Progetto di sistemi fuzzy applicati alla classificazione di stati patologici fetali

During the fetal life, some growth and nutrition alterations often appear and they can bring chronic fetal suffering conditions. We would like to analyze in this work the cases of intra uterine growth retardation (IUGR = IntraUterine Growth Retardation) or nutrition alterations caused by maternal diabetes (DM) that imply serious consequences for the fetus.
Nevertheless, many among these suffering conditions are very difficult to be diagnosed so it is very important to find out all the data concerning the fetus' health conditions. Moreover, the very delicate structure of the fetus requires the use of non-invasive methods to find out the necessary data.
The fetal hearth rate signal (FHR = Fetal Heart Rate) is rich in information about the fetus' health conditions. The fetal heart rate signal can be obtained through electrocardiographic, phonocardiographic and ultrasonocardiographic approaches. We have considered in this study all signals derived from ultrasonocardiographic recordings that is to say by means of Doppler probes, thus in a spread and noninvasive way. By means of Hewlett Packard (Böblingen) ultrasound cardiotocograph “HP M135XA”, the cardiotocographic signal was extracted that is to say the cardiographic signal (heartbeats) together with the tocographic signal (contractile activity of the uterus).
The "at sight" exam of the cardiotocographic signal, used in the clinical practice, has been reducing intra-delivery and premature neonatal death for a number of years. There are still many problems connected with interindividual and intraindividual variability concerning the interpretation of the curve [Van Geijn 1996].
A number of systems to extract the parameters starting from the cardiotocographic signal were developed and in this way intraindividual variability was highly reduced. But the clinical meaning of the different parameters isn't still completely clear so the problems concerning interindividual variability haven't been solved yet.
We used for this project all the data that were recorded by the Obstetrics Unit of prof. Domenico Arduini of the University of Rome "Tor Vergata". A number of parameters were extracted by FHR layouts (curves), and we used some parameters' vectors concerning each separate recording. Different kinds of parameters were used: the morphologic ones, such as the baseline, accelerations, decelerations and others measuring the time scale signal characteristics, parameters from the autoregressive, frequency domain spectral analysis (such as signal power and frequency components) and the approximate entropy. Moreover, even the gestational age was considered as a parameter, taking into account the great importance of the fetus growing conditions in finding out all suffering signals. We considered only fetuses with more than 34 gestational weeks since the available data concerning lower ages, due to recording difficulties, were only a few. All data concerning populations different from the Normal, IUGR and maternal diabetes were excluded. According to the results of an advanced statistical analysis, previously developed in the same project, we identified a set of 10 parameters (mean, std, number of low amplitude accelerations, Interval Index, LF power spectrum normalized %, MF power spectrum normalized %, HF power spectrum normalized %, LF/(MF+HF), ApEn(2, 0.1), gestational weeks) out of all extracted ones. The statistical analysis was based on the assessment of correlation degree among parameters trying to identify less correlated ones in a multiparametric domain.
So it was decided to use for the analysis of each subject a vector of 10 parameters only
Since it was impossible to find out some thresholds of normality for each extracted parameter, it was decided to find out in this work the clusters for the subjects belonging to the three populations (normal, IUGR and maternal diabetes), inside the parameters' hyperspaces.