Biochemistry Publications
Towards large scale automated interpretation of cytogenetic biodosimetry data
Document Type
Conference Proceeding
Publication Date
12-1-2012
Journal
ICIAFS 2012 - Proceedings: 2012 IEEE 6th International Conference on Information and Automation for Sustainability
First Page
30
Last Page
35
URL with Digital Object Identifier
10.1109/ICIAFS.2012.6420039
Abstract
Cytogenetic biodosimetry is the definitive test for assessing exposure to ionizing radiation. It involves manual assessment of the frequency of dicentric chromosomes (DCs) on a microscope slide, which potentially contains hundreds of metaphase cells. We developed an algorithm that can automatically and accurately locate centromeres in DAPI-stained metaphase chromosomes and that will detect DCs. In this algorithm, a set of 200-250 metaphase cell images are ranked and sorted. The 50 top-ranked images are used in the triage DC assay (DCA). To meet the requirement of DCA in a mass casualty event, we are accelerating our algorithm through parallelization. In this paper, we present our finding in accelerating our ranking and segmentation algorithms. Using data parallelization on a desktop system, the ranking module was up to 4-fold faster than the serial version and the Gradient Vector Flow module (GVF) used in our segmentation algorithm was up to 8-fold faster. Large scale data parallelization of the ranking module processed 18,694 samples in 11.40 hr. Task parallelization of Image ranking with parallelized labeling on a desktop computer reduced processing time by 20% of a serial process, and GVF module recoded with parallelized matrix inversion reduced time by 70%. Overall, we estimate that the automated DCA will require around 1 min per sample on a 64-core computing system. Our long-term goal is to implement these algorithms on a high performance computer cluster to assess radiation exposures for thousands of individuals in a few hours. © 2012 IEEE.