Quantification and visualization of carotid segmentation accuracy and precision using a 2D standardized carotid map

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Computational Anatomy

Background:

• Accurate vascular segmentation is crucial for assessing vessel wall and plaque burden.
• Existing evaluation methods may not sufficiently capture segmentation accuracy for vessel wall thickness (VWT).
• Segmentation of lumen and wall boundaries requires simultaneous consideration for reliable VWT assessment.

Purpose of the Study:

• To propose a statistical framework for evaluating vascular image segmentation algorithms.
• To quantify the difference in local vessel wall thickness (VWT) between manual and automated segmentation.
• To decompose VWT differences into contributions from lumen and wall boundary segmentation errors.

Main Methods:

• Developed statistical metrics to evaluate local VWT differences (ΔT) and boundary segmentation variability.
• Decomposed ΔT into local wall (ΔW) and lumen (ΔL) boundary differences.
• Segmented 3D carotid ultrasound images from 21 subjects five times manually and using a level-set algorithm.
• Computed and visualized difference measures and pooled standard deviations on 2D standardized maps.

Main Results:

• Quantified local accuracy and variability of the segmentation algorithm across subjects.
• Visualized segmentation performance on 2D standardized maps, highlighting areas of error.
• Demonstrated that ΔT can be approximated by ΔW and ΔL, indicating contributions of lumen and wall segmentation to VWT error.

Conclusions:

• The proposed framework effectively quantifies vascular segmentation accuracy and variability.
• The decomposition of VWT error provides insights into specific segmentation challenges (lumen vs. wall).
• Results on 2D maps can guide targeted improvements in segmentation algorithms, such as adjusting anchor points or force weights.