|
|
Segmentation of Electrocardiographic Signals Using Deep Learning Methods
Hejč, Jakub ; Černý, Martin (referee) ; Halámek, Josef (referee) ; Kolářová, Jana (advisor)
The thesis deals with deep learning methods for the segmentation of surface and intracardiac electrocardiographic recording with focus on atrial activity. The theoretical part introduces current segmentation aproaches of electrocardiographic signals. Issues related to the development of deep learning models in context of standard ECG databases were also discussed. We proposed a pipeling for processing multimodal electrophysiology data from interventional procedures in order to build reliable training datasets. A deep model for segmentation of intracardiac recordings based on a modified residual architecture was proposed. A series of experiments was conducted to evaluate the effect of both model and dataset properties on segmentation quality. The annotation methodology of recordings with atrial fibrillation proved to be a crucial factor. Properties of loss function and type of data augmentation were revealed as secondary important parameters. A novel P wave segmentation method for incomplete references was proposed in the thesis. The approach was inspired by the deep contrast learning. It was modified to distinguish local segments of signals at different levels of abstraction of the extracted feature maps. Results were analyzed using standard quality metrics and post-hoc visual analysis. In some cases, a statistical comparison of experiments for different settings was performed. The results of the work showed that it is possible to use intracardiac signals for embedding a vector representation of local atrial activation into deep models.
|
|
|
Ischemic thrombus analysis in multiphasic brain stroke CT data
Mikešová, Tereza ; Holeček, Tomáš (referee) ; Jakubíček, Roman (advisor)
This master thesis deals with analysis of ischemic thrombus in brain CT scans. In theoretical part, a review of methods, especially thrombus segmentation, is developed. Furthermore, the anatomy of cerebral arteries and acute ischemic stroke is summarized. Selected methods from the field of image processing are briefly described. The practical part results in a comparison of thrombus segmentation methods. The segmentation itself was preceded by data preprocessing, which is described in the theses, and the creation of a manual annotation database. The best implemented method was found to be the adaptive thresholding method, which achieved a Dice score of 0,4555. By combining the methods appropriately, a final Dice score of 0,5145 was achieved. Thrombus parameters were then calculated from the segmented volumes. The median intensity value was 51,55~HU, the median length was 15,16 mm, and the median volume was determined to be 65,34 mm3. Subsequent correlation analysis showed no significant relationship between the derived parameters.
|
|
|
Deep Learning in Historical Geography
Vynikal, Jakub ; Pacina, Jan
In relation to the rapid development of artificial intelligence, the possibilities of automatic processing of spatial data are increasing. Scanned topographical maps are a valued source of historical information. Neural networks allow us to extract information quickly and efficiently from such data, eliminating the difficult and repetitive work that would otherwise have to be done by a human. The article presents two case studies exploring the possibilities of using deep learning in historical geography. The first one is concerned with detecting and extracting swamps from topographic maps, while the second one attempts to automatically vectorize contours from the State Map 1 : 5 000
|
|
| |
|
|
Methods for biomedical image signal segmentation
Krumpholc, Lukáš ; Šmirg, Ondřej (referee) ; Přinosil, Jiří (advisor)
This work deals with methods of segmentation of biomedical image signals. It describes, sums up and compares representative methods of digital image processing. Segmentation based on parametric representation is one of the mentioned methods. So as the basic parameter can be chosen for example luminance and the final binary image is obtained by thresholding. Next described method is segmentation based on edge representation. This method can be divided into edge detection by the help of edge detectors and of Hough transformation. Edge detectors work with the first and second derivation. The following method is region-based segmentation, which can be used for a image with noise. This category can be divided into three parts. The first one is segmentation via splitting and merging regions, when the image is split and the created regions are tested on a defined condition. If the condition is satisfied, the region merges and doesn’t continue splitting. The second one is region growing segmentation, when adjacent pixels with a similar intensity of luminance are grouped together and create a segmentated region. Third one is watershed segmentation algorithm based on the idea of water diffusion on uneven surface. The last group of methods is segmentation via flexible and active contours. Here is described an active shape model proceeding from a possibility to deform models so that they match with sample shapes. Next I also describe method Snakes, where occurs gradual contour shaping up to the edge of the object in the image. For the final editing is used mathematical morphology of segmentated images. I aimed to meet methods of image signals segmentation, to cover the chosen methods as a script in programming language Matlab and to check their properties on images.
|
|
|
Detection of face parts in the thermographic spectrum
Šujan, Miroslav ; Smékal, Zdeněk (referee) ; Mekyska, Jiří (advisor)
Master´s thesis deals with current problems of face detection and its parts in the infrared thermographic spectrum. Most previously published literature deals with the detection in the visible spectrum, making the thermographic detection range an interesting alternative. The work deals with the processing of image signals, images and faces in thermographic spectrum, selected methods of face detection and its parts and also deals with practical system design for detecting facial parts in this spectrum and its subsequent testing.
|
|
|
Detection of Boxes in Image
Soroka, Matej ; Bartl, Vojtěch (referee) ; Herout, Adam (advisor)
The aim of this work is to experiment and evaluate different approaches of computer vision with the aim of automatic detection of boxes-blocks in the image, for this purpose, approaches based on neural networks were used in the solution. Experiments were performed with classification using our own data set, classification using our own convolutional neural network, detection using a window, YOLO detector and in the last part a proposal for improvement using U-net and MirrorNet networks.
|
|
|
Knowledge Discovery from Time Series
Krutý, Peter ; Burget, Radek (referee) ; Bartík, Vladimír (advisor)
This thesis is focused on the field of knowledge discovery from data, specifically from time series. Main objective is to research Python programming language support in this area and then design and implement an application that will allow to demonstrate and compare selected methods. Methods are demonstrated in experiments using appropriate data set. The output of the thesis is a comparison of methods for specific tasks and the application implementing selected methods.
|
|
|
Scene Analysis Based on the 2D Images
Hejtmánek, Martin ; Drahanský, Martin (referee) ; Orság, Filip (advisor)
This thesis deals with an object surface analysis in a simple scene represented by two-dimensional raster image. It summarizes the most common methods used within this branch of information technology and explains both their advantages and drawbacks. It introduces the design of an surface profile analysis algorithm based on the lighting analysis using knowledge and experiences from previous work. It contains a detailed description of the implemented algorithm and discusses the experimental results. It also brings up options for the possible enhancement of the projected algorithm.
|
|
| |
guest ::
