SEI: Hierarchical Dependency Graphs for Col-Space Design with Application to Leukocyte Detection and Tracking

University Of Virginia Main Campus, Charlottesville VA

Investigators

Abstract

There are two barriers to the automatic data collection and analysis from video microscopy. The algorithms grow more sophisticated and the current implementations fall short of real-time analysis, which requires a speedup of three or four orders of magnitude over current implementations. The traditional design methodology is inherently limited in the quality of systems it can produce, restricting the capabilities of both the engineers and the application experts. This projects co-optimizes both algorithms and custom system architectures in a collaborative space, where changes in implementation can be considered at the same time as changes in the algorithms, leading to enhanced performance. The system uses hierarchical dependency graphs as a common language to establish the bi-directional relationship between and algorithm and an implementation. The interdisciplinary team from design, biomedical image analysis, computer architecture and medicine, is developing a prototype system for in vivo leukocyte detection and tracking and establishment of a collaborative space optimization design. This research will facilitate a new approach and design infrastructure for designing systems that will have dramatic improvements in system efficiency and speeds. The work will be publicly released and lead not only to training of students but a new collaborative approach to design of algorithms and implementations.