Fetal Risk Classification Based on Cardiotocography Data: A Kernel-Based Approach

Abstract

The use of classification-based approaches for health state diagnosis is still an ongoing issue in medicinal applications. In this paper, we address the problem of fetal risk anticipation using cardiotocography (CTG) measurements. We adopt, for this purpose, a classification approach based on Support Vector Domain Description (SVDD). Indeed, the SVDD is an efficient kernel method employed to solve one-class problems known also as novelty detection problems. The fundamental goal of one-class learning is to generate a rule that distinguishes between a set of typical objects called target class and aberrant objects designated as outliers. Based on CTG data, the abnormalities anticipation/detection can be basically perceived as multi-classification problems since it consists in recognizing CTG patterns and classifying them into several classes according to the risk category. To do so, we have used a modified version of SVDD algorithm endowed with useful tools to manage the multi-classification problems. The aim is to generate, from a small number of CGT samples, a classification model that can be generalized and applied on a wider number of unknown samples. The proposed approach is assessed on real CTG database to prove its effectiveness.