Fuzzy Segmentation of the Left Ventricle in Cardiac MRI Using Physiological Constraints

We describe a general framework for adapting existing segmentation algorithms, such that the need for optimisation of intrinsic, potentially unintuitive parameters is minimized, focusing instead on applying intuitive physiological constraints. This allows clinicians to easily influence existing tools of their choice towards outcomes with physiological properties that are more relevant to their particular clinical contexts, without having to deal with the optimisation specifics of a particular algorithm’s intrinsic parameters. This is achieved by a structured exploration of the parameter space resulting in a subspace of relevant segmentations, and by subsequent fusion biased towards segmentations that best adhere to the imposed constraints. We demonstrate this technique on an algorithm used by a validated, and freely available cardiac segmentation suite (Segment – http://segment.heiberg.se).