Related Experiment Video
Updated: Jan 7, 2026
Automated Midline Shift and Intracranial Pressure Estimation based on Brain CT Images
Published on: April 13, 2013
Open area segmentation in CT images based on pixel displacement and multi-view with application in the axillary and
Tiange Liu1,2, Yajie Wang2, Drew A Torigian3
1School of Intelligence Science and Technology, University of Science and Technology Beijing, Beijing, China.
Background:
In medical image segmentation, the segmentation of open regions presents challenges and remains relatively unexplored. Unlike objects such as organs that have well-defined boundaries, open regions typically lack clear boundaries, complicating the segmentation process.
Purpose:
This study aims to propose a new segmentation algorithm to address these challenges in the segmentation of open regions and to improve the accuracy and reliability of segmentation.
Approach:
This paper introduces a segmentation algorithm inspired by the concept of optical flow, which leverages pixel displacement information within a 2.5D CNN framework. Our framework fully exploits the potential of multi-view networks, achieving precise segmentation of open areas in axillary and lower cervical regions by integrating information from multiple views and pixel tracking capabilities.
Results:
We validated the proposed algorithm using two CT datasets from the axillary region and the lower cervical region. This dataset includes a total of 100 patient studies, which are randomly selected and allocated, with 50 allocated for training, 10 for validation, and 40 for testing purposes. Experimental results show that the proposed algorithm achieves Dice coefficients of 90.25% and 80.43% in these two regions, surpassing the state-of-the-art methods by over 1.32% and 1.85%, respectively. The ablation experiments confirm that both the multi-channel ensemble and pixel displacement guidance strategies enhance segmentation accuracy in open regions.
Conclusions:
The algorithm captures the dynamic changes occurring in the region by analyzing the intensity changes information of pixels between adjacent slices, effectively addressing the segmentation challenges of open region boundaries. This is particularly evident in areas with step-like changes, where the algorithm incorporates pixel displacement information as prior knowledge along with a multi-channel ensemble strategy to enhance segmentation accuracy and robustness.
