Title: Gabor fully convolutional network and ellipse fitting technique for foetal head segmentation and biometry measurement

Authors: Ahmed Zaafouri; Hanene Sahli; Radhouane Rachdi; Mounir Sayadi

Addresses: ENSIT, SIME, University of Tunis, 1008, Montfleury, Tunis, Tunisia ' ENSIT, SIME, University of Tunis, 1008, Montfleury, Tunis, Tunisia ' Department of Maternity, Military Hospital, Tunis, Tunisia ' ENSIT, SIME, University of Tunis, 1008, Montfleury, Tunis, Tunisia

Abstract: This paper introduces a new approach for foetal head segmentation and biometry measurement based on Gabor fully convolutional networks (G-FCN) along with the ellipse fitting technique. A fully convolutional network (FCN) training process based on Gabor features is presented. The new approach tends to accelerate the training stage and gives successful results. The Gabor wavelets with their steerable properties (i.e., their scales and orientations) are able to reinforce the robustness of G-FCN and reduce the training complexity. The proposed model is applied for foetal US image segmentation and foetal head circumference (HC) measurement using the elliptical fit technique. Our datasets are provided from a radiographic sequence of the foetus during different periods of pregnancy. An experimental study is conducted to prove the usefulness of the proposed algorithm for foetal biometric purposes. In addition, the automated approach makes it easier for doctors to diagnose US images.

Keywords: ultrasound images; ellipse fitting; fully convolutional network; Gabor CNNs; Gabor wavelets.

DOI: 10.1504/IJBET.2024.140561

International Journal of Biomedical Engineering and Technology, 2024 Vol.45 No.4, pp.354 - 367

Received: 20 Aug 2023
Accepted: 03 Dec 2023
Published online: 23 Aug 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article