|
Marie-Odile Cordier and Guy Carrault and René Quiniou and Feng Wang
Intelligent data analysis for diagnosing and monitoring
, IDAMAP 2002: Intelligent Data Analysis in Medicine and Pharmacology - workshop at ECAI'2002
, 19-23
, july
, 2002
, Document
|
|
Abstract
We present a novel approach to cardiac data analysis for monitoring patients in intensive care units. Recorded data are transfomed into symbolic event streams. Then a chronicle recognizer attempts to detect, on the fly, interesting temporal patterns called chronicles on this stream. Chronicles including temporal relationships are high level representations of temporal phenomena. They can be learnt from example data by relational learning methods such as Inductive Logic Programming. On the one hand, they are expected to be more accurate and robust for diagnosis, even in the presence of a noisy input and, on the other hand, are easier to understand by the clinical staff. The feasibility of the approach is illustrated on learning and recognizing cardiac arrhythmias from electrocardiograms. We hope to use this same approach for characterizing less known temporal phenomena or for interpreting input from new sensors.
|
|