Computational Intelligent Diagnostic System

1
Computational Intelligent Diagnostic System
in predicting preeclampsia

• Computational Intelligence
• Artificial neural networks
• Evolutionary systems / Genetic algorithms
• Artificial immune systems
• Fuzzy systems

Andreas Neocleous Kypros Nicolaides Christos
Schizas Kleanthis Neokleous Natasa Schiza Costas
Neocleous
FMF, University of Cyprus, Cyprus University of
Technology, Cyprus
2
Computational Intelligent System in predicting
preeclampsia
Objective Use computational intelligence to
predict preeclampsia at 11-13 wks
All data Total singleton pregnancies 13,538 No
preeclampsia 13,118 (96.9) Preeclampsia
420 (3.1)
Data for training and validations Unbalanced
data Training various ANNs 335 PE, 10,496
unaffected by PE Totally unknown cases used for
validations 85 PE, 2,622
unaffected by PE Balanced data Training various
ANNs 335 PE, 352 unaffected by PE Totally
unknown cases used for validations
85 PE, 88 unaffected by PE
3
Computational Intelligent System in predicting
preeclampsia
Results on the unknown validation (verification)
data set
4
Computational Intelligent System in predicting
preeclampsia
Conclusions

• Very encouraging findings regarding
  classification by means of Computational
  Intelligence
• Future work should aim to reduce the normal data
  in a systematic way so that the new reduced set
  will reflect the whole database of cases.
• Some unaffected cases were repeatedly wrongly
  categorized as PE in almost all of the examined
  networks. When we compared the inputs of these
  cases with the inputs of some true PE cases, we
  observed that they were quite similar.

Thank you