Intelligent Adaptive Monitoring for Cardiac Surveillance - Computational Intelligence in Healthcare: Advanced Methodologies - page 330

hand. Using a symbolic approach for knowledge and (temporal) reasoning makes it
possible to provide understandable explanations to the user. This is very important in
medicine for letting clinician operators trust such systems.
In our proposal, the adaption functionality relies on a centralized pilot which has to
take its decisions apriori by analyzing the context. This approach was motivated by
limited computing resources. In case of no limitation on computing resources, an en-
semble of detectors approach could be used [7]. This approach would consist in running
all the detectors concurrently and merge their results. Though requiring more comput-
ing resources, ensemble methods show good performance and should be assessed for
adaptive monitoring. Currently, there is a strong interest in using such methods for
change detection [18,1]. We will investigate ensemble methods for self-adaptation in a
near future.
We have advocated the use of a supervised learning method, ILP, for learning chron-
icles. In many situations, collecting labelled examples is difficult and unsupervised
learning methods are more relevant. We are investigating data mining methods for dis-
covering temporal patterns containing numerical constraints that could be easily trans-
lated to chronicles [14].
Until now Calicot has used a static diagnostic knowledge: decision rules and chron-
icles are learnt offline and are not modified during monitoring. To perform full self-
adaptation Calicot should be able to learn or adapt its diagnostic knowledge online in
order to cope with context change or with important changes in the state of the patient.
In an other application domain, intrusion detection, we are investigating solutions for
dealing with concept change and adaptation [36,15]. These solutions could be tried on
patient monitoring as well.
This research could not have been achieved without an active and fruitful collabora-
tion with the medical staff. Working with clinicians in hospital is not always easy for
computer scientists: experts are overbooked, getting data is sometimes difficult as pro-
tocols for recording data are very strict, especially they should not introduce any risk
for the patient or any violation of data privacy. But, confronting ideas and views from
different research, knowledge and practice domains is particularly rewarding.
References
1. Bach, S.H., Maloof, M.A.: Paired learners for concept drift. In: ICDM, pp. 23-32. IEEE
Computer Society Press, Los Alamitos (2008)
2. Bremond, F., Thonnat, M.: Issues of representing context illustrated by video-surveillance
applications. International Journal of Human-Computer Studies Special Issue on Context 48,
375-391 (1998)
3. Carrault, G., Cordier, M.-O., Quiniou, R., Wang, F.: Temporal abstraction and inductive logic
programming for arrhythmia recognition from ECG. Artificial Intelligence in Medicine 28,
231-263 (2003)
4. Cordier, M.-O., Dousson, C.: Alarm driven monitoring based on chronicles. In: Safeprocess
2000, pp. 286-291 (2000)
5. de Kleer, J., Mackworth, A., Reiter, R.: Characterizing diagnoses and systems. Artificial
Intelligence 56(2-3), 197-222 (1992)
6. Dechter, R., Meiri, I., Pearl, J.: Temporal constraint networks. Artificial Intelligence 49(1-3),
61-95 (1991)