Related Experiment Video
Updated: Aug 26, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
Deep supervision and atrous inception-based U-Net combining CRF for automatic liver segmentation from CT.
Peiqing Lv1, Jinke Wang2,3, Xiangyang Zhang1
1School of Automation, Harbin University of Science and Technology, Harbin, 150080, China.
Scientific Reports
|October 10, 2022
Summary
This study presents an improved deep learning method for segmenting livers in CT scans, enhancing accuracy for challenging low-contrast images. The novel approach effectively addresses blurred boundaries, leading to more reliable automated liver segmentation.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Liver segmentation in CT images is challenging due to low contrast and blurred boundaries.
- Existing methods struggle with accurate segmentation of liver tissue and neighboring organs with similar intensity values.
Purpose of the Study:
- To develop an improved deep learning method for accurate liver segmentation in CT images.
- To address the challenges of low contrast and fuzzy boundaries in liver segmentation.
Main Methods:
- Utilized deep supervision (DS) and atrous inception (AI) technologies with conditional random fields (CRF).
- Incorporated residual blocks in the encoder, an AI module for multi-scale features, and DS in the decoder.
- Employed Tversky loss function and dense CRF for refinement.
Main Results:
- The proposed method significantly increased segmentation accuracy for livers with low contrast and fuzzy boundaries.
- Achieved superior performance compared to state-of-the-art methods on public datasets (LiTS17, 3DIRCADb, SLiver07).
Conclusions:
- The novel approach enhances automatic liver segmentation, particularly in challenging low-contrast scenarios.
- The method offers a more robust and accurate solution for clinical applications requiring precise liver segmentation.
