Abstract:
Introduction: Atrial fibrillation (AF) ablation requires a precise reconstruction of the left atrium (LA) and pulmonary veins (PV). Model-based FAM (m-FAM) is a novel module recently developed for the CARTO system which applies machine-learning techniques to LA reconstruction. We aimed to evaluate the feasibility and safety of a m-FAM guided AF ablation as well as the accuracy of LA reconstruction using the cardiac computed tomography angiography (CTA) of the same patient LA as gold standard, in 32 patients referred for AF ablation.
Methods: Consecutive patients undergoing AF ablation. The m-FAM reconstruction was performed with the ablation catheter (Group 1) or a Pentaray catheter (Group 2). The reconstruction accuracy was confirmed prior to the ablation by verification of pre-specified landmarks of the LA and PVs by intracardiac echocardiogram (ICE) visualization and fluoroscopy. A cardiac CTA performed before the ablation was used as gold standard of LA anatomy. For each patient, the m-FAM reconstruction was compared to his/her cardiac CTA.
Results: The m-FAM reconstruction was accurate in all patients regardless the catheter used for mapping. In 12% re acquisition of the LA landmarks was necessary to improve the accuracy. m-FAM time was shorter in group 2 while the M-Fam fluoroscopy time was similar. Pulmonary vein isolation was achieved in 100% of patients without major complications. The m-FAM reconstructions accurately resemble the cardiac CTA of the same patients.
Conclusions: The m-FAM module allows for rapid and precise reconstruction of the LA and PV anatomy, which can be safely used to guide AF ablation.
Introduction :
A precise reconstruction of the left atrium (LA), the pulmonary veins (PV) and the left atrial appendage (LAA) is of critical importance in order to achieve efficient and safe results during atrial fibrillation (AF) ablation. The procedure is currently performed with the guidance of three-dimensional mapping systems. In the CARTO 3 mapping (CARTO System, Biosense Webster, Inc, Diamond Bar, CA) the LA reconstruction methods most frequently used are the Fast Anatomical Mapping (FAM) and cardiac computed tomography angiography (CTA) integration with the electroanatomic mapping (CARTO Merge).1-6
The existing FAM algorithm currently used in CARTO constructs a three-dimensional reconstruction of the atrial anatomy by applying a standard ball-pivoting algorithm to the point cloud acquired by the catheter. FAM mapping requires catheter manipulation by a skilled operator and it is frequently time consuming for less experienced operators. The result does not consider the underlying anatomy, provides no information in areas where the catheter has not yet visited, and contains anatomically incorrect artifacts due to deformation of the atria by the catheter during the mapping procedure, such as merging of the appendage and left superior pulmonary vein. Therefore, the raw anatomy obtained with FAM requires post imaging refinement of the LA surfaces, achieved by shaving the image with “sculpting tools” prior to initiation of the ablation and throughout the procedure. In the CARTO Merge, the CTA scan of the left atrium is integrated with the FAM reconstruction. CARTO Merge provides a far better definition of the LA anatomy, including the areas mentioned above, which are critical for a successful ablation and provides a clear definition of anatomical variants, such as PV common ostium, separated branches, or additional PVs. The drawback of CARTO Merge approach is mainly related to the exposure of the patients to contrast media and (low dose) radiation during the cardiac CTA, plus the additional costs of the CTA scan, which are added to the total cost of the procedure.
Model-based FAM (m-FAM) is a novel module recently developed for the CARTO 3 system (Biosense Webster). 7 8 The scope of the mFAM algorithm is generating an improved, anatomically correct reconstruction of left atrial anatomy, while at the same time requiring fewer samples than are currently needed. This is carried out by defining a parametric model representing a shape of a portion of a heart, and constructing a statistical prior of the shape from a dataset of other instances of the portion. The method is further carried out by fitting the parametric model to the point clouds data and statistical prior to produce an isosurface of the portion of the heart of the subject, which is then display in CARTO. m-FAM applies machine-learning techniques to LA reconstruction using an adaptive model trained from over 300 LA anatomies obtained from CTA scans. Training was performed to define statics of relevant proportions and sizes to represent a “realistic” and anatomically-accurate left atria. Examples of these features are the angle between the PVs, distances from atria center to the valve and PV ostium or the relations of those values to the volume. Having the geometrical primitives associated with any of the left atria anatomical structures, as well as keeping the proportions and sizes trained by a large left atria population, provides a way to represent the most probable left atria even with very limited information. The algorithm selects the model that best fits all the available information: points locations, contact force magnitude and direction, user tagging of anatomical parts, and a statistical score measuring anatomical correctness of the model. The reconstruction adapts smoothly to new points as they are mapped by the catheter, and allows assignment of special points where the catheter is in contact with the LA surface (“magnets”), such as ablation points, that constrain the surface while retaining overall anatomical correctness. The model deforms to best align with those locations and model reconstruction.
Mathematically, the algorithm can be defined as an optimization problem in which model parameters that best fit the data and prior statistical knowledge are estimated in an iterative process. The model is a result of fitting a parametric shape model to the point cloud acquired by catheter positions. The model includes a geometrical primitive for each of the left atria parts, main chamber, pulmonary veins, appendage, and valve. The geometrical primitives are blended to form the atria structure. Once the mapping catheter or the ablation catheter introduced in the LA engage the PVs, their location is tagged on the software and model fitting begins. A few additional force- and respiration-gated points are required to further refine the surface and volume of the LA at key locations, such as the roof, back wall and anterior LA, this tagged points are called magnet points. After acquisition of enough magnet points, the software produces a reconstruction of the LA that includes the LAA ostium and PVs. The model continually self-adjusts with added ablation tags during the procedure. Importantly, the model is trained to adjust for non-classical left atrial and pulmonary veins anatomy as for common ostium or additional side branches. In case of side branches, these branches are incorporated in the PVs of the closest major PV. The potential advantages of the m-FAM include: reduction of the mapping time required to achieve an entire LA reconstruction and better shaping of the LA anatomy compared to FAM.
In view of the fact that the m-FAM is a totally new feature of the CARTO 3 system, we decided to investigate two different workflows to obtain a m-FAM reconstruction of the LA using either the ablation catheter (3.5mm Smart-Touch or 3.5mm ST-SF ablation catheters, Biosense Webster) or a Pentaray catheter (Biosense Webster) to collect the location tags required by the software. We opted for a mapping using either the ablation catheter or a Pentaray catheter because these are the most commonly catheters used to create the initial LA reconstruction. Of note, m-FAM reconstruction can also be obtained with a Lasso NAV catheter (Biosense Webster) if preferred. The accuracy of the m-FAM reconstruction was confirmed by confirmation of the tagged magnet points both with fluoroscopy and intra-cardiac echocardiography (ICE) at the drop-points of the pulmonary veins to precisely define the pulmonary vein antrum and avoid ablations either too distal or too proximal to the pulmonary veins as well as the roof, anterior and posterior LA. Moreover, the m-FAM reconstruction was compared to a cardiac computed tomography angiography (CTA) of the LA which is routinely performed in our center prior to ablation of AF. Importantly, for the purpose of this study, the operator remained blinded to the CTA scan imaging of the patient during the m-FAM mapping reconstruction and reconstruction confirmation with ICE and fluoroscopy.
The aim of the present study was to evaluated the feasibility and safety of the ablation procedure when guided by the m-FAM reconstruction and the time required to achieve the m-FAM reconstruction and fluoroscopy time. We also compared the LA reconstruction obtained with m-FAM obtained with different catheters to the patient’s cardiac CTA.