Left ventricular four-dimensional rotational angiography with low radiation dose through interphase registration

    loading  Checking for direct PDF access through Ovid

Abstract

Aims

Ventricular tachycardia ablations could benefit from four-dimensional (4D) (dynamic 3D) visualization of the left ventricle (LV) as roadmap for anatomy-guided procedures. Our aim was to develop an algorithm that combines information of several cardiac phases to improve signal-to-noise ratio in low-dose, noisy rotational angiography [three-dimensional rotational angiography (3DRA)] image datasets, enabling semi-automatic segmentation and generation of 4D rotational angiography (4DRA) LV surface models.

Methods and results

We developed a novel slow pacing protocol for low-dose 4DRA imaging and applied interphase registration (IPR) to improve contrast-to-noise ratio (CNR) such that 4D LV segmentation could be achieved using a single iso-intensity value (ISO). The method was applied to construct four-phase dynamic LV models from five porcine experiments. Optimal choice of IPR and ISO parameters and resulting LV model accuracy were assessed by comparison with ‘groundtruth’ manual LV delineations using surface distance measures [root mean square distance (RMSD), Hausdorff distance (HD), fraction of surface distances ≤3 mm (d3 mm)]. Using IPR with optimized parameters, CNR improved by 88% (P < 0.0001) and increased segmentation accuracy was proven irrespective of ISO. Significant improvement was achieved in RMSD [mean at optimal ISO: −28.3% (95% confidence interval (CI) −21.7 to −35.0, P < 0.0001)], HD [−21.4% (95% CI −18.6 to −24.1, P < 0.0001)], and d3 mm [+7.8% (95% CI +4.6 to +10.9, P < 0.0001)]. An average d3 mm of 95.6 ± 2.8% was reached at optimal ISO. Time to generate a 4D model was ±11.5 min with IPR vs. ±22 min without.

Conclusion

Interphase registration significantly improves 4DRA image quality and facilitates semi-automatic segmentation, resulting in clinically useful accuracy despite low-dose image acquisition protocols, while shortening 4D model generation time. This opens the prospect of 4D imaging in clinical settings.

Related Topics

    loading  Loading Related Articles