnoses regardless of operator experience.
System Architecture and Methods
The proposed framework transforms a handheld 2D ultrasound scanner into a 3D diagnostic tool through a multi-stage pipeline involving data acquisition, segmentation, and advanced 3D reconstruction.
- Data Acquisition: A handheld probe is paired with an electromagnetic tracking system to record spatial coordinates in real-time.
- Segmentation: A U-Net architecture extracts the Media-Adventitia Boundary (MAB, outer wall) and the Lumen-Intima Boundary (LIB, inner wall) from 2D frames.
- 3D Reconstruction with Regularization: To correct for manual scanning artifacts like hand-shaking, the team utilized a Fast Dot Projection (FDP) method integrated with Total Variation (TV) regularization. A "pose re-rank" algorithm using Principal Component Analysis (PCA) was introduced to handle non-linear probe movements, such as accidental backward motion during a scan.
Automated Diagnosis and Quantitative Analysis
A key feature of the system is its dual-stream Convolutional Neural Network (CNN) for classification. This module processes both the raw image data and the segmented masks to provide morphological context, improving the detection of diseased cases.
Beyond qualitative diagnosis, the system enables quantitative clinical metrics. It automatically calculates the stenosis grade (the degree of arterial narrowing) and supports "virtual sonography." This allows clinicians to generate longitudinal views of the 3D volume at specific angles (e.g., ±30°) that may have been missed during the physical examination, ensuring the most severe narrowing is identified.
Key Results
The system was evaluated using 127 3D scans from 83 subjects. Key performance metrics include:
- Segmentation Accuracy: The U-Net achieved a Dice Similarity Coefficient (DSC) of 95.67% for the outer wall and 94.17% for the inner wall.
- Diagnostic Performance: The system achieved a sensitivity of 0.714, a specificity of 0.851, and an overall accuracy of 0.803 for detecting atherosclerosis.
- Clinical Correlation: Automatically measured stenosis grades showed a strong correlation (r = 0.762) with measurements taken by experienced sonographers using high-end clinical equipment.
Conclusion
The proposed deep-learning based technique offers a convenient and reliable method for carotid atherosclerosis examination. By automating the reconstruction and diagnosis process, the system decreases dependence on clinician experience and provides a portable solution for 3D arterial assessment. Future work will focus on expanding the dataset and improving sensitivity for detecting smaller plaques.
Summary updated