Non-Rigid Volume Registration For DCE-MRI Using Deep Learning
(2021) In Master's Theses in Mathematical Sciences FMAM05 20202Mathematics (Faculty of Engineering)
- Abstract
- In dynamically contrast enhanced magnetic resonance imaging (DCE-MRI) mul-tiple images are generated for a single scan. Before further analysis can be done,the images needs to be registered. Traditional registration techniques can yieldimpressive results, but are often very computationally expensive.In this thesis a neural network model using spatial transformer networks isproduced. This model aims to solve the registration problem, while not beingtoo computationally expensive.Several versions of the model are trained, on a limited data set of DCE-MRI scans of rabbit livers. The registration accuracy of these versions are thencompared to each other, and to two baseline methods: ANTs and Elastix.The model is shown – with a similarity metric... (More)
- In dynamically contrast enhanced magnetic resonance imaging (DCE-MRI) mul-tiple images are generated for a single scan. Before further analysis can be done,the images needs to be registered. Traditional registration techniques can yieldimpressive results, but are often very computationally expensive.In this thesis a neural network model using spatial transformer networks isproduced. This model aims to solve the registration problem, while not beingtoo computationally expensive.Several versions of the model are trained, on a limited data set of DCE-MRI scans of rabbit livers. The registration accuracy of these versions are thencompared to each other, and to two baseline methods: ANTs and Elastix.The model is shown – with a similarity metric and visual inspection – to reg-ister the volumes less accurately than ANTs, but more accurately than Elastix.Notably this is done with orders of magnitude less computational expense aftertraining (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9062388
- author
- Ferm, Jonatan LU
- supervisor
- organization
- course
- FMAM05 20202
- year
- 2021
- type
- H2 - Master's Degree (Two Years)
- subject
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- LUTFMA-3447-2021
- ISSN
- 1404-6342
- other publication id
- 2021:E28
- language
- English
- id
- 9062388
- date added to LUP
- 2021-08-18 17:13:34
- date last changed
- 2021-08-18 17:13:34
@misc{9062388, abstract = {{In dynamically contrast enhanced magnetic resonance imaging (DCE-MRI) mul-tiple images are generated for a single scan. Before further analysis can be done,the images needs to be registered. Traditional registration techniques can yieldimpressive results, but are often very computationally expensive.In this thesis a neural network model using spatial transformer networks isproduced. This model aims to solve the registration problem, while not beingtoo computationally expensive.Several versions of the model are trained, on a limited data set of DCE-MRI scans of rabbit livers. The registration accuracy of these versions are thencompared to each other, and to two baseline methods: ANTs and Elastix.The model is shown – with a similarity metric and visual inspection – to reg-ister the volumes less accurately than ANTs, but more accurately than Elastix.Notably this is done with orders of magnitude less computational expense aftertraining}}, author = {{Ferm, Jonatan}}, issn = {{1404-6342}}, language = {{eng}}, note = {{Student Paper}}, series = {{Master's Theses in Mathematical Sciences}}, title = {{Non-Rigid Volume Registration For DCE-MRI Using Deep Learning}}, year = {{2021}}, }