Paper Info
Title
Interpretation of Hidden Node Methodology in Automated Classification of Neural Cell Morphology
Abstract
Developmental biologists are interested in classifying the development of cells in culture and oligodendrocytes are a class of cell that is frequently studied. In studies, biologists view culture dishes under a microscope and attempt to count the cells using a small number of classes, for example, progenitors, immature type 1, immature type 2 and differentiated. In this way they can determine the effects of pollutants (or other reagents) on growth. This is, however, a difficult, inaccurate and subjective method that could be greatly improved by using computer vision. Previous empirical results in a computer vision application to automate the classification of neural cells showed that a modified network with two hidden nodes added where conditional independencies were seriously violated achieved a significantly improved performance with an average prediction accuracy of 84% compared to 59% achieved by the original network. In this paper we justify the improvement of performance by examining the changes in network accuracy using four network accuracy measurements; the Euclidean accuracy, the Cosine accuracy, the Jensen-Shannon accuracy and the MDL score. Our experimental results consistently demonstrate that the introduction of hidden nodes results in the improvement of network accuracy, and thus contributes to the improvement of prediction accuracy.
Year
Venue
Keywords
2003
METMBS'03: PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON MATHEMATICS AND ENGINEERING TECHNIQUES IN MEDICINE AND BIOLOGICAL SCIENCES
cell morphology
Field
DocType
Citations 
Cell morphology,Pattern recognition,Computer science,Artificial intelligence,Hidden node problem
Conference
0
PageRank 
References 
Authors
0.34
1
4
Name
Order
Citations
PageRank
Jung-wook Bang100.68
Alexandros Pappas231.25
Duncan Fyfe Gillies39717.86
Stephen Muggleton43915619.54