Space-time segmentation using level set active contours applied to myocardial gated SPECT

This paper presents a new region based (opposed to classically pixel based) space-time segmentation method applied to gated SPECT sequences. We make the assumption that such sequences are composed of a moving (or warping) object with a constant (in first approximation) pixel value (the myocardium) on a motionless generally non-constant background. Then, the model was modified to take into account the variability in time of the myocardium value. In accordance with this definition, we define a variational criterion which involves all the frames at once providing a dependence between individual frame segmentations. The region based characteristic is implied by the explicit involvement in this criterion of the spatial domains of moving and warping objects in each frame. Then, we propose an original minimization algorithm to compute the desired segmentation of the endocardial and epicardial limits. It is based on alternate resolutions of several coupled partial derivative equations, one of which is implemented with active contours using the level set technique (thus the computed segmentation contours are necessarily closed curves (2D) or surfaces (3D)). We have tested this space-time segmentation method on two 3D clinical gated SPECT sequences