This thesis explores the application of diffusion models for the segmentation of brain tumors in 3D MRI images. By leveraging the robust generative capabilities of diffusion models, the research investigates how these models can accurately identify and segment tumor regions in volumetric MRI data. The study focuses on enhancing tumor detection accuracy and addressing challenges associated with 3D medical image segmentation, such as complex tumor shapes and varying intensities.
Thesis Type |
|
Student |
Ritabrata Sanyal |
Status |
Running |
Presentation room |
Seminar room I5 6202 |
Supervisor(s) |
Stefan Decker |
Advisor(s) |
Yongli Mou |
Contact |
mou@dbis.rwth-aachen.de |
Background
Brain tumor segmentation from MRI images is a crucial task in medical imaging, aiding in diagnosis, treatment planning, and patient monitoring. Traditional segmentation methods often struggle with the inherent variability in tumor morphology and the high dimensionality of 3D data. Diffusion models, known for their ability to generate complex data distributions, offer a promising approach to overcome these challenges by capturing the nuanced spatial features present in 3D MRI scans. This project aims to evaluate the efficacy of diffusion models for reliable and precise segmentation of brain tumors, comparing them to conventional and deep learning-based segmentation techniques.
Objectives
- Investigate the effectiveness of diffusion models for 3D MRI-based brain tumor segmentation
- Enhance segmentation accuracy and robustness in 3D medical imaging applications
- Compare diffusion models with traditional and contemporary segmentation methods
Tasks
Pre-process and curate a dataset of 3D MRI images annotated with brain tumor regions
- Implement and fine-tune diffusion models for the segmentation task
- Evaluate the performance of diffusion models against baseline methods
- Optimize model parameters to improve segmentation accuracy and computational efficiency
- Basic skills in Python
- Basic knowledge of machine learning and NLP concepts