Methods to be used

The project will include a number of related work packages, some running concurrently, and will involve health service researchers working together with technology experts and computer scientists. The architecture of the system will be based (at least loosely) on that of the BabyTalk project and the tasks for the work packages will be as follows:

1) Developing an ontology and knowledge base for remote road traffic accident emergency management

2) Simulation and processing of sensor data

3) Use of NLG and summarisation for decision support

4) Design of user interface and mode of operation

5) Evaluation of summaries

6) Development of follow-up proposal

Work Package 1: Developing an ontology and knowledge base for remote road traffic accident emergency management

This will involve detailed consultations with healthcare workers at all levels. Working with these individuals, we will establish the most common and time critical acute scenarios likely to be managed in traffic accidents in rural settings. For each scenario, we will develop an understanding of the key elements of history, physical signs and physiological measures that are required to manage each scenario. This will support the development of appropriate automated reasoning mechanisms (probably ontology plus rules) for this setting.

Work Package 2: Simulation and processing of sensor data

In anticipation of later application of actual sensor technology, we will create simulated sensor data for different accident scenarios that can be used to drive the development of the summarisation/ decision support system and also to evaluate that system. We will ensure that this data is as realistic as possible by making use of output from medical simulation systems and using expert knowledge (and our previous experience with medical sensors) to adjust the data to simulate abnormal conditions. However, we will assume a stationary casualty and hence minimal movement noise. We will build software to extract medically relevant patterns (as determined by Work Package 1) from this data.

Work Package 3: Use of NLG and summarisation for decision support

Using reasoning mechanisms from Work Package 1, data patterns generated from Work Package 2 and experience from the BabyTalk project, we will develop a method to automatically generate summaries with associated recommendations/ advice for a range of plausible remote road traffic accident situations.  These summaries will require to be configured differently for different levels of training and expertise among healthcare workers, for instance incorporating a different balance between informing and directing according to the past experience of the user.

Work Package 4: Design of user interface and mode of operation

With our collaborators, we will produce a use-case analysis and initial interface design for a portable device implementing the above functionalities, which advises a person on the scene, logs important events and is able to produce handover summaries. We will investigate and make recommendations about how someone would use such a device and how it would inter-operate with established road traffic accident management procedures.

Work Package 5: Evaluation of summaries

Using data and summaries generated in work package 3, we will generate a number of scenarios and present system output to rescue personnel not involved in the earlier work packages to assess the acceptability of the interface and the appropriateness of recommended actions/ information given.

Work Package 6: Development of follow-up proposal

Assuming that our technology is promising, this WP carries out additional background research so as to be able to propose a further project (possibly partly concurrent with this one) which uses real sensor data (possibly with moving casualties), makes use of real mobile devices and evaluates in realistic simulated accident situations. For the sensor work, we will build on our current mountain rescue work with a sensor company in Belfast (Intelesens), with whom we are trialling a wireless sensor patch which measures heart rate, skin temperature and movement.  Our findings to date suggest that the addition of respiratory monitoring is highly desirable and we will continue to explore whether this can be built into one composite sensor patch.  We also need to consider at this stage the networking requirements for the proposed system and how they might be met.