Fetal Distress Classification with Deep Convolutional Neural Network

Abstract

Objective: Our study aimed to provide an improvised model that classifies the fetal distress using a two-dimensional Convolution neural network (CNN). It also helps in improving the visualization of FHR and UC signals. Background: Hypoxia or Fetal Distress is the main cause of death in the newborns. Cardiotocography is used to detect hypoxia in which fetal heart rate and uterine contraction signals are observed. Setting: Department of Computer Engineering and Technology, Guru Nanak Dev University, India. Subjects: The CTG-UHB database was used for classification purpose and 552 records were analyzed for classification purposes. Methods: Convolutional Neural Network was used for the classification purpose and HoloViz was used for the visualization of data in which HvPlot and HoloViews libraries are used in python. The CTG-UHB database was used for the analysis purpose. A total of 552 records were used for classification purposes. The classification was performed on the Keras software. Results: The accuracy achieved by our model was above 70%. Three classes were obtained named Normal Hypoxia (pH>7.15), Mild Hypoxia (7.05<pH<7.15), and Severe Hypoxia (pH<7.05). The accuracy achieved by normal hypoxia was 70%, mild hypoxia achieved 71.4%, whereas severe hypoxia class achieved 70% accuracy. Conclusion: This model describes the state of the fetus by using the classification based on pH values and also our proposed model visualizes the signals using HoloViz. The accuracy achieved outperforms other models mentioned in the literature.