|
|
Optimizing Image Reconstruction in Electrical Impedance Tomography
Dušek, Jan ; Brančík, Lubomír (oponent) ; Sikora, inž Jan (oponent) ; Mikulka, Jan (vedoucí práce)
The thesis presents, analyzes, and discusses the optimization of algorithms that reconstruct images of unknown specific conductivity from data acquired via electrical impedance tomography. In this context, the author provides a brief mathematical description of the forward and inverse tasks solved by using diverse approaches, characterizes relevant measurement techniques and data acquisition procedures, and discusses available numerical tools. Procedurally, the initial working stages involved analyzing the methods for optimizing those parameters of the model that influence the reconstruction accuracy; demonstrating approaches to the parallel processing of the algorithms; and outlining a survey of available instruments to acquire the tomographic data. The obtained knowledge then yielded a process for optimizing the parameters of the mathematical model, thus allowing the model to be designed precisely, based on the measured data; such a precondition eventually reduced the uncertainty in reconstructing the specific conductivity distribution. When forming the numerical model, the author investigated the possibilities and overall impact of combining the open and closed domains with various regularization methods and mesh element scales, considering both the character of the conductivity reconstruction error and the computational intensity. A complementary task resolved within the broader scheme outlined above lay in parallelizing the reconstruction subalgorithms by using a multi-core graphics card. The results of the thesis are directly reflected in a reduced reconstruction uncertainty (through an optimization of the initial conductivity value, placement of the electrodes, and shape deformation of the domains) and accelerated computation via parallelized algorithms. The actual research benefited from an in-house designed automated tomography unit.
|
|
|
Optimizing Image Reconstruction in Electrical Impedance Tomography
Dušek, Jan ; Brančík, Lubomír (oponent) ; Sikora, inž Jan (oponent) ; Mikulka, Jan (vedoucí práce)
The thesis presents, analyzes, and discusses the optimization of algorithms that reconstruct images of unknown specific conductivity from data acquired via electrical impedance tomography. In this context, the author provides a brief mathematical description of the forward and inverse tasks solved by using diverse approaches, characterizes relevant measurement techniques and data acquisition procedures, and discusses available numerical tools. Procedurally, the initial working stages involved analyzing the methods for optimizing those parameters of the model that influence the reconstruction accuracy; demonstrating approaches to the parallel processing of the algorithms; and outlining a survey of available instruments to acquire the tomographic data. The obtained knowledge then yielded a process for optimizing the parameters of the mathematical model, thus allowing the model to be designed precisely, based on the measured data; such a precondition eventually reduced the uncertainty in reconstructing the specific conductivity distribution. When forming the numerical model, the author investigated the possibilities and overall impact of combining the open and closed domains with various regularization methods and mesh element scales, considering both the character of the conductivity reconstruction error and the computational intensity. A complementary task resolved within the broader scheme outlined above lay in parallelizing the reconstruction subalgorithms by using a multi-core graphics card. The results of the thesis are directly reflected in a reduced reconstruction uncertainty (through an optimization of the initial conductivity value, placement of the electrodes, and shape deformation of the domains) and accelerated computation via parallelized algorithms. The actual research benefited from an in-house designed automated tomography unit.
|
host ::
RSS.