|
|
Automatic AIF voxel detection in DCE-MRI using machine learning
Frolíková, Štěpánka ; Jiřík, Radovan (referee) ; Vitouš, Jiří (advisor)
Dynamická kontrastní magnetická rezonance (DCE-MRI) je metoda používaná k hodnocení perfuze tkání. Vyžaduje co nejpřesnější identifikaci a detekci arteriální vstupní funkce (AIF). Manuální značení AIF není dostatečně rychlé a může být dost pracné. Obvykle se ale algoritmy automatické detekce nepoužívají. Cílem této práce je přesné určení AIF pomocí natrénovaného modelu využívající strojového učení. Výsledky natrénovaného modelu jsou porovnány s jinými přístupy, jako je k-means shlukování. Data použitá pro trénování a testování modelu jsou jak reálná, tak syntetická. Syntetická data jsou simulována pomocí farmakokinetického modelu DCATH. Shlukovací model je založen na algoritmu k-means, který je optimalizován pro lidská a myší data. Model strojového učení využívá klasifikátor založený na metodě náhodného lesa v kombinaci se shlukováním. Tato práce hodnotí přesnost použitých metod, jejich výhody a nevýhody. Plně automatická detekce AIF by značně ušetřila délku trvání perfúzní analýzy a kvalitu odhadu AIF.
|
|
|
Analysis of time-frequency characteristics of signals
Vitouš, Jiří ; Ředina, Richard (referee) ; Poměnková, Jitka (advisor)
This thesis focuses on time-frequency analysis of discrete signals. The aim of this work is to compare the most well known methods for spectro/scalegram estimation. The two main topics discussed are: The compromise between time and frequency resolution and the effect of noise in input data on the quality of estimated spectrograms. To achieve this a database has been created. This database consists of real and artificial signals on which the analysis can be performed and evaluated. This database is used in created demonstration application. This application is also used in a created laboratory task.
|
|
|
MRI of Rats - Quantification of T1 in Myocardium
Vitouš, Jiří ; Mézl, Martin (referee) ; Jiřík, Radovan (advisor)
This thesis focuses on cardiac imaging and quantification of T1 relaxation time in rat hearts. Its main focus is to investigate available methods for such quantification and their application in the development of quantification tools. The large impact is given to methods of acquisition synchronization, mainly with respect to cardiac motion and breathing using retrospective gating, where the navigator signal is obtained solely from the acquired data, so without any external equipment such as the ECG or respiratory sensors. This paper takes into account situations where steady-state has been reached and also those where it has not, by means of contrast agent injection or by inversion pulses.
|
|
|
Quantification of Relaxation Time T1 in DCE-MRI
Vitouš, Jiří ; Mézl, Martin (referee) ; Jiřík, Radovan (advisor)
This bachelor thesis deals with methods of T1 relaxation time quantification and B1 magnetic field inhomogeneities corrections. For purposes of the required measurements, the phantom was created using a 3D printer. Firstly the thesis presents information about a given theme, then it hands out the results of measurements taken using standard quantification methods. These methods are consequently analyzed and compared. Next part of the bachelor thesis deals with the fast method based on a variable flip angle (VFA), which needs a magnetic field B1 correction maps to give reasonable T1 estimates. For this purpose, two methods of B1 correction are presented. The first method is a double angle method and second is based on mapping magnetic field B1 using a reference T1 map. To decide which T1 map of which method should be used as a reference, a simple simulation is proposed, which describes the impact of unfavorable effects, such as B1 itself, on the T1 estimate of standard methods. In the end, the conclusions are made and the results of VFA methods are compared with standard methods of quantification.
|
|
|
Comparison of Pharmacokinetic Models for Quantification of Blood-Brain-Barrier Opening Induced by Focused Ultrasound
Hývlová, Denisa ; Vitouš, Jiří ; Jiřík, Radovan
This paper focuses on quantification of blood-brainbarrier(BBB) opening induced by focused ultrasound. Dynamiccontrast-enhanced MRI (DCE-MRI) with two different pharmacokineticmodels, 2CX and ETK, is used to evaluate BBB opening.The ETK model is commonly used in BBB-opening studies. Wepropose use of a more advanced model, 2CX, that provides moreaccurate estimation of the BBB permeability. The accuracy of themodels is compared using simulated and real data. The resultsdemonstrate similar performance of the 2CX and ETK models inthe brain region with a highly permeable BBB, but the ETK hasproven unable to correctly detect low level of permeabilization.Thus, the 2CX model is more suitable for the studies of BBBpermeabilitydynamics.
|
|
|
Unfolded Low-rank + Sparse Reconstruction for MRI
Mokrý, O. ; Vitouš, Jiří
We apply the methodology of deep unfolding on the problem of reconstruction of DCE-MRI data. The problem is formulated as a convex optimization problem, solvable via the primal-dual splitting algorithm. The unfolding allows for optimal hyperparameter selection for the model. We examine two approaches - with the parameters shared across the layers/iterations, and an adaptive version where the parameters can differ. The results demonstrate that the more complex model can better adapt to the data.
|
|
|
Correction of gradient pulse shape distortions in radial MRI
Vitouš, Jiří
This paper focuses on the optimization of gradient-pulse shapes in MRI measurement. The main topic investigated in this paper is optimization of slice-selective gradient, where imperfections may produce phase distortion in the resulting image and also signal loss in the acquired signal. A method for correction is proposed based on the Nelder-Mead algorithm followed by coordinate ascent search in the neighborhood of the found solution. The adjustment is evaluated using a simple Fast low angle shot (FLASH) sequence with radial readout. The results show a significant improvement in the Free induction decay (FID) signal magnitude, echo stability, and an improvement in the homogeneity of image phase.
|
|
|
Unfolded Low-rank + Sparse Reconstruction for MRI
Mokrý, Ondřej ; Vitouš, Jiří
We apply the methodology of deep unfolding on the problem of reconstruction of DCE-MRI data. The problem is formulated as a convex optimization problem, solvable via the primal–dual splitting algorithm. The unfolding allows for optimal hyperparameter selection for the model. We examine two approaches – with the parameters shared across the layers/iterations, and an adaptive version where the parameters can differ. The results demonstrate that the more complex model can better adapt to the data.
|
|
|
Correction of gradient pulse shape distortions in radial MRI
Vitouš, Jiří
This paper focuses on the optimization of gradient-pulse shapes in MRI measurement. The main topic investigated in this paper is optimization of slice-selective gradient, where imperfections may produce phase distortion in the resulting image and also signal loss in the acquired signal. A method for correction is proposed based on the Nelder-Mead algorithm followed by greedy search in the neighborhood of the found solution. The adjustment is evaluated using a simple Fast low angle shot (FLASH) sequence with radial readout. The results show a significant improvement in the Free induction decay (FID) signal magnitude, echo stability, and an improvement in the homogeneity of image phase.
|
|
|
Analysis of time-frequency characteristics of signals
Vitouš, Jiří ; Ředina, Richard (referee) ; Poměnková, Jitka (advisor)
This thesis focuses on time-frequency analysis of discrete signals. The aim of this work is to compare the most well known methods for spectro/scalegram estimation. The two main topics discussed are: The compromise between time and frequency resolution and the effect of noise in input data on the quality of estimated spectrograms. To achieve this a database has been created. This database consists of real and artificial signals on which the analysis can be performed and evaluated. This database is used in created demonstration application. This application is also used in a created laboratory task.
|
guest ::
