Final Report Summary - MRISIMUL (ENHANCED MRI PHYSICS SIMULATOR)
The MRISIMUL project involves the design and development of an enhanced comprehensive computer simulation platform of Magnetic Resonance imaging (MRI) physics, which aims to integrate realistic aspects of the MRI experiment from signal generation up to and including image reconstruction for the explicit purposes of 1) better understanding the mechanisms involved in artifact and contrast generation in cardiovascular MRI and 2) effectively teaching MRI physics to non-technical personnel.
This comprehensive computer simulation of MRI physics and image reconstruction will allow us to better understand realistic artifact generating mechanisms in cardiovascular MRI and will help in defining exam protocols that are optimized for suppressing such artifacts without resorting to animal and/or human experimentation. This research will allow for detailed evaluation of cardiovascular pulse sequences. With the design and development of the proposed simulator platform, the parameter space of different protocols will be tested and evaluated in detail without the time and cost constraints imposed by experimentation in an MRI scanner. We will also apply this tool to allow researchers, technologists, physicians and students to visualize the effects of protocol parameter changes in models of acute myocardial infarction. This will allow for further research in optimizing MRI protocols and also for training of non-technical personnel.
First, the basic computational core of the simulation platform was designed and developed based on the solution of the Bloch equations in three dimensions. We constructed a comprehensive MATLAB platform which allows the development of custom MRI pulse sequences and their application on virtual objects of simple shapes. The MRI simulator takes into account the main static field value and allows the introduction of an inhomogeneity field map to simulate the artifacts associated with the static magnetic field spatial variance, like those induced by susceptibility variation among different tissues. Moreover, it allows for the observation of the total magnetization vector components as well as for the three dimensional display of the total magnetization vector over time. Next, for educational purposes, a basic interactive interface was developed to display the temporal evolution of a number of magnetization vectors after the application of various RF pulse types in the presence of static field inhomogeneity.
The computational power required to accommodate reasonable execution times of the MRI simulator on a high end personal computer was evaluated next. Prohibitively long execution times on the order of several days were recorded. Furthermore, main random access memory size limitations hindered the application of the simulator algorithm to realistic size datasets. An alternative approach to CPU based processing was selected for subsequent evaluation; namely the use of a large number of processing cores within a graphic processing unit (GPU) in a single personal computer. Two high end CUDA-based graphic cards were used for this specific purpose. The MATLAB simulator was modified to accommodate a direct link to the GPU kernel so that pulse sequences could be designed with ease within the higher language environment while computationally intense algorithms could be executed in parallel within the GPU environment. A speedup of about 228 times when compared to serially executed C-code on the CPU was observed. The speedup was between 31 to 115 times when compared to the OpenMP parallel executed C-code on the CPU, depending on the number of threads used in multithreading (2 to 8 threads). Moreover, in multi-node, multi-GPU systems, MRISIMUL demonstrated an almost linear scalable performance along with the increasing number of the available GPU cards. Last, a detailed 3D model of the anatomy of the human heart and torso was developed based on the segmentation of high resolution tomographic images of the Visible Human Project. Motion, strain and respiration were implemented as well. The development of three different motion models, including a respiratory motion model, a heart motion model and a simple flow model, along with their integration with the simulator platform running on CUDA, allowed for motion simulations. The web access version of the simulator supports motion simulations (http://mri.dib.uth.gr).
The MRI simulator now allows for complex analysis of pulse sequences and protocols, which will advance our understanding and aid in the dissemination of cardiovascular MRI in a cost effective manner. We expect that the simulation platform will become an indispensable tool for medical imaging centers. We also expect that the proposed enhanced simulation platform will change the way by which MRI protocols are optimized. Moreover, it should change how MRI physics are taught to non-technical personnel (i.e. physicians, technologists). We expect that it will allow for a shift from classroom-teaching to a laboratory-style approach or a web-based approach without the need for access to an MRI scanner. The hands-on experience and 3D visualization that this simulation platform will provide will allow for easier understanding of basic MRI physics concepts and for direct feedback with respect to pulse sequence and protocol selection.
This comprehensive computer simulation of MRI physics and image reconstruction will allow us to better understand realistic artifact generating mechanisms in cardiovascular MRI and will help in defining exam protocols that are optimized for suppressing such artifacts without resorting to animal and/or human experimentation. This research will allow for detailed evaluation of cardiovascular pulse sequences. With the design and development of the proposed simulator platform, the parameter space of different protocols will be tested and evaluated in detail without the time and cost constraints imposed by experimentation in an MRI scanner. We will also apply this tool to allow researchers, technologists, physicians and students to visualize the effects of protocol parameter changes in models of acute myocardial infarction. This will allow for further research in optimizing MRI protocols and also for training of non-technical personnel.
First, the basic computational core of the simulation platform was designed and developed based on the solution of the Bloch equations in three dimensions. We constructed a comprehensive MATLAB platform which allows the development of custom MRI pulse sequences and their application on virtual objects of simple shapes. The MRI simulator takes into account the main static field value and allows the introduction of an inhomogeneity field map to simulate the artifacts associated with the static magnetic field spatial variance, like those induced by susceptibility variation among different tissues. Moreover, it allows for the observation of the total magnetization vector components as well as for the three dimensional display of the total magnetization vector over time. Next, for educational purposes, a basic interactive interface was developed to display the temporal evolution of a number of magnetization vectors after the application of various RF pulse types in the presence of static field inhomogeneity.
The computational power required to accommodate reasonable execution times of the MRI simulator on a high end personal computer was evaluated next. Prohibitively long execution times on the order of several days were recorded. Furthermore, main random access memory size limitations hindered the application of the simulator algorithm to realistic size datasets. An alternative approach to CPU based processing was selected for subsequent evaluation; namely the use of a large number of processing cores within a graphic processing unit (GPU) in a single personal computer. Two high end CUDA-based graphic cards were used for this specific purpose. The MATLAB simulator was modified to accommodate a direct link to the GPU kernel so that pulse sequences could be designed with ease within the higher language environment while computationally intense algorithms could be executed in parallel within the GPU environment. A speedup of about 228 times when compared to serially executed C-code on the CPU was observed. The speedup was between 31 to 115 times when compared to the OpenMP parallel executed C-code on the CPU, depending on the number of threads used in multithreading (2 to 8 threads). Moreover, in multi-node, multi-GPU systems, MRISIMUL demonstrated an almost linear scalable performance along with the increasing number of the available GPU cards. Last, a detailed 3D model of the anatomy of the human heart and torso was developed based on the segmentation of high resolution tomographic images of the Visible Human Project. Motion, strain and respiration were implemented as well. The development of three different motion models, including a respiratory motion model, a heart motion model and a simple flow model, along with their integration with the simulator platform running on CUDA, allowed for motion simulations. The web access version of the simulator supports motion simulations (http://mri.dib.uth.gr).
The MRI simulator now allows for complex analysis of pulse sequences and protocols, which will advance our understanding and aid in the dissemination of cardiovascular MRI in a cost effective manner. We expect that the simulation platform will become an indispensable tool for medical imaging centers. We also expect that the proposed enhanced simulation platform will change the way by which MRI protocols are optimized. Moreover, it should change how MRI physics are taught to non-technical personnel (i.e. physicians, technologists). We expect that it will allow for a shift from classroom-teaching to a laboratory-style approach or a web-based approach without the need for access to an MRI scanner. The hands-on experience and 3D visualization that this simulation platform will provide will allow for easier understanding of basic MRI physics concepts and for direct feedback with respect to pulse sequence and protocol selection.