Related Experiment Video
Updated: Jun 1, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.6K
Mixed reality infrastructure based on deep learning medical image segmentation and 3D visualization for bone tumors
Kun Wang1, Yong Han2, Yuguang Ye3,4,5
1Institute of Design, Quanzhou Normal University, Quanzhou 362000, China.
Journal of Bone Oncology
|January 22, 2025
Summary
A novel deep learning model, DCU-Net, enhances 3D bone tumor segmentation and reconstruction accuracy. This improves surgical planning and spatial understanding for clinicians using mixed reality, boosting diagnostic and treatment efficacy.
Area of Science:
- Oncological imaging
- Medical image analysis
- 3D reconstruction
Background:
- Accurate segmentation and 3D modeling of bone tumors from 2D images are crucial for diagnosis and treatment.
- Existing methods struggle with low distinguishability between tumors and surrounding tissues, impacting accuracy and stability.
Purpose of the Study:
- To develop an advanced deep learning model (DCU-Net) for improved oncological image segmentation and 3D reconstruction of bone tumors.
- To explore the application of a mixed reality (MR) infrastructure integrating deep learning segmentation with MR for bone tumor diagnosis and treatment.
Main Methods:
- A U-Net model with double dimensionality reduction and channel attention gating (DCU-Net) was optimized for feature extraction and target space clustering.
- Experiments utilized a hospital dataset for bone tumor segmentation and 3D reconstruction, evaluated using DSC, R, P, and VDE.
- Clinical examination compared 2D image viewing versus MR infrastructure for surgical planning effectiveness using Likert scale (LS).
Main Results:
- The DCU-Net model achieved segmentation performance metrics (DSC, R, P) exceeding 90%, outperforming U-Net and Attention-U-Net.
- 3D reconstruction demonstrated high fidelity in reflecting individual tumor characteristics.
- Clinicians using the DCU-Net-based MR system showed enhanced spatial awareness for preoperative planning compared to 2D methods.
Conclusions:
- The DCU-Net deep learning model significantly improves tumor CT image segmentation and 3D model reconstruction accuracy.
- The integrated MR system enhances clinicians' understanding of tumor morphology and spatial relationships, showing great potential for clinical practice.
- This approach is expected to promote advancements in the diagnosis and treatment of bone tumors, ultimately improving patient outcomes.
