A whole 3D liver reconstruction for personalised preoperative surgery based on FCN U-net model segmentation
by Amina Benahmed; Kamila Khemis; Loudjedi Salim
International Journal of Biomedical Engineering and Technology (IJBET), Vol. 44, No. 2, 2024

Abstract: The vascular mapping of liver is part of personalised medicine which allows optimal management of the patient during surgery. The work that we propose allows 3D reconstruction of the liver with its vascular tree. The main steps followed to achieve this processing are image pre-processing based on Hounsfield windowing, automatic segmentation of liver, then vascular tree extraction. Liver automatic segmentation is a challenging stage because of the inter-patient variability of the liver shape and its similar grey level with neighbouring organs. The deep learning approach fits this problem. We applied fully convolutional networks (FCN) 'U-Net' with 38 layers, 40 connections, Adam's algorithm optimiser and learning rate (LR) of 0.001. We tested this model on two CT scan databases: 3Dircadb and Task08_HepaticVessels. The results obtained, validated quantitatively, are very satisfactory (dice coefficient 98% and loss function 2%), comparable to the literature and confirm the robustness of U-Net for automatic liver segmentation.

Online publication date: Thu, 29-Feb-2024

The full text of this article is only available to individual subscribers or to users at subscribing institutions.

 
Existing subscribers:
Go to Inderscience Online Journals to access the Full Text of this article.

Pay per view:
If you are not a subscriber and you just want to read the full contents of this article, buy online access here.

Complimentary Subscribers, Editors or Members of the Editorial Board of the International Journal of Biomedical Engineering and Technology (IJBET):
Login with your Inderscience username and password:

Username:        Password:          Forgotten your password?

Want to subscribe?
A subscription gives you complete access to all articles in the current issue, as well as to all articles in the previous three years (where applicable). See our Orders page to subscribe.

If you still need assistance, please email subs@inderscience.com