Resumen de Left atrial parameterisation and multi-modal data analysis: Application to atrial fibrillation
Many aspects related to the pathogenesis and progression of atrial fibrillation (AF or AFib), the most common type of arrhythmia, are still not fully understood. Besides, current AF treatments, such as ablation strategies, have shown sub-optimal success rates showing the need of developing novel research approaches and computational methodologies. Non-invasive imaging techniques, such as late-gadolinium magnetic resonance imaging (LGE-MRI), have recently emerged as powerful tools to identify structural changes of atrial tissue allowing patient-specific strategies, and extending ablation approaches from generic techniques (e.g. pulmonary vein isolation (PVI)) to substrate-based approaches. Atrial fibrosis imaging using LGE-MRI can be used to improve the understanding of the underlying mechanisms of atrial structural remodeling, contributing significantly to understand the pathophysiology and progression of AF. In order of application, the first contribution of this thesis is an automatic framework to segment the left atrium from LGE-MRI data. The method is based on a whole-heart multi-atlas segmentation scheme that benefits from a shape-based atlas selection strategy to identify the most convenient set of atlases, prior to registration, avoiding the high computational cost of the latter. Second, a framework for obtaining a standardised two-dimensional representation of the left atrial cavity that simplifies data inspection and allows comparison of different atria. Unfolding methods, such as conformal flattening, applicable to the cardiac ventricles are not suitable for the atria due to the complex nature of the constraints (i.e. fixed position of the pulmonary veins and left atrial appendage holes). With our method, auricular cavities of any shape can be mapped to a 2D standardised representation in a few seconds. Third, a methodology to detect and measure the incompleteness of ablation lesions after PVI. PVI stops the fibrillation by ablating the surroundings of the PV avoiding the transmission of the abnormal electrical signals that cause AF and are frequently originated in the veins. The procedure is known to fail in many cases, possibly due to the appearance of gaps in the ablation lesion leading to re-connection. Currently, there is no standard and observer independent way of assessing the magnitude of the gap. Using graph theory, we have provided a unique definition of the gap and a way of quantifying its magnitude, which is highly reproducible and may help in predicting the outcome of PVI. Finally, we have applied the previously mentioned tools and other computational techniques to real clinical datasets showing how our methods can be transferred and included into clinical research practice. We have investigated the regional distribution of ablation gaps in a cohort of AF patients who undergone radiofrequency PVI, the reproducibility of scar imaging with LGE-MRI, the intra- and inter-observer variability of manual LA segmentation tools, the preferential regional distribution of fibrosis in AF patients, and the relation between electroanatomical information (e.g. voltage) and LGE-MRI data.
