Abstract
Identification of vascular structures from medical images is integral to many clinical procedures. Most vessel segmentation techniques ignore the characteristic pulsatile motion of vessels in that formulation. In a recent effort to automatically segment vessels that are hidden under fat, we motivated the use of the magnitude of local pulsatile motion extracted from surgical endoscopic video. In this paper we propose a new approach that leverages the local orientation, in addition to magnitude of motion, and demonstrate that the extended computation of motion vectors can improve the segmentation of vascular structures. We implement our approach using two alternatives to magnitude-only motion estimation by using traditional optical flow and by exploiting the monogenic signal for fast flow estimation. Our evaluations are conducted on both synthetic phantoms as well as real ultrasound data showing improved segmentation results (0.36 increase in DSC and 0.11 increase in AUC) with negligible change in computational performance.
Introduction
Accurate identification of vascular structures is a cornerstone of many clinical procedures, particularly in robotic-assisted surgery where vessels may be obscured by layers of adipose tissue. Traditional segmentation methods typically rely on static image features such as intensity gradients or geometric "vesselness" filters. However, these approaches often falter in low-contrast environments or when vessels are physically hidden.
This research shifts the focus from static appearance to dynamic behavior. The defining characteristic of a vessel is its pulsation—a periodic radial distension. While previous motion-based methods focused solely on the magnitude of movement, they often struggled to distinguish true vessel pulsation from the bulk translation of surrounding tissues. By incorporating the direction of motion, specifically the radial expansion and contraction, vascular structures can be more uniquely identified.
Methodology
The proposed approach models vessels as tubular structures that expand and contract radially. This motion creates a vector field where the divergence is highest at the vessel's centerline during the expansion phase. To capture this, the authors developed two distinct pipelines: one based on traditional Optical Flow (OF) and another utilizing the Monogenic Signal (MON) framework.
The workflow consists of four primary stages:
- Motion Estimation: Calculating local motion vectors using either the high-accuracy OF or the computationally efficient MON framework.
- Temporal Bandpass Filtering: Isolating motion frequencies corresponding to the human heart rate to filter out physiological drift and noise.
- Radial Distension Modeling: Computing the divergence of the filtered motion field to highlight the vessel centerline.
- Multi-scale Integration: Applying the divergence operator across multiple spatial scales to detect vessels of varying diameters.
The MON framework is particularly noteworthy as it uses a 2D extension of the analytic signal (via Riesz transforms) to extract local orientation and phase information far more efficiently than the optimization-heavy OF method.
Experimental Results
The methods were evaluated using both synthetic phantoms and real-world ultrasound data of the Common Carotid Artery (CCA). The synthetic tests included scenarios with complex noise textures to simulate tissue clutter.
The results demonstrate a significant leap in performance over previous Phase-Based (PB) Eulerian methods. On real ultrasound data, the MON approach improved the average Area Under the Curve (AUC) from 0.84 to 0.95 and increased the Dice Similarity Coefficient (DSC) by 0.36.
Crucially, the MON method offers a massive computational advantage. For a standard 40-frame sequence, the MON pipeline completed in just 15 seconds, compared to nearly 20 minutes for the OF pipeline. This efficiency makes the MON-based radial distension model a viable candidate for real-time clinical applications.
Conclusion and Future Work
By leveraging the directional components of pulsatile motion, this work provides a robust and efficient means of segmenting vessels that are otherwise difficult to detect. The use of the monogenic signal framework allows for high-accuracy detection with near-real-time performance.
Future research will focus on addressing "gross out-of-plane motion" and extending the framework to 3D volumetric data, such as Cine MRI or 3D Ultrasound. Additionally, applying these algorithms directly to raw Radio Frequency (RF) ultrasound data could further enhance accuracy by avoiding the information loss inherent in standard B-mode image conversion.
Summary updated