Eigenfish



We propose to take high-resolution 3-d tomographic images of 60 fish a day over a 60 day period in their early development cycle (360 images in all).  Using modern image compression technology, we will then decompose these images into representations based on fundamental shapes in the images with the goal of tracking changes in these shapes as the fish develop.  We expect to be able to model fish development as a process of a small number of base shapes which are allowed to morph using a dozen or so shape-altering pathways (such as expanding radially, developing a cavity, elongating and bending).  By characterizing these pathways, we expect to confirm biological intuition that all of morphology arises through a small number of ways in which genes express themselves during animal development.   Characterization will be through base shape, morphological pathway and a “velocity vector” which describes how fast and in what direction the shape moves through the pathway.  This vector might change and new base shapes might be added to the organism as development progresses.



This simple experiment could revolutionize developmental biology. It would give geneticists and biologists a defined set of parameters which could be used to describe any developing organism.  It would be much simpler to tie these few independent parameters to the genome than it is to attempt to explain fully formed limbs and organs in complicated organisms.  Furthermore, this project would establish a workflow for measuring these parameters which would be straightforward and relatively low cost and which could then be applied to any animal.  Being able to characterize early development in a manner that is independent of any particular animal would make it easier to compare development across species and individuals and would point the way to being able to make such characterizations while the animals are still alive, leading the way to understanding and being able to monitor early development of human beings and healthy organs as well as anomalous tissues such as cancer tumors.



It has been suggested that the shape of animals does indeed develop along a handful of pathways, though data has never been collected nor a methodology put in place to formalize this hypothesis.  Although a great deal of computer science and artificial intelligence has been applied to problems in genetic analysis, particularly the recognition of important patterns in DNA, no attention has been paid to defining the shape of animals as they grow.   Despite the thousands of papers which have been published on object recognition and tracking for such applications as traffic monitoring, video surveillance and high-end gaming, biological development through shape changes has not been attempted.  Since object tracking is such a well-developed field, however, infrastructure exists to make the computer science aspect of this proposal a relatively simple application of off-the-shelf algorithms.