Related Experiment Video
Updated: Oct 14, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
HFRU-Net: High-Level Feature Fusion and Recalibration UNet for Automatic Liver and Tumor Segmentation in CT Images
Devidas T Kushnure1, Sanjay N Talbar2
1Department of Electronics and Telecommunication Engineering, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded, Maharashtra, India; Department of Electronics and Telecommunication Engineering, Vidya Pratishthan's Kamalnayan Bajaj Institute of Engineering and Technology, Baramati, Maharashtra, India.
This study introduces HFRU-Net, a deep learning model for automatic liver and tumor segmentation in CT scans. The model achieves high accuracy, improving hepatic disease detection and treatment planning.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Accurate liver and tumor segmentation in CT scans is crucial for hepatic disease management.
- Manual segmentation is time-consuming, operator-dependent, and challenging due to image complexities.
- Deep learning offers a promising solution for automated medical image analysis.
Purpose of the Study:
- To develop an advanced deep learning model, HFRU-Net, for precise automatic segmentation of liver and tumors in CT images.
- To improve upon existing segmentation methods by enhancing feature representation and reducing computational complexity.
- To validate the model's performance on public datasets and compare it with state-of-the-art techniques.
Main Methods:
- The proposed HFRU-Net model modifies the UNet architecture with local feature reconstruction and feature fusion in skip pathways.
- Incorporates squeeze-and-Excitation network (SENet) for adaptive recalibration of fused features.
- Utilizes atrous spatial pyramid pooling (ASPP) in the bottleneck layer for multiscale feature representation.
Main Results:
- HFRU-Net achieved high Dice Similarity Coefficients: 0.966/0.972 for liver and 0.771/0.776 for tumor segmentation on LiTS and 3DIrcadb datasets.
- Demonstrated robust performance on the independent LiTS challenge test dataset with 95.0% liver and 61.4% tumor segmentation accuracy.
- Outperformed existing methods in segmentation accuracy while reducing computational complexity.
Conclusions:
- HFRU-Net significantly enhances automatic liver and tumor segmentation accuracy in CT images.
- The model's architecture effectively captures detailed contextual and multiscale features for improved segmentation.
- HFRU-Net represents a robust and efficient tool for clinical applications in hepatic disease diagnosis and treatment planning.
