Nephrology Dialysis Transplantation, Vol 13, Issue 1 59-66, Copyright © 1998 by Oxford University Press
W Van Biesen, G Sieben, N Lameire and R Vanholder
Background: A Kohonen topological map is an artificial
intelligence system of the connectionist school (neural networks). The map
learns the typical features of the subclasses in the learning set by means
of a shortest Euclidean distance algorithm, after which self adaptation of
the neurons occurs. By its ability of self-organization and generalization,
a Kohonen map is useful for pattern recognition, and its application in the
medical field as an aid for decision making seems promising. This study
describes the use of a Kohonen topological mapping system in the
classification of renal diseases as being glomerular or tubular on basis of
clinical characteristics and laboratory results.
Methods: Forty-one parameters from 75 patients were
retrospectively retrieved and used to train four different Kohonen maps of
10 x 10 neurons. For reference diagnostic classification, we referred to
the results of the light-microscopic examination. The classification of the
patient by the four different Kohonen networks was compared to the
classification by a rule-based system and by three nephrologists. We also
developed a 'hybrid' decision system that makes a classification on basis
of the opinion of the four networks and that of the rule-based system.
Results: The results show that a Kohonen map is
capable of classifying the patients as having glomerular or tubular disease
with a higher sensitivity and predictive value than the nephrologists and
the rule-based system, and that the best classification was performed by
the hybrid system: sensitivity and predictive value for the diagnosis
'glomerular' respectively 100 and 88% for the network with the most
adequate results, 90 and 83% for the nephrologists, 90 and 95% for the
rule-based system, and 95 and 96% for the hybrid system; sensitivity and
predictive value for the diagnosis 'tubular' respectively 50 and 100% for
the neural networks, 31 and 45% for the nephrologists, 81 and 68% for the
rule-based system, and 87 and 82% for the hybrid system).
Conclusion: We conclude that a Kohonen map is capable
of classifying the patients as having glomerular or tubular disease with a
high sensitivity and predictive value. The rule-based system performs worse
than the neural networks. The most adequate results were obtained with the
hybrid system. Key words: artificial intelligence;
neural networks; medical decision making; kidney; nephrology
ORIGINAL ARTICLES
Application of Kohonen neural networks for the non-morphological distinction between glomerular and tubular renal disease
Departments of Nephrology and Internal Medicine, Renal Division, University Hospital Gent, De Pintelaan 185, B-9000 Gent, Belgium; Department of Control Engineering and Automation, University of Gent, Belgium
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  L. Gabutti, D. Vadilonga, G. Mombelli, M. Burnier, and C. Marone Artificial neural networks improve the prediction of Kt/V, follow-up dietary protein intake and hypotension risk in haemodialysis patients Nephrol. Dial. Transplant., May 1, 2004; 19(5): 1204 - 1211. [Abstract] [Full Text] [PDF]
|
|