Feature selection for computerized fetal heart rate analysis using genetic algorithms.

During birth, timely and accurate diagnosis is needed in order to prevent severe conditions such as birth asphyxia. The fetal heart rate (FHR) is often monitored during labor to assess the condition of fetal health. Computerized FHR analysis is needed to help clinicians identify abnormal patterns and to intervene when necessary. The objective of this study is to apply Genetic Algorithms (GA) as a feature selection method to select a best feature subset from 64 FHR features and to integrate these best features to recognize unfavorable FHR patterns. The GA was trained on 408 cases and tested on 102 cases (both balanced datasets) using a linear SVM as classifier. 100 best feature subsets were selected according to different splits of data; a committee was formed using these best classifiers to test their classification performance. Fair classification performance was shown on the testing set (Cohen's kappa 0.47, proportion of agreement 73.58%). To our knowledge, this is the first time that a feature selection method has been tested for FHR analysis on a database of this size.