Biomimetic Ultrasound: A Computational Approach

Rolf M¨uller, The Mærsk Institute, University of Southern Denmark

Biomimetic engineering has a proven track record of turning inspirations from biology into products. Here, the potential of biosonar to advance technology will be considered. Biosonar is one of the most integrated and capable sensorimotor systems in nature. It enables bats to thrive in a multitude of ecological niches and demonstrates how a wealth of information about a communication channel can be obtained parsimoniously from two sensory inputs only. Insights into the underlying mechanisms can be used to build compact powerful sensing and communication devices for application areas such as biomedical diagnosis, household appliances, wireless communication, non-destructive testing, and autonomous robotics. In biosonar systems,built-in intelligence and task-specific knowledge are distributed across all stages and integration levels. An understanding of these solutions requires a comprehensive analysis of the sensory tasks and system properties. Furt! hermore, physical data from real-world sensing scenarios should be considered in order to ensure the validity and robustness of the results. All these requirements can be met by combining computational analysis with hardware realizations of biomimetic systems. It will be shown how the various challenges posed by this endeavor can be addressed. A computational approach to the extraction of smart antenna knowledge from biological samples will receive particular attention: a semi-automated tool-chain has been devised for generating digital representations of natural antenna shapes and predicting their acoustic properties numerically. In addition, digital shape manipulation methods open up new opportunities to study the relationship between morphology and function. This will be illustrated by the tragus, a frontal protrusion of the outer ear, which was found to control the formation of a strong asymmetric side-lobe in the directivity pattern. To highlight the potential of neurom! imetic signal processing, a computational theory for classification of natural landmarks will be presented. A short feature vector which follows naturally from the statistical properties of a parsimonious neuromimetic echo encoding was found to support an unprecedented classification performance in this demanding estimation task. The task and its solution both address issues relevant to diagnostic biomedical ultrasound. Future work will use the functional analysis of biological smart antenna and sensory array shapes to derive quantitative rules for the systematic design of biomimetic technology. It will also investigate the integration of acoustic and neuromimetic signal processing in order to understand how sensory systems can be jointly optimized in both domains.

Biography

Rolf M¨uller is an assistant professor at the Mærsk Mc-Kinney Møller Institute for Production Technology at the University of Southern Denmark. He earned his master (Neuroscience, Genetics, and Electronics, in 1995) and Ph.D. degree (summa cumlaude and dissertation award, in 1998) from the University of T¨ubingen, Germany. He has been a postdoctoral fellow in the Department of Electrical Engineering at Yale University (1998-2000) and project leader at the University of T¨ubingen (2000- 2003). In T¨ubingen, he established the Biosonar Lab and participated in extending Germany’s first accredited bioinformatics program. Over the last five years, he has authored 15 peer-reviewed works and 17 conference contributions. He has been awarded scholarships by the German National Merit Foundation, the German ScienceFoundation, and NATO. His research has been funded by grants from the State of Baden-W¨urtt! emberg, the German Science Foundation, and the European Union.