|
|
Estimation of bone mineral density of cancellous vertebral bone in multi-energy CT data
Líška, Martin ; Jakubíček, Roman (referee) ; Chmelík, Jiří (advisor)
The principle of the BMD estimation method presented in this thesis consists in the tomographic scanning of the axial skeleton by a CT system with two different energies. The BMD estimation method was applied to acquisitions scanned by CT system IQon Spectral CT (Philips) on seven patients, two men and five women, in the lumbo-sacral region. For the functionality of the method, it is necessary to know the standardized amounts of selected elemental components contained in a given tissue, specifically in the cancellous bone of the vertebra. In the first part, the thesis deals with the theoretical part of solving the estimation of BMD from dual-energy CT data, two equations with several unknowns and their modification. The practical part deals with the program solution of the method of calculating the estimation of bone minerals in dual-energy CT data. The outputs of the presented BMD estimation method were processed and statistically compared with the other two phantom-less BMD estimation methods. The functionality of the method and statistical processing were solved in MATLAB and STATISTICA softwares.
|
|
|
Analýza vzťahov medzi radiomickými priznakmi heterogenity trombu v akútnych ischemických mozgových príhodách
Nemčeková, Petra ; Škrváň, Adam ; Henk, Marquering ; Chmelík, Jiří ; Jakubíček, Roman
Cievne mozgové príhody sú jedným z najznámejších patológií mozgu. Prvotnou diagnostickou metódou je použitie počítačovej tomografie (CT). Avšak pre správne určenie liečby by bolo potrebné vedieť bližšie charakteristiky trombu, na základe ktorých by bol lekár schopný usúdiť najmenej riskantnú cestu pre pacienta. Táto štúdia sa zameriava na analýzu heterogenity trombov na CT snímkach u pacientov s ischemickou mozgovou príhodou. Na základe extrahovaných radiomických príznakov získaných z reprezentatívnych masiek trombov bolo získané rozmiestnenie voxelov jednotlivých trombov v novom parametrickom priestore. To bolo následne podrobené vizualizačným technikám tSNE a UMAP. Na základe vyhodnotenia morfologickej štruktúry jednotlivých vytvorených zhlukov u pacientov by bolo možné určiť počet častí trombu s rôznym zložením, na základe čoho by lekár mohol byť schopný predikovať záťaž pre pacienta pri trombektómii, ako napríklad pomocou počtu pokusov potrebných na spriechodnenie cievy.
|
|
|
Možnosti přístupu k obrazovým datům v rámci projektů ÚBMI ve spolupráci s klinickými pracovišti
Jakubíček, Roman ; Nemčeková, Petra ; Ouředníček, Petr ; Chmelík, Jiří
Tento článek zkoumá výzvy a možnosti spojené se zpracováním a sdílením obrazových dat v kontextu biomedicíny a počítačem podporované diagnostiky. S rostoucím výpočetním výkonem a využitím strojového učení se metody analýzy obrazů stávají stále efektivnějšími, ale potýkají se s problémy dostupnosti dat a obtížnou interpretovatelností. Autoři diskutují legislativní a etické aspekty ochrany osobních údajů a upozorňují na význam spolupráce mezi akademickými institucemi a klinickými pracovišti. Článek také prezentuje dva aktuální výzkumné projekty ÚBMI v oblasti analýzy obrazů, tj. analýza trombu v CT mozku a kardiovaskulární zobrazování magnetickou rezonancí, ve kterých se aktuálně využívají pokročilé algoritmy strojového učení. Spolupráce mezi ÚBMI a klinickými pracovišti přináší nové možnosti pro zlepšení diagnostiky a léčby pacientů.
