Related Experiment Video
Updated: Oct 12, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
U-Net: A valuable encoder-decoder architecture for liver tumors segmentation in CT images
Hanene Sahli1, Amine Ben Slama2, Salam Labidi2
1Laboratory of Signal Image and Energy Mastery (SIME), LR13ES03, University of Tunis, ENSIT, 1008, Tunis, Tunisia.
Journal of X-Ray Science and Technology
|November 22, 2021
Summary
This study introduces an advanced segmentation method for detecting liver tumors in CT scans. The approach enhances diagnostic accuracy for metastasis detection, improving patient treatment outcomes.
Area of Science:
- Medical Imaging
- Radiology
- Oncology
Background:
- Accurate localization of liver metastasis lesions in computed tomography (CT) images is critical for diagnosis and treatment planning.
- Existing methods face challenges in precise lesion identification, impacting patient management and therapeutic follow-up.
- Improving diagnostic processes is essential for increasing the success rate of liver cancer management.
Purpose of the Study:
- To propose and evaluate a novel predictive segmentation method for liver tumor detection in CT images.
- To develop an automated algorithm for segmenting liver tumors using deep learning architectures.
- To enhance the volumetric analysis and precise localization of metastatic liver lesions.
Main Methods:
- A computerized approach utilizing an encoder-decoder structure, specifically Seg-Net and U-Net architectures, was developed.
- The method was applied to segment liver tumors from metastasis CT images.
- A dataset of 8,297 CT slices from 200 pathologically confirmed metastasis cancer cases was used for training and validation (85% training, 15% validation).
Main Results:
- The proposed segmentation method achieved high performance, indicated by an F1-score of 0.9573, Recall of 0.9520, and IOU of 0.9654.
- Low binary cross-entropy (0.0032) and a p-value < 0.05 demonstrate statistical significance.
- The approach showed superior segmentation performance compared to state-of-the-art techniques, yielding higher precision in identifying metastasis tumor positions.
Conclusions:
- The developed automatic segmentation algorithm demonstrates significant potential for accurate liver tumor detection and localization in CT imaging.
- The encoder-decoder based approach, leveraging Seg-Net and U-Net, offers a robust solution for volumetric analysis of liver metastasis.
- This method improves diagnostic accuracy and precision, contributing to better patient management and therapeutic strategies in liver cancer care.
