Statistical modeling and segmentation in cardiac MRI using a grid computing approach

Grid technology is widely emerging as a solution for wide-spread applicability of computerized analysis and processing procedures in biomedical sciences. In this paper we show how a cardiac image analysis task can substantially benefit from Grids, making use of a middleware service tailored to the needs of common application tasks. In a first part we describe a methodology for the construction of three-dimensional (3D) statistical shape models of the heart, from a large image database of dynamic MRI studies. Non-rigid registration is needed for the automatic establishing of landmark correspondences across populations of healthy and diseased hearts; but when dealing with large databases, the computational load of current algorithms becomes a serious burden. Our Grid service API provided an easy way of taking benefit from our computing resources, by allowing for pipelining the distributed and non-distributed steps of the algorithm. As a second part of this work we show how the constructed shape models can be used for segmenting the left ventricle in MRI studies. To this aim we have performed an exhaustive tuning of the parameters of a 3D model-based segmentation scheme, also in a distributed way. We run a series of segmentation tests in a Monte Carlo fashion, but only making use of the Grid service web portal, as this time the pipeline was simpler. Qualitative and quantitative validation of the fitting results indicates that the segmentation performance was greatly improved with the tuning, combining robustness with clinically acceptable accuracy.