HomeNews & MediaPress ReleasesPress Release 

April 16, 2008

Media Contacts:

Michael Buckley, The Johns Hopkins University Applied Physics Laboratory
240-228-7536 or 443-778-7536

Kristi Marren, The Johns Hopkins University Applied Physics Laboratory
240-228-6268 or 443-778-6268

The Johns Hopkins University Applied Physics Laboratory
Office of Technology Transfer
Inventions of the Year for 2007

Bayesian Information Fusion Networks (BIFNs) for Electronic Surveillance System for Early Notification of Community-Based Epidemics (ESSENCE)

Need: Timely and accurate detection of both naturally occurring and bioterrorism-related disease outbreaks in a population is crucial for executing an efficient public health response to reduce mortality and morbidity. Current techniques for detecting disease outbreaks suffer from a high level of false-positive alerts. Thus, there is an ever-pressing need to reduce false-alarm rates. The Bayesian Information Fusion Network (BIFN) meets this need.

Technical Description: Electronic Surveillance System for Early Notification of Community-Based Epidemics (ESSENCE), developed and deployed throughout the United States by The Johns Hopkins University Applied Physics Laboratory (JHU/APL), is an automated disease-surveillance system that (i) collects electronic information representing indicators of health in a population [e.g., over-the-counter drug sales and hospital emergency room pre-diagnosis information, such as ICD-9 (International Classification of Disease, Ninth Revision) codes] and (ii) applies techniques to determine epidemiologically significant events (e.g., disease outbreaks) based on the collected information. The BIFN is an important improvement to these techniques in that it significantly reduces false alarms.

Mathematical anomalies derived from information that is indicative of the health of a population are very often irrelevant to public health events, and epidemiologists can rule out obvious false alerts by slicing and looking at the information from a different perspective. Essentially, the BIFN automates the epidemiologist's "rule-out" process. Unlike other techniques, it fuses information from multiple sources to estimate relevance of statistical anomalies that may be epidemiological events.

In a first step, multiple data sources (e.g., emergency room reports and outpatient reports, which are geographically distributed) are searched for possible alert indications. In a second step, the alert indications are synchronized in time. In a third step, the synchronized alert indications are input into a Cognitive Analysis and Information Fusion Network, where data fusion occurs in a Bayesian Belief Network (BBN).

An exemplary BBN is depicted in Figure 1. The BBN fuses multisource information in a manner comparable with an epidemiologist's decision-making process. This has not been done before. The BBN is a probabilistic graphical model typically visualized as a directed graph. Its nodes represent variables, and directed edges represent probabilistic dependencies; it offers a compact representation of the relationships among all variables in the graph. Expert (e.g., epidemiological) knowledge is embedded in the BBN. From the BBN, the probability of occurrence of an epidemiologically significant event is determined.

Stage of Development: Reduced to practice; the BIFN has been implemented and successfully tested in an ESSENCE system operated in the National Capital Region.

bayesian infromation fusion networks

Figure 1. The structure of Cognitive Analysis and Information Fusion Network for Influenza Detection.


The Applied Physics Laboratory, a division of The Johns Hopkins University, meets critical national challenges through the innovative application of science and technology. For information, visit www.jhuapl.edu.