Machine visual screening

Arman Garakani
Chief Technology Officer
Reify Corporation


Biologists have always relied on viewing biological systems as they evolve in space and time. Most of us -- even non biologists -- can recognize a dynamic biological event such as a beating heart, a dividing, migrating or spreading cell. How is it that we can "label" these events without imposing a model? One explanation may be that the entire event projects information about its dynamic nature. Oscillating events may be recognized by their repetitive presentation. Similarly, spreading events may be recognized by their continuously enlarging presentation. A visual event is captured digitally into a snapshot sequence. Measuring similarity between any snapshot and all other snapshots is particularly hard. The transformation between two snapshots has to describe geometric changes of deformable objects, illumination changes, intensity changes, and so on. If there were no mutual information in a visual event, it would be impossible to measure similarity among its snapshots. In our experience, most biological events, when captured appropriately, are consistent and contain measurable mutual information.

During this talk, I will describe our work at Reify corporation. We have developed a set of technologies to extract and measure biological events enabling "machine visual screening" of biology. We measure the mutual information inherent in an image sequence digital capture of a biological event. We call this measurement the Aggregate Change Index (ACI). ACI can in many cases be directly used as a functional readout. In many other cases it can form a robust basis of other functional readouts. Whether measuring mechanical work of motor proteins or heart function of a young zebra fish for a phenotypic compound profiling, ACI and other mutual information based measurements present the possibility of unbiased, robust, and repeatable functional readouts of the marvellously consistent set of systems that is Biology.


Arman Garakani has over 18 years of experience in all aspects of computational methods applied to measurement of physical systems. Arman was Director of Vision Tools and Applications at Cognex Corp., the world's leading machine vision company, and has 13 image analysis patents in his name and others pending. Arman was a primary contributor to Cognex's creation of more than $1.0 billion in shareholder value. Since leaving Cognex, he has worked in multiple video and image processing startups. Arman is the inventor of Reify's patent-pending Visible Discovery platform, and is responsible for Reify's technology development. Arman's graduate research was in computational methods in Fluid Mechanics at Alden Laboratory; one of the oldest continuously operating hydraulic laboratories in the world. Prior to joining Cognex, Arman worked at IBM Poughkeepsie on the first generation automated equipment for semiconductor manufacturing. Arman holds a BS in Mechanical Engineering from Worcester Polytechnic Institute and an MS in Management of Technology from MIT.