Ventricular structure in ARVC: going beyond volumes as a measure of risk

Background Altered right ventricular structure is an important feature of Arrhythmogenic Right Ventricular Cardiomyopathy (ARVC), but is challenging to quantify objectively. The aim of this study was to go beyond ventricular volumes and diameters and to explore if the shape of the right and left ventricles could be assessed and related to clinical measures. We used quantifiable computational methods to automatically identify and analyse malformations in ARVC patients from Cardiovascular Magnetic Resonance (CMR) images. Furthermore, we investigated how automatically extracted structural features were related to arrhythmic events. Methods A retrospective cross-sectional feasibility study was performed on CMR short axis cine images of 27 ARVC patients and 21 ageing asymptomatic control subjects. All images were segmented at the end-diastolic (ED) and end-systolic (ES) phases of the cardiac cycle to create three-dimensional (3D) bi-ventricle shape models for each subject. The most common components to single- and bi-ventricular shape in the ARVC population were identified and compared to those obtained from the control group. The correlations were calculated between identified ARVC shapes and parameters from the 2010 Task Force Criteria, in addition to clinical outcomes such as ventricular arrhythmias. Results Bi-ventricle shape for the ARVC population showed, as ordered by prevalence with the percent of total variance in the population explained by each shape: global dilation/shrinking of both ventricles (44 %), elongation/shortening at the right ventricle (RV) outflow tract (15 %), tilting at the septum (10 %), shortening/lengthening of both ventricles (7 %), and bulging/shortening at both the RV inflow and outflow (5 %). Bi-ventricle shapes were significantly correlated to several clinical diagnostic parameters and outcomes, including (but not limited to) correlations between global dilation and electrocardiography (ECG) major criteria (p = 0.002), and base-to-apex lengthening and history of arrhythmias (p = 0.003). Classification of ARVC vs. control using shape modes yielded high sensitivity (96 %) and moderate specificity (81 %). Conclusion We presented for the first time an automatic method for quantifying and analysing ventricular shapes in ARVC patients from CMR images. Specific ventricular shape features were highly correlated with diagnostic indices in ARVC patients and yielded high classification sensitivity. Ventricular shape analysis may be a novel approach to classify ARVC disease, and may be used in diagnosis and in risk stratification for ventricular arrhythmias. Electronic supplementary material The online version of this article (doi:10.1186/s12968-016-0291-9) contains supplementary material, which is available to authorized users.


Spatial and temporal alignment of anatomical models
In order to quantitatively compare the anatomical models of different subjects, the models were first aligned in space and the correct temporal frames were extracted. Spatial alignment reduced the bias in the construction of the mean (described in the next section) to allow differences in the anatomy to be computed rather than differences in position and orientation. Temporal alignment ensured that corresponding time frames were compared. The spatial alignment was performed on the extracted point clouds, rather than on the images themselves, given that points are simpler to align than images.
The spatial alignment used in this work was rigid; essentially aligning all subjects to the same physical space without any stretching/shortening or scaling. In order to align the different subjects, the barycentre of each ventricle over all slices was computed and a line-segment joining the barycentre of the left ventricle (LV) to the barycentre of the right ventricle (RV) was computed. The line-segments were then aligned pair-wise to a chosen reference subject for all other subjects via rotation and translation operations. The computed line-segment transformation was then applied to all points. Note that the choice of reference subject was arbitrary in this formulation.
The temporal alignment used in this work was designed to account for the fact that CMR sequences are gated by the electrocardiogram (ECG) signal, while taking into consideration the different heart rates and ES phase from one subject to another. Assuming the ECG-gating was accurate, the first and last frames of the CMR sequence should correspond to the ED phase. Thus, we chose the first frame as the ED phase. The ES frame was taken from the LV volume curves computed from Segment for each subject by choosing the frame that gave the minimum volume.
Once all sets of point clouds were aligned to a common spatial frame and corresponding temporal frames were extracted, an anatomical surface model was created from each point cloud using the Gmsh software [3], and smoothed to obtain a more physiological shape using the Visualisation Toolkit (VTK) [4].

Mean model construction
In the present work, we were interested in computing a mean shape model without a parameterisation of the surfaces (namely without defining any landmarks, and point-correspondences from one anatomical model to another). A non-parametric (no point labelling) method was used to describe each anatomical model as a set of 'currents', from which statistical analysis was performed on the set of currents rather than on the anatomical models themselves. The currents representation of shapes essentially describes a distribution of shape features by characterising the surfaces by how they 'interact' with vector fields; which is a mathematical concept that can be considered intuitively as the same action that a laser scanner has to describe the shape of a 3D object by moving a laser over the object. Describing the shapes (surfaces) in this way has the advantage of removing the parameterisation of points while providing a framework where two shapes can be quantitatively compared.
A forward model was then used to describe each subject shape by how the mean shape was stretched/shortened to obtain this shape (the deformation required to transform shape A to shape B), plus some residual information not included in the analysis (such as errors/bias due to image quality, segmentation, etc.). Using this formulation, the mean was computed iteratively by minimising the 'distance' between the mean and all subjects (or more precisely, by minimising the deformations from each subject to the mean). The deformations were computed using the large deformation diffeomorphic metric mapping method (LDDMM). The full mean shape model construction pipeline is described in [5], the application of this method for right ventricular shape analysis is described in [6] and the extension of this to bi-ventricular analysis is described in [7]. Figure 1 The pre-processing pipeline to go from a CMR image to a point cloud describing the surfaces of the endocardium and epicardium of the left ventricle and endocardium of the right ventricle (ventricle segmentation), then to correct the slice misalignment due to breath-holds (slice misalignment correction), then to detect the frames corresponding to the ED and ES phases (extract ED / ES phases), and finally to align all subjects spatially to remove differences in position and orientation (spatial alignment).