prof. Ing. Valentine Provazník, Ph.D.

2. round (applications submitted from 01.07.2019 to 31.07.2019)

1. Advanced algorithms for monitoring human activity using mobile sensors

   The theme of this dissertation is aimed on monitoring and evaluation of activities performed by individuals using sensors commonly available in mobile devices (accelerometer, GPS, microphone, heartbeat sensor) and it can be divided into two parts. The goal of the first part is to analyze possibilities of mobile devices available on consumer market, become familiar with the types of sensor data and assess their potential. The goal of the second part is to design advanced algorithms for processing of captured data to identify different types of performed activities (sitting, standing, walk, run, fall). Applicants are expected to be familiar with Matlab programming and have an overview in the area of processing and analysis of one-dimensional digital signals. PhD students will completed six-month internships at attractive partner universities abroad.

   Tutor: Vítek Martin, Ing., Ph.D.

2. Advanced methods for cardiac arrhythmias classification

   The theme of the dissertation is aimed on design and development of new sophisticated methods for parametrization and classification of ECG records in order to timely diagnose cardiac arrhythmias. It will be mainly focused on the automatic detection of arrhythmias, which are poorly recognized by common ECG criteria and are often confused with other types of arrhythmias. Among such arrhythmias, they are atrial fibrillation and atrial flutter, which may have paroxysmal character and, consequently, may not be detected via existing methods. Multidimensional data analysis, time-frequency or nonlinear analysis are expected to be useful for addressing the topic. Automatic recognition of different types of cardiac arrhythmia implies the use of advanced machine learning methods, including state-of-the-art deep learning approaches yielding excellent results in image data classification. To develop and test the algorithms, publicly available data as well as data collected in frame of research projects at DBME will be used. The work will be a follow-up to the conducting research. PhD students will completed six-month internships at attractive partner universities abroad. UBMI provides doctoral students with a stipend and/or a part-time contract beyond the state stipend when joining a grant project or engaging in teaching.

   Tutor: Filipenská Marina, Ing., Ph.D.

3. Advanced methods for ECG signal quality estimation

   The topic of dissertation thesis is focused on continuous quality monitoring in long-term ECG records. The first part is to evaluate the quality of ECG signal recorded from different locations on the body using mobile recorder and possibilities of simultaneously recorded physical activity by gyroscope. The second part is the design of advanced algorithms for continuous and real-time estimation of the ECG quality and subsequent identification of the section of the same quality. Applicants are expected to programming skills in Matlab and base knowledge of the processing and analysis of 1D signal. PhD students will complete a six-month internships at attractive partner universities abroad. UBMI provides doctoral students with a stipend and/or a part-time contract beyond the state stipend when joining a grant project or engaging in teaching.

   Tutor: Smital Lukáš, Ing., Ph.D.

4. Deep learning methods for processing of image sequences

   This topic deals with the application of deep artificial neural networks in the area of video processing and analysis. The primary focus will be on the detection, segmentation and tracking of people in video sequences with the focus on face segmentation. Two main applications will be considered - biometrics (e.g. person recognition, sex and age recognition) and biomedicine (e.g. heart rate or respiratory detection, facial expression recognition). It is expected that deeper study will be needed in the following areas: convolution neural networks, transfer learning for application in other tasks, progressive learning to solve new and complex problems and application of recurrent neural networks for segmentation and tracking objects in the image data. PhD students will complete a six-month internship at attractive partner universities abroad. UBMI provides doctoral students with a stipend and/or a part-time contract beyond the state stipend when joining a grant project or engaging in teaching.

   Tutor: Kolář Radim, doc. Ing., Ph.D.

5. Genetic variation in cardiomyopathy and coronary artery disease

   Patients suffering cardiovascular diseases such as cardiomyopathy and coronary artery disease tend to cluster in families due to underlying monogenic or polygenic genetic architectures. The main aim of the project is search for genetic variation in these diseases in order to find causative genes and susceptibility loci. Distribution of the allele frequencies of the selected set of loci in a sample population will be analyzed and modelled. The study will be extended to identify loci that implicate pathways in blood vessel morphogenesis and inflammation related to the diseases. Data from 1000 Genomes Project and from CARDIoGRAMplusC4D Consortium project will be used to conduct large genome-wide bioinformatics analysis. There will opportunities to develop and apply research methodologies in statistical genetics and bioinformatics, develop skills in programming in high-level analysis packages, and develop skills in high-performance computing. PhD students will complete a six-month internship at attractive partner universities abroad. UBMI provides doctoral students with a stipend and/or a part-time contract beyond the state stipend when joining a grant project or engaging in teaching.

   Tutor: Provazník Valentine, prof. Ing., Ph.D.

6. Medical image segmentation using deep learning techniques

   The theme of this thesis is aimed on medical image segmentation and classification using deep learning methods. The first aim of this thesis is to improve on actual methods for segmentation of 2D medical images. In next step these methods will be adapted for segmentation of 3D volume images, especially images from microCT system. The classification of images using deep learning methods will be also part of this thesis. Machine learning methods, especially neural networks, which represents new and perspective algorithms of image processing, will be used for the solution of this thesis. The main aim of this thesis is extend possibilities of automatic processing and classification of large volume of data like images from CT scanners. PhD students will complete a six-month internship at attractive partner universities abroad.

   Tutor: Harabiš Vratislav, Ing., Ph.D.

7. New algorithms and data structures supporting handling of multidimensional data representing interrelated MR spectra

   The work will be based on the state-of-the-art techniques of spectroscopic signal analysis as implemented in software jMRUI, licensed to about 3000 teams and practically used by many MR sites worldwide for the quantification of metabolite concentrations from in vivo MR spectra. The weak point of the existing solution is treating signals as a one-dimensional array of signals of the same type. This approach is insufficient for proper handling multidimensional MR spectroscopic signals, undersampled spectroscopic imaging, series of diffusion-weighted spectroscopic signals etc. It is also impossible to display such data appropriately. This work aims primarily at the development of a new information infrastructure that will be able to support these (and maybe other) techniques and submit such data to specialized algorithms for processing. Verified, often mathematically sophisticated algorithms will have to be revised and transformed into the new environment. The multidimensional data will require a new graphical user interface and application programming interface, which will provide the functionality for visual inspection of the data analysis steps, valuable in method design and concrete analyses as well, and for independent analysis of specific restricted subsets of the complete data set, ideally utilizing all available parallel-calculation capabilities. The development of prototypes of such function will be part of this project. As necessary side tasks to the main goal, modules for loading data in various formats coming from various MR-scanner vendors will be worked upon. The work will be part of Marie-Curie ITN project INSPiRE-MED and will be perfomed in interaction with the other partners. The student will participate in the workshops and review meetings within this project, he/she will be seconded for 3 months to the University of Bern and the company MRC, and will closely collaborate with a local student ESR5, whose work will benefit from the results described above. The student will be employed for 36 months by ÚPT AV ČