Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9574
Title:	Neural Nets for Detection of Down'S Syndrome
Authors:	Sordo Sanchez,MM
Date:	1995
Presented:
Keywords:
Abstract:	Of great practical importance are all the statistical methods developed so far, that help in the diagnose of Down's syndrome in unborn babies. However, they can not provide a clear "yes/no" response to an input data. Artificial Neural Networks are an alternative approach to this problem, since they are capable to provide, with some extent, this required response. A Radial Basis Function network with 3-50-1 processing elements for the input, hidden and output layer provided a 100orrect classification during learning as well as during training stage with neither false positive nor false negative responses, was selected as the final network. However, when tested with independent data, the network correctly classified 8435777771060f the Down's cases in the high risk group -"YES" response- but with a false positive rate of 35.5 The overlapping distributions of both the unaffected and Down's cases complicated the classification. Comparing the previous figures with the current statistical methods, which provide classification rates around 60-70with false positive rates close to 5-7, it is clear that additional work will be required before a network could produce reliable results that could be acceptable in clinical practice. Even though it is hoped that this work will stimulate further work in the field.
Download:	NO ONLINE COPY

Division of InformaticsForrest Hill & 80 South Bridge