Object separation in CT images using random walk
(2025) BMEM01 20251Department of Biomedical Engineering
- Abstract
- The segmentation and separation of objects in 3D medical imaging, particularly in Computed Tomography (CT), are essential for accurate diagnostics and preoperative planning. This master's thesis presents a tool for the interactive and automatic separation of segmented bones in CT scans, using the Random Walker algorithm and neural networks. The tool was developed to address challenges such as fused bones and computational efficiency while maintaining user-friendliness.
The interactive component enables users to manually place seed points, guiding the Random Walker algorithm to achieve precise separations based on voxel intensity and connectivity. For automation, a neural network identifies inter-bone regions, enabling the tool to separate... (More) - The segmentation and separation of objects in 3D medical imaging, particularly in Computed Tomography (CT), are essential for accurate diagnostics and preoperative planning. This master's thesis presents a tool for the interactive and automatic separation of segmented bones in CT scans, using the Random Walker algorithm and neural networks. The tool was developed to address challenges such as fused bones and computational efficiency while maintaining user-friendliness.
The interactive component enables users to manually place seed points, guiding the Random Walker algorithm to achieve precise separations based on voxel intensity and connectivity. For automation, a neural network identifies inter-bone regions, enabling the tool to separate all bones within a CT scan autonomously. Optimizations such as bounding boxes and GPU acceleration ensure computational feasibility, achieving interactive splitting times of under 20 seconds.
Results demonstrate the tool's robustness and versatility, performing well across various anatomies and clinical cases. It is integrated into regulatory-approved software, Segment 3DPrint, and has received positive feedback from beta testers. Limitations include dependence on pre-segmented images and computational demands for automatic separation.
This work contributes a clinically applicable, efficient solution for complex 3D CT segmentation, with potential applications in diverse medical and research contexts. (Less) - Popular Abstract
- Unlocking the Secrets of CT Scans: AI in Bone Separation
In the world of modern medicine, imaging technologies like Computed Tomography (CT) scans are indispensable. These high-resolution 3D images allow doctors to see inside the body with incredible detail, aiding in everything from diagnosing diseases to planning surgeries. But while these images provide invaluable insights, they also come with challenges — one of the biggest being the accurate separation of different structures, such as bones, within a single scan.
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9187314
- author
- Sobel, Rasmus LU
- supervisor
- organization
- alternative title
- Objektseparation i CT-bilder med hjälp av random walk
- course
- BMEM01 20251
- year
- 2025
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- CT images, Bone segmentation, Random Walker algorithm, Neural network, U-Net, Segment3D Print
- language
- English
- additional info
- 2025-06
- id
- 9187314
- date added to LUP
- 2025-06-11 13:56:27
- date last changed
- 2025-06-11 13:56:27
@misc{9187314, abstract = {{The segmentation and separation of objects in 3D medical imaging, particularly in Computed Tomography (CT), are essential for accurate diagnostics and preoperative planning. This master's thesis presents a tool for the interactive and automatic separation of segmented bones in CT scans, using the Random Walker algorithm and neural networks. The tool was developed to address challenges such as fused bones and computational efficiency while maintaining user-friendliness. The interactive component enables users to manually place seed points, guiding the Random Walker algorithm to achieve precise separations based on voxel intensity and connectivity. For automation, a neural network identifies inter-bone regions, enabling the tool to separate all bones within a CT scan autonomously. Optimizations such as bounding boxes and GPU acceleration ensure computational feasibility, achieving interactive splitting times of under 20 seconds. Results demonstrate the tool's robustness and versatility, performing well across various anatomies and clinical cases. It is integrated into regulatory-approved software, Segment 3DPrint, and has received positive feedback from beta testers. Limitations include dependence on pre-segmented images and computational demands for automatic separation. This work contributes a clinically applicable, efficient solution for complex 3D CT segmentation, with potential applications in diverse medical and research contexts.}}, author = {{Sobel, Rasmus}}, language = {{eng}}, note = {{Student Paper}}, title = {{Object separation in CT images using random walk}}, year = {{2025}}, }