Artificial neural networks for non-invasive chromosomal abnormality screening of fetuses

Abstract

A large number of different neural network structures have been constructed, trained and tested to a large data base of pregnant women characteristics, aiming at generating a classifier-predictor for the presence of chromosomal abnormalities in fetuses, namely the Trisomy 21 (Down syndrome), Trisomy 18 (Edwards syndrome), Trisomy 13 (Patau syndrome) and the Turner syndrome. The database was composed of 31611 cases of pregnant women. 31135 women did not show any chromosomal abnormalities, while the remaining 476 were confirmed as having a chromosomal anomaly of T21, T18, T13, or Turner Syndrome. From the total of 31611 cases, 8191 were kept as a totally unknown database that was only used for the verification of the predictability of the network. In this set, 7 were of the Turner syndrome, 14 of the Patau syndrome, 42 of the Edwards syndrome and 71 of the Down syndrome. For each subject, 10 parameters were considered to be the most influential at characterizing the risk of occurrence of these types of chromosomal anomalies. The best results were obtained when using a multi-layer neural structure having an input, an output and three hidden layers. For the case of the totally unknown verification set of the 8191 cases, 98.1% were correctly identified. The percentage of abnormal cases correctly predicted was 85.1%. The unknown T21 cases were predicted by 78.9%, the T18 by 76.2%, the T13 by 0.0% and the Turner syndrome by 42.9%.