RSNA 2014 

Abstract Archives of the RSNA, 2014


INE026-b

Scalable Distributed Open Source Medical Image Storage, Compute, and Search Platform

Education Exhibits

Presented on December 3, 2014
Presented as part of INS-WEA: Informatics Wednesday Poster Discussions

Participants

David William Piraino MD, Presenter: Medical Advisory Board, Agfa-Gevaert Group
Daniel W. Palmer PhD, Abstract Co-Author: Nothing to Disclose
Naveen Subhas MD, Abstract Co-Author: Research Grant, Siemens AG
Nancy A. Obuchowski PhD, Abstract Co-Author: Research Consultant, Siemens AG Research Consultant, Hologic, Inc Research Consultant, CVUS Research Consultant, Elucid Bioimaging Inc
Daniel Felipe Gonzalez, Abstract Co-Author: Nothing to Disclose
Michael Ciancibello, Abstract Co-Author: Nothing to Disclose
Katie Hulme, Abstract Co-Author: Nothing to Disclose

BACKGROUND

Picture Archiving and Communication Systems (PACS) and Radiology Information Systems (RIS) have been optimized for consistency and transactional performance. New concepts for processing and search of “Big Data” have been developed that favor availability and scalability over consistency. These “Big Data” technologies can be used to process, store, and search medical imaging information. This poster presents an architecture that uses an open source non-SQL scalable distributed database (Riak) as its base storage, search, and processing architecture. We demonstrate how this architecture can be used for search, image processing, and analysis. Riak is a key value database that uses consistent hashing (figure) to equally distribute information and processing to many nodes. Information is located on more than one node for redundancy, fault tolerance, and distributed processing. This distributed database provides map reduce, link walking, and secondary indexes for querying and processing. The open source search engine Solr is included for free text and discrete value search. Open source applications (Mirth, DMC4CHE, and MySQL) are included on each node to provide distributed DICOM and HL7 connectivity and relational database functionality. ImageJ and R are also included on each node to provide image processing functionality and statistics processing. The physical nodes run behind a load balancer to distribute external processing requests.

EVALUATION

The system is presently receiving de-identified radiology reports and DICOM images. DICOM images are stored in DICOM, JPEG, and PNG formats. DICOM header elements are stored as simplified JSON documents. Search over greater than 7 million reports takes 1-2 seconds from a Web frontend. During failure of any node the system continues to receive, process, and provide search results.

DISCUSSION

The system is able to store, process and search radiology reports and DICOM images in a distributed manner.

CONCLUSION

“Big Data” and non-SQL information processing technologies can be used for distributed medical image and report storage, processing, and search.

FIGURE (OPTIONAL)

http://abstract.rsna.org/uploads/2014/14008143/14008143_m9re.jpg

Cite This Abstract

Piraino, D, Palmer, D, Subhas, N, Obuchowski, N, Gonzalez, D, Ciancibello, M, Hulme, K, Scalable Distributed Open Source Medical Image Storage, Compute, and Search Platform.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14008143.html