University of Twente Student Theses
Automatic 3D pelvic landmark detection and 3D bladder segmentation from low-field MRI using 2.5D ForkNet
Schulte, H.J. (2023) Automatic 3D pelvic landmark detection and 3D bladder segmentation from low-field MRI using 2.5D ForkNet.
|
PDF
3MB |
| Abstract: | Pelvic organ prolapse is a component of pelvic floor dysfunction and is a big issue as it affects half of all women over the age of 50 years. With POP surgery, the risk of recurrence is about 10-30%. To evaluate the effect of surgery the patients are scanned with MRI before and after the operation. Manual delineation can be labour-intensive, and having a deep learning model that is able to do this properly, saves a lot of time and is more consistent than manual selection. This study uses the MRI data from women who participated in the TORBO study at the University of Twente. The U-Net model uses all the slices of one orientation in one batch and trains the model on patterns in 2D images in three directions, this is called 2.5D. ForkNet is used to integrate the 2.5D landmark detection and bladder segmentation in one model. The 2.5D U-Net model outperforms the 3D U-Net in bladder segmentation but falls short in landmark detection, while ForkNet presents an opportunity to integrate different anatomical features, albeit with the need for optimization in POP assessment. |
| Item Type: | Essay (Bachelor) |
| Faculty: | TNW: Science and Technology |
| Programme: | Biomedical Technology BSc (56226) |
| Link to this item: | https://purl.utwente.nl/essays/96834 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page