|
|
| |
|
|
Automatic diagnosis of the 12-lead ECG using deep learning
Blaude, Ondřej ; Chmelík, Jiří (referee) ; Provazník, Valentine (advisor)
The aim of this diploma thesis is to investigate the problematics of automatic ECG diagnostics, namely on twelve-lead recordings. This problem is solved by standard methods such as random forest, artificial neural networks or K-nearest neighbors. However, thanks to its ability to independently extract symptoms, deep learning methods are also popular. All these methods are described in the theoretical part. In the practical part, deep learning models were designed, functionality support was verified using data from the PhysioNet database. Two pilot models were created and subsequently optimized. From the entire parameter optimization procedure, three models are available, of which the best accuracy achieves an F1 score of 87.35% and 83.7%, and the second best achieves an F1 score of 77.74% and an accuracy of 84.53%. The results achieved are discussed and compared with those of similar publications.
|
|
|
Deep-learning based model implementation for pathological tissue characterization in brain MR images
Malík, Michael ; Nemčeková, Petra (referee) ; Chmelík, Jiří (advisor)
This bachelor thesis focuses on the issue of image segmentation by using a deep learning model. The theoretical part describes the anatomy and selected pathology of brain. The thesis also deals with the construction of MR device and creation of an MR image. In the closing section of theoretical part, the main focus is on describing the possibilities of image segmentation with the use of deep learning architectures and selected publicly available dataset. The aim of the practical part is to put the mentioned dataset and pre-processed data to the test and acquire results of image segmentations of individual patients from attached model of neural network. In conclusion, the achieved results are appropriately discussed.
|
|
|
Implementation of a deep learning model for vertebral segmentation in CT data
Blažková, Lenka ; Chmelík, Jiří (referee) ; Nohel, Michal (advisor)
This bachelor’s thesis deals with the problem of vertebrae segmentation in CT data with the use of deep learning. Firstly, there is a theoretical review focused on the anatomy and the pathologies of the spine and the vertebrae, the CT systems, and the deep learning models for vertebrae segmentation in 3D data. The following section contains a more detailed description of the chosen model. The fifth section describes the implementation of the chosen model and the proposed modification, together with the results on the relevant database. In the end, the model with the modification is used on the clinical data provided by the supervisor and its evaluation is described.
|
|
|
Analysis of diagnostic parameters from 4D CINE MRI data
Panáček, Oldřich ; Harabiš, Vratislav (referee) ; Chmelík, Jiří (advisor)
The thesis is focused on calculation and analysis of heart function parameters which could be calculated from image data obtained by magnetic resonance. Manually annotated image data in short axis of heart were used for calculation. Volumetric parameters were calculated directly from image data by counting voxels which were parts of specific heart structure and then multiplied by voxel volume. Modified radial method was used to compute functional parameters whereas myocardial contractility was calculated by using segmental strain analysis. Obtained results were given into table and visualised in box– plots. Correlation analysis of parameters was also performed and decision tree classifier was used to test discrimination capacity of estimated parameters.
|
|
|
Deep-learning based segmentation of pathological tissue in brain MR images
Nantl, Ondřej ; Kolář, Radim (referee) ; Chmelík, Jiří (advisor)
This diploma thesis deals with the topic of segmentation of ischemic tissue in T1 weighted MRI image data using deep learning methods. The theoretical part deals with the anatomy of brain, brain imaging using MRI, available datasets for automatic segmentation of pathological brain tissue and automatic deep learning methods for segmentation of ischemic brain tissue. In the practical part the used dataset and its preprocessing, as well as the proposed deep learning methods (U-Net) and their training, are described. The models were implemented using Python. Finally, the results of the models are presented and discussed.
|
|
|
Advanced registration of image sequences from video-ophthalmoscope
Dufková, Barbora ; Chmelík, Jiří (referee) ; Kolář, Radim (advisor)
This master's thesis deals with the issue of registration of ophthalmic video sequences. It describes basic geometric transformations that can be used for registration. The basic methods of image registration are also presented, from which the most suitable variant for this application is selected. This is then implemented using a script created in the MATLAB environment. The proposed method is further evaluated objectively using the brightness profile method, using mutual information and correlation, and using retinal vessel skeleton. The effect of polynomial transformation on registration and possible optimizations of the algorithm are discussed.
|
guest ::
