Pierre Ravaux

Towards symbolization using data-driven extraction of local trends for ICU monitoring

We propose a methodology for the extraction of local trends from a stream of data. It has been designed to suit the needs of interpretation-oriented visualization and symbolization from ICU monitoring data. After giving implementation details for efficient computation of local trends, we propose the use of a characteristic analysis span for each variable. This characteristic span is obtained from a set of criteria that we compare and evaluate in regard of analysis of ICU monitoring data gathered within the Aiddaig project. The processing results in a rich visual representation and a framework for the local symbolization of the data stream based on its dynamics.

Computer-assisted evaluation of respiratory data in ventilated critically ill patients

In intensive care unit, a lot of data are currently available but remain unused by nurses and residents because of complexity of analysis. We have developed a system for interpretation of respiratory data (RESPAID) in order to improve monitoring of patients under respiratory support and also to provide a high level of information. RESPAID is a real-time system which interprets quantitative and qualitative aspects of the usual respiratory data at different levels of information. Initial knowledge base was built from data given by four specialists in intensive care. Major attention was paid to different aspects of the system: monitor interface, user interface and time representation. Data are issued from standard respirators and/or monitors used in the intensive care unit. Informations provided by RESPAID are alarm identification, ventilator settings modification and proposal for physiological evolution of the patient or suspected complication.RESPAID runs on IBM PCAT3 with 1st class shell. It is currently in clinical validation procedure.

INFORM: integrated support for decisions and activities in intensive care

SummaryMany medical decision support systems that have been developed in the past have failed to enter routine clinical practice. Often this is because the developers have failed to analyse in sufficient detail the precise user requirements, because they have produced a system which takes too narrow a view of the patient, or because the decision support facilities have not been sufficiently well integrated into the routine clinical data handling activities. In this paper we discuss how the AIM-INFORM project is setting out to deal with these issues, in the context of the provision of decision support in the intensive care unit.

RESPAID: computer aided decision support for respiratory data in ICU

A prototype developed to support interpretation of respiratory data for ventilated patients is described. Data are taken from standard respirators and/or monitors used in intensive care units (ICUs). An initial knowledge base was built from data given by four specialists in intensive care. Both quantitative and qualitative aspects were taken into account. Major attention was paid to different aspects of the system: monitor interface to produce valid data, user interface and time representation. The prototype, called RESPAID, runs in real-time on an IBM PC AT3 with first class shell. This shell uses the concept of learning from examples. RESPAID is currently undergoing clinical validation.<<ETX>>

Introduction of knowledge bases in patient's data management system: role of the user interface*

As the number of signals and data to be handled grows in intensive care unit, it is necessary to design more powerful computing systems that integrate and summarize all this information. The manual input of data as e.g. clinical signs and drug prescription and the synthetic representation of these data requires an ever more sophisticated user interface. The introduction of knowledge bases in the data management allows to conceive contextual interfaces.The objective of this paper is to show the importance of the design of the user interface, in the daily use of clinical information system. Then we describe a methodology that uses the man-machine interaction to capture the clinician knowledge during the clinical practice. The different steps are the audit of the users actions, the elaboration of statistic models allowing the definition of new knowledge, and the validation that is performed before complete integration. A part of this knowledge can be used to improve the user interface. Finally, we describe the implementation of these concepts on a UNIX platform using OSF/MOTIF graphical interface.

A neural approach to knowledge base systems

An implementation for Knowledge Base systems using an Object Oriented data structure and a Neural type inference motor. Advantages are : Knowledge structuration, speed, compatibility with parallel computing, dynamic evolution of the knowledge base.

ICU Patient State Characterization Using Machine Learning in a Time Series Framework

We present a methodology for the study of real-world time-series data using supervised machine learning techniques. It is based on the windowed construction of dynamic explanatory models, whose evolution over time points to state changes. It has been developed to suit the needs of data monitoring in adult Intensive Care Unit, where data are highly heterogeneous. Changes in the built model are considered to reflect the underlying system state transitions, whether of intrinsic or exogenous origin. We apply this methodology after making choices based on field knowledge and ex-post corroborated assumptions. The results appear promising, although an extensive validation should be performed.

Si-GARCH: Construction and validation of a new method for the detection of breaking points in models

ABSTRACT In this article, we define a new method (Si-GARCH) for signal segmentation based on a class of models coming from econometrics. We make use of these models not to perform prediction but to characterize portions of signals. This enables us to compare these portions in order to determine if there is a change in the signal’s dynamics and to define breaking points with an aim of segmenting it according to its dynamics. We, then, expand these models by defining a new coefficient to improve their accuracy. The Si-GARCH method was tested on several thousands of hours of biomedical signals coming from intensive care units.