Deep Learning Measurement Model to Segment the Nuchal Translucency Region for the Early Identification of Down Syndrome
Published Online: May 14, 2022
Page range: 187 - 192
Received: Nov 22, 2021
Accepted: Apr 19, 2022
DOI: https://doi.org/10.2478/msr-2022-0023
Keywords
© 2022 Mary Christeena Thomas et al., published by Sciendo
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Down syndrome (DS) or Trisomy 21 is a genetic disorder that causes intellectual and mental disability in fetuses. The most essential marker for detecting DS during the first trimester of pregnancy is nuchal translucency (NT). Effective segmentation of the NT contour from the ultrasound (US) images becomes challenging due to the presence of speckle noise and weak edges. This study presents a Convolutional Neural Network (CNN) based SegNet model using a Visual Geometry Group (VGG-16) for semantically segmenting the NT region from the US fetal images and providing a fast and affordable diagnosis during the early stages of gestation. A transfer learning approach using AlexNet is implemented to train the NT segmented regions for the identification of DS. The proposed model achieved a Jaccard index of 0.96 and classification accuracy of 91.7 %, sensitivity of 85.7 %, and a Receiver operating characteristic (ROC) of 0.95.