Automated identification of abnormal cardiotocograms using neural network visualization techniques

The cardiotocogram (CTG) is a display of the fetal heart rate and maternal uterine activity over time. An automated system for CTG analysis can be used as a decision support tool in a clinical setting. We present an automated system for the identification of abnormal patterns in the intrapartum (labor) CTG. We extract discriminating features from the CTG and then use techniques based upon the Neuroscale algorithm to project these features onto a two-dimensional visualization space. The locations of the projected features in the visualization space correlate retrospectively with an expert's assessment of the CTG's pattern.