Related Experiment Video
Updated: Feb 18, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
Multi-encoder U-Net benchmarking for LiTS17 Liver-Tumor segmentation: accuracy-efficiency trade-offs across training
Yuguang Ye1,2,3, Kavimbi Chipusu1,2,3, Taisheng Zeng1,2,3
1Faculty of Mathematics and Computer Science, Quanzhou Normal University, Quanzhou, China.
Biomedizinische Technik. Biomedical Engineering
|February 16, 2026
Summary
Accurate liver and tumor segmentation using CT scans is crucial for liver cancer care. This study benchmarks U-Net and V-Net models, finding that longer training and stronger encoders significantly improve tumor segmentation accuracy.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Radiology
- Computational Pathology
Background:
- Accurate liver and tumor segmentation from CT scans is essential for liver cancer diagnosis, treatment planning, and monitoring.
- Existing U-Net variants and volumetric architectures like V-Net lack standardized comparison, hindering optimal model selection.
Purpose of the Study:
- To establish a unified benchmarking framework for evaluating 2D U-Net models with various encoders and a 3D V-Net baseline.
- To analyze the impact of encoder selection, training duration, and computational cost on segmentation performance.
Main Methods:
- A standardized framework using 3-fold cross-validation on the LiTS17 dataset.
- Evaluation of multiple U-Net backbones (VGG, ResNet, MobileNetV2) and a V-Net baseline.
- Assessment of 15, 50, and 100-epoch training schedules, reporting overlap (Dice, IoU) and detection (precision, recall) metrics alongside efficiency indicators.
Main Results:
- Liver segmentation performance quickly reached near-optimal levels across all tested models.
- Tumor segmentation accuracy significantly improved with extended training durations and more powerful encoder networks, particularly for challenging small or low-contrast lesions.
Conclusions:
- The study provides a reproducible protocol for benchmarking medical image segmentation models.
- Offers practical guidance for selecting models that effectively balance segmentation accuracy, robustness, and computational cost for liver cancer applications.
