University of Twente Student Theses

Login

Towards stable feature detection for 3D reconstruction of the placental surface during the FLOVA procedure - Convolutional neural network based feature detection

Buser, Myrthe (2020) Towards stable feature detection for 3D reconstruction of the placental surface during the FLOVA procedure - Convolutional neural network based feature detection.

Full text not available from this repository.

Full Text Status:Access to this publication is restricted
Embargo date:1 January 2022
Abstract:Twin pregnancy can be complicated by Twin-to-Twin-Transfusion Syndrome (TTTS), which can be cured with Fetoscopic Laser Occlusion of Vascular Anastomoses (FLOVA). During the FLOVA procedure, the placental anastomoses causing TTTS are coagulated under fetoscopic view. The field of view during FLOVA is limited due the need of using instruments of a small size. To artificially aid the field of view, 3D FLOVA-SLAM was developed as a 3D placental surface reconstruction method. However, the feature detection method Oriented BRIEF and Rotated FAST (ORB) in 3D FLOVA-SLAM shows certain problems when applied on placental images. To solve these problems, the new feature detection method F-net was developed, based on a Convolutional Neural Network (CNN). To create a training set for F-net, stable features were selected, using the information of the reconstruction of 3D FLOVA-SLAM. The performance of F-net was compared to the performance of ORB. It was shown that F-net detects less features and has a higher prediction time than ORB, but detected features are more distinct. The training of F-net on selected features increases feature distinctiveness. Implementing F-net in 3D FLOVA-SLAM is likely to increase the quality of the placental reconstruction.
Item Type:Essay (Master)
Clients:
Radboudumc
Faculty:TNW: Science and Technology
Subject:44 medicine, 54 computer science
Programme:Technical Medicine MSc (60033)
Link to this item:http://purl.utwente.nl/essays/80704
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page