Abstract
The proposed method combines a highly accelerated single-breathhold compressed-sensing multi-slice CMR technique with a non-model-based 3D reconstruction to accurately and reproducibly measure LA volumes and function.
Introduction
The left atrium (LA) is a critical indicator of cardiac health, with its dilation linked to heart failure, hypertension, and stroke risk. Despite its importance, obtaining accurate 3D measurements of LA volume remains challenging. Standard Cardiovascular Magnetic Resonance (CMR) typically requires multiple breath-holds to cover the atrium, leading to "misregistration" errors if the patient's breath-hold position varies. To circumvent this, clinicians often rely on "model-based" shortcuts, such as the biplane area-length method, which assume the atrium is a simple ellipsoid. However, real hearts are irregularly shaped, and these models often significantly underestimate true volumes.
Methods
This study introduces a two-part solution to improve LA volume assessment:
- CS-cineCMR Acquisition: A highly accelerated compressed sensing (CS) cine sequence was employed. This technique allows the scanner to skip significant portions of data during acquisition and reconstruct the missing information mathematically. Consequently, five high-resolution slices covering the entire LA can be captured within a single 20-second breath-hold, eliminating inter-slice motion artifacts.
- Non-Model-Based Reconstruction: Rather than using geometric formulas, the researchers utilized a 3D reconstruction algorithm. LA contours were delineated on 2D slices to generate a 3D point cloud. A triangle mesh algorithm was then applied to "shrink-wrap" these points, calculating the volume of the resulting 3D object. This method respects the actual, often irregular, anatomy of the patient.
The method was validated using five manually carved potato-based phantoms designed to mimic various LA morphologies (e.g., oval, spherical, curved) and three patients with diverse cardiac conditions.
Results and Discussion
A key finding was the method's robustness to slice orientation. Traditional model-based imaging fails if slices are not perfectly aligned with the heart's long axis. In contrast, this non-model-based 3D reconstruction yielded accurate results regardless of whether slices were aligned or misaligned.
In phantom studies, the CS-cineCMR method showed a mean difference of only -4.73 ± 1.75 ml (approximately 8.7% underestimation) compared to gold-standard water displacement. In the patient cohort, mid-diastolic volumes closely matched high-resolution 3D references, with a mean difference of -2.66 ± 6.5 ml (3.0% underestimation). Furthermore, the cine nature of the sequence allows for the generation of full time-volume curves, enabling the calculation of functional indices like active emptying fraction from a single scan.
Conclusion and Limitations
The proposed single breath-hold CS-cineCMR approach provides a fast, accurate, and orientation-independent method for measuring LA volume and function. While promising, the study is limited by a small patient sample size (n=3) and the use of manual segmentation. Future developments focusing on automated, machine-learning-based segmentation and validation in larger clinical cohorts are necessary to integrate this workflow into routine clinical practice.
Summary updated