Clustering: Is it the future in CTG evaluation?
Eleni Drosou, Václav Chudáček · 2015
Purpose of this paper is to present a computerized way to evaluate CTG recordings, and more specifically use of feature clustering for the classification process. We used a database which contained 552 records and 20 features. Matlab (version R2012a) was used for the experiments. First we performed a reduction of the number of features used in order to end up only with the most useful ones. That set up was done by performing a Kruskal-Wallis test and application of a correlation procedure. Next step was the actual clustering of the remaining features using a k-means function. The resulting clusters were plotted. According to the values of sensitivity, specificity and F-score the best results (clusters containing pathological data) were picked and evaluated. The ultimate goal is a better understanding of the natural clusters in the CTG recordings without dependence onto obstetricians' assessment which is based more on their experience and less on objective technical or clinical features.