|
|
Modelling of perfusion curves in dynamic magnetic resonance
Ochodnický, Erik ; Mangová, Marie (referee) ; Rajmic, Pavel (advisor)
Perfusion MRI can provide information about perfusion characteristics of the observed tissue, which makes it a widely applicable medical procedure. Measuring process of MRI is very time-consuming, and therefore, using classical reconstruction methods, we are often not able to obtain enough samples to accomplish the needed time and space resolution for perfusion analysis. That is why it is necessary to use compressed sensing, which allows reconstruction from under-sampled data by solving an optimization model. In this work, several models for reconstruction of an image sequence are verified on real and artificial data, along with multiple algorithms capable of solving these models. Among the optimization models used in this work are two L+S models with different regularization of the S component that are solved by Forward-Backward and Chambolle-Pock algorithm. The quality of reconstruction for various models was compared especially by their perfusion curves. In the last section, we explore possible modifications of the SASS model in order to increase quality of reconstruction and resistance to under sampling for the purpose of better adaptation for dynamic data.
|
|
|
Modelling of perfusion curves in dynamic magnetic resonance
Ochodnický, Erik ; Mangová, Marie (referee) ; Rajmic, Pavel (advisor)
Perfusion MRI can provide information about perfusion characteristics of the observed tissue, which makes it a widely applicable medical procedure. Measuring process of MRI is very time-consuming, and therefore, using classical reconstruction methods, we are often not able to obtain enough samples to accomplish the needed time and space resolution for perfusion analysis. That is why it is necessary to use compressed sensing, which allows reconstruction from under-sampled data by solving an optimization model. In this work, several models for reconstruction of an image sequence are verified on real and artificial data, along with multiple algorithms capable of solving these models. Among the optimization models used in this work are two L+S models with different regularization of the S component that are solved by Forward-Backward and Chambolle-Pock algorithm. The quality of reconstruction for various models was compared especially by their perfusion curves. In the last section, we explore possible modifications of the SASS model in order to increase quality of reconstruction and resistance to under sampling for the purpose of better adaptation for dynamic data.
|
guest ::
