Related Experiment Video
Updated: Sep 6, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
Segmentation of liver tumors with abdominal computed tomography using fully convolutional networks
Chih-I Chen1,2,3,4,5, Nan-Han Lu6,7,8, Yung-Hui Huang8
1Division of Colon and Rectal Surgery, Department of Surgery, E-DA Hospital, Kaohsiung City, Taiwan.
Journal of X-Ray Science and Technology
|June 27, 2022
Summary
This study developed an automated deep learning model for liver tumor segmentation on CT scans, achieving over 90% accuracy. While effective for many tumors, segmenting very small ones remains challenging.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Manual segmentation of liver tumors on CT scans is time-consuming and costly.
- Automated segmentation is crucial for efficient tumor staging and treatment planning.
Purpose of the Study:
- To develop and validate a deep learning network for automatic liver tumor segmentation.
- To fine-tune parameters for optimal performance of the segmentation model.
Main Methods:
- Utilized 7,190 2D CT images from 131 patients with liver tumors.
- Employed Fully Convolutional Networks (FCN) with various backbones (Xception, InceptionresNetv2, MobileNetv2, ResNet18, ResNet50).
- Investigated optimizers, epoch size, and batch size, evaluating with metrics like Mean IoU and BF Score.
Main Results:
- Achieved high accuracy with Mean IoU of 0.954 and Mean BF Score of 0.962 using ResNet50 with SGDM optimizer.
- Top FCN models demonstrated Mean IoU exceeding 0.900.
- InceptionresNetv2 showed superior performance among the tested backbones.
Conclusions:
- Developed an automated FCN-based model for liver tumor segmentation with high accuracy (>90%).
- Deep learning models show significant potential for liver tumor segmentation from CT images.
- Accurate segmentation of small and tiny liver tumors remains a challenge for current FCN models.
