Related Experiment Video
Updated: Nov 2, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.1K
MIDeepSeg: Minimally interactive segmentation of unseen objects from medical images using deep learning
Xiangde Luo1, Guotai Wang1, Tao Song2
1School of Mechanical and Electrical Engineering, University of Electronic Science and Technology of China, Chengdu, China.
Medical Image Analysis
|June 12, 2021
Summary
This study introduces a new deep learning method for medical image segmentation that requires minimal user clicks for accurate results. The approach effectively segments both familiar and novel organs or lesions, improving efficiency and generalization.
Area of Science:
- Medical Imaging
- Computer Vision
- Machine Learning
Background:
- Accurate medical image segmentation is crucial for diagnosis and treatment planning.
- Convolutional Neural Networks (CNNs) excel at automatic segmentation but struggle with accuracy and robustness in complex cases.
- Interactive segmentation offers a practical alternative but often demands extensive user input or performs poorly on unseen objects.
Purpose of the Study:
- To develop a novel deep learning-based interactive segmentation method that is efficient and generalizes to unseen objects.
- To reduce the number of user interactions required for accurate medical image segmentation.
- To improve the robustness of interactive segmentation methods for diverse clinical applications.
Main Methods:
- Proposed a deep learning framework utilizing user-provided interior margin points encoded via exponentialized geodesic distance for initial segmentation.
- Implemented a novel information fusion technique to refine segmentation using minimal additional user clicks.
- Validated the framework on 2D and 3D medical image segmentation tasks involving a wide range of unseen objects.
Main Results:
- The proposed method achieved accurate segmentation results with significantly fewer user interactions and less time compared to state-of-the-art interactive frameworks.
- Demonstrated strong generalization capabilities, performing well on previously unseen organs and lesions not present in the training data.
- The framework proved effective for both 2D and 3D medical image segmentation tasks.
Conclusions:
- The novel deep learning-based interactive segmentation method offers high efficiency and excellent generalization for medical imaging.
- This approach addresses the limitations of traditional and existing CNN-based interactive segmentation methods.
- The framework shows significant potential for improving clinical workflows in diagnosis and treatment planning through accurate and rapid segmentation.
