ABSTRACT
We present an emerging, artificial life framework for medical image analysis. It was originally introduced as an extension to the established physics-based image segmentation approach known as deformable models. Although capable of extracting coherent, smooth boundaries from low-level image information, classical deformable models rely on human guidance to produce satisfactory segmentation results. Our artificial life approach augments the bottom-up, data-driven deformable model methodologies with top-down, intelligent deformation control mechanisms, yielding intelligent deformable models that we call deformable organisms. This is achieved by adding behavioral and cognitive modeling layers atop the physical and geometrical layers of the classical models. The resulting organisms evolve according to behaviors driven by decisions based on their perception of the image data as well as their internally encoded anatomical knowledge. Thus, deformable organisms are autonomous agents that can automatically segment, label, and quantitatively analyze anatomical structures of interest in medical images. This chapter motivates and overviews our novel framework, describes its fundamental principles, demonstrates prototype instances of deformable organisms, and summarizes recent advances.
