Loading...

Use of Deep Learning in Ultrasound Videos to Improve the Automatic Detection of the Nuchal Translucency in the First Trimester of Pregnancy

Alimoradi , Fatemeh | 2021

258 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 54267 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Jahed, Mehran; Kaveh Vash, Zahra
  7. Abstract:
  8. Identifying and interpreting standard fetal plates in first trimester examinations with two-dimensional ultrasound is a very complex task that requires years of experience. Aside from the challenge to position and direct the probe to required positions, there is an understandable difficultly for inexperienced sonologists to identify the corresponding structures in the emerging images. Automated image processing can be helpful both to the experienced as well as inexperienced operators to confront these challenges. In this study, we used data from three different devices SAMSUNG WS80A, VOLUSON E6, VOLUSON E8, which includes a total of 2230 images to measure the thickness of the space behind the fetal neck, of which 600 images were standard images of the fetus in the sagittal view, in addition to 15 videos. Our research consists of two stages. First, we propose a convolutional neural network-based method, which can automatically detect the standard fetal sagittal plane to measure the thickness of the nuchal translucency in ultrasound videos, which is crucial in estimating fetal malformations. The accuracy of the trained grid reached 93.125%, which promises an efficient method for automatic detection of the standard plane to measure the thickness of the nuchal translucency. Notably, we did not find any research that considered the standard plane of ultrasound videos using deep learning. In the second step, we found the desired area on the standard plane obtained from the previous step, which contained a nuchal translucency. The basis of the methods used to find the desired area is deep convolutional neural networks that in this thesis, we specifically used Faster Region-based Convolutional Neural Network
  9. Keywords:
  10. Deep Learning ; Transfer Learning ; Object Detection ; Sonography ; Region-Based Convolutional Neural Network ; Nuchal-Translucency Thickness ; Standard Plane Detection

 Digital Object List

 Bookmark

No TOC