Reconstruction of Velocity Fields from MRI Data Using Pyramidal Lucas–Kanade Sparse Optical Flowfor Abdominal Aortic Aneurysm Assessment

Technical Summary: Reconstruction of Velocity Fields from MRI Data Using Pyramidal Lucas–Kanade Sparse Optical Flow for Abdominal Aortic Aneurysm Assessment

Problem Statement
Abdominal Aortic Aneurysm (AAA) is a critical condition characterized by the dilatation of the abdominal aorta, carrying a high risk of rupture with significant mortality rates. While Phase-Contrast MRI (PC-MRI) is the gold standard for assessing hemodynamic velocity fields required for risk stratification, it necessitates specialized acquisition protocols. This study addresses the challenge of extracting essential hemodynamic information from standard, routinely acquired cardiac cine-MRI sequences without additional acquisition costs. The specific problem is the accurate reconstruction of two-dimensional velocity fields (both linear and angular) from these standard sequences to assess AAA hemodynamics.

Methodology
The research implements and evaluates the Pyramidal Lucas–Kanade (PLK) sparse optical flow algorithm to estimate motion vectors from cine-MRI data. The methodology is structured as follows:

• Data Source: Two 2D cine-MRI sequences (short-axis and long-axis views) from a single AAA patient were analyzed. Each sequence contained 25 frames covering one cardiac cycle.
• Feature Detection: The Shi–Tomasi corner criterion was used to identify robust features within the aortic lumen. Two initialization strategies were employed: "Line mode" (10 equidistant points) and "Full-lumen mode" (all mask pixels).
• Algorithm Configuration: The PLK algorithm utilizes a Gaussian image pyramid (4 levels) to handle large inter-frame displacements, processing from coarse to fine resolutions. Key parameters included 3 iterations per level and a convergence threshold of 0.03.
• Sensitivity Analysis: A systematic evaluation of the search window size (winSize) was conducted, testing ranges from (4, 4) to (30, 30).
• Drift Mitigation: A critical methodological contribution is the implementation of a "periodic reset" strategy. After each frame transition, tracked points are re-anchored to their initial positions to prevent the accumulation of tracking drift and the formation of artificial voids in the velocity field.
• Validation: The algorithm was first validated against synthetic data (a moving red disk with known ground truth) to determine optimal parameters. Subsequently, linear and angular velocities were computed for the patient data, converting pixel displacements to physical units (mm/s) using DICOM metadata.

Key Results

• Algorithm Validation: On synthetic data, the PLK algorithm achieved a minimum tracking error of approximately 2.79% at a winSize of (11, 11). The error rate was significantly lower than that of smaller windows (e.g., 7.98% for winSize (4, 4)), and convergence was achieved within 3 iterations.
• Optimal Window Size: Sensitivity analysis on patient data identified winSize values in the range of (10, 10) to (15, 15) as optimal. Smaller windows produced spurious large displacements, while larger windows (>20) over-smoothed the motion field, suppressing physiological peaks.
• Velocity Reconstruction:
  • Short-Axis View: The reconstruction revealed a consistent anti-clockwise vortical flow pattern. Average linear velocities ranged from 8.6 to 14.7 mm/s depending on the window size. A characteristic systolic peak was identified at T = 0.44 s.
  • Long-Axis View: The analysis captured pulsatile axial flow with peak systolic velocities between 23 and 33 mm/s. The systolic peak occurred earlier in this view, at T ≈ 0.20–0.25 s.
• Angular Velocity: Angular velocity analysis confirmed an anti-clockwise rotation (ω ≈ +0.13 to +0.22 rad/s) in the ascending aorta. The study noted that without the periodic reset strategy, the angular velocity exhibited a systematic clockwise drift, which is physiologically inconsistent.

Significance and Claims
The paper claims to demonstrate the feasibility of reconstructing hemodynamic velocity fields from standard cine-MRI sequences using the Pyramidal Lucas–Kanade algorithm, offering a cost-effective and non-invasive alternative to specialized PC-MRI.

The study asserts that the proposed framework provides:

1. Physiologically Consistent Data: The identification of anti-clockwise vortical flow and distinct systolic peaks aligns with known aortic hemodynamics and published literature on helical flow.
2. Robust Parameterization: The establishment of an optimal winSize range [(10, 10)–(15, 15)] and the necessity of a periodic reset strategy for accurate angular velocity estimation.
3. Clinical Potential: The ability to extract these metrics from routine scans suggests direct potential for clinical AAA surveillance and patient-specific rupture risk stratification without the need for additional imaging protocols.

The authors maintain a modest stance, acknowledging limitations such as the use of data from a single patient, the restriction to 2D analysis (lacking through-plane components), and the reliance on manual segmentation. They note that future work requires multi-centre validation, integration with deep learning for automation, and quantitative comparison with PC-MRI ground truth.