(+48) 500 333 969, (+48) 502 318 888

Ginekologia i Położnictwo medical project
ISSN 1896-3315 e-ISSN 1898-0759

Use of Advanced Methods of Statistical Modelling in Estimating Fetal Dystrophy


Author(s): Dariusz Daniluk, Andrzej Kisiel, Agata Smoleń, Artur Czekierdowski

Introduction. Completion of pregnancy with surgical procedure is most frequently used in women with dystrophic feta. In carried own studies, the method of searching non-linear correlation of variables was suggested, based on computer-generated artificial neural network (ANN) in order to improve the estimation of fetus weight. Objective of the work. Examining whether introduction of advanced methods of statistical modelling in the estimation of ultrasonographic biometry may be more useful in early diagnostics of fetus dystrophy than currently applied methods.Material and methods. Investigated group was composed of 2,869 women in unifetal pregnancy of uncomplicated history. Examinations were made with use of Toshiba Capasee (Japan) and Echoson-Spinel (Poland) apparatus with transabdominal probes of 3.5 MHz frequency and transvaginal probes of 7 MHz frequency, Results. An average age of examined women was 27.2 years (SD ± 4.9 years). Among examined women primigravidas accounted for 46.5%. In examined population 8..5% of deliveries occurred before completion of 37th week of pregnancy. Most deliveries took place in 41st week of pregnancy and they accounted for 24.9%. In examined population the analysis of difference between estimated values and actual fetal weight obtained by way of modelling with use of multiple regression indicated that an average error in estimating the birth weight was 6.39%. Of numerous ANN networks, four with the highest prognostic values have been selected finally. For the best from among constructed network an average error in prognosing the fetal weight was 3.36%/. Conclusion. Among currently applied methods of statistical modelling of ultrasonographic biometry results, the highest prognostic accuracy of dystrophy is indicated by models based on computer-generated artificial neural networks.