Clinical decision-support for diagnosing stress-related disorders by applying psychophysiological medical knowledge to an instance-based learning system

  • Authors:

  • Markus Nilsson;Peter Funk;Erik M. G. Olsson;Bo von Schéele;Ning Xiong

  • Affiliations:

  • Department of Computer Science and Electronics, Mälardalen University, Högskoleplan 1, P.O. Box 832, 721 22 Västerås, Sweden;Department of Computer Science and Electronics, Mälardalen University, Högskoleplan 1, P.O. Box 832, 721 22 Västerås, Sweden;Department of Psychology, Uppsala University, P.O. Box 1225, 751 42 Uppsala, Sweden;The Institute for Psychophysiological Behavioral Medicine, Box 230, 171 77 Stockholm, Sweden and PBM StressMedicine Systems AB, Heden 110, 821 31 Bollnäs, Sweden;Department of Computer Science and Electronics, Mälardalen University, Högskoleplan 1, P.O. Box 832, 721 22 Västerås, Sweden

  • Venue:
  • Artificial Intelligence in Medicine
  • Year:
  • 2006

Quantified Score

Hi-index 0.00

Visualization

Abstract

Objective: An important procedure in diagnosing stress-related disorders caused by dysfunction in the interaction of the heart with breathing, i.e., respiratory sinus arrhythmia (RSA), is to analyse the breathing first and then the heart rate. Analysing these measurements is a time-consuming task for the diagnosing clinician. A decision-support system in this area would reduce the analysis task of the clinician and enable him/her to give more attention to the patient. We have created a decision-support system which contains a signal classifier and a pattern identifier. The system performs an analysis of the physiological time series concerned which would otherwise be performed manually by the clinician. Methods: The signal-classifier, HR3Modul, classifies heart-rate patterns by analysing both cardio- and pulmonary signals, i.e., physiological time series. HR3Modul uses case-based reasoning (CBR), using a wavelet-based method for retrieving features from the signals. The system searches for familiar shapes in the signals by comparing them with shapes already stored. We have applied a best fit scheme for handling signals of different lengths, as the length of a breath is highly dynamic. We also apply automatic weighting to the features to obtain a more autonomous system. The classified heart signals indicate if a patient may be suffering from a stress-related disorder and the nature of the disorder. These classified signals are thereafter sent to the second subsystem, the pattern-identifier. The pattern-identifier analyses the classified signals and searches for familiar patterns by identifying sequences in the classified signals. The identified sequences give clinicians a more complete analysis of the measurements, providing them with a better basis for diagnosis. Results and conclusion: We have shown that a case-based classifier with a wavelet feature extractor and automatic weighting is a viable option for building a decision-support system for the psychophysiological domain, as it is at par, or even outperforms other retrieval techniques and is less complex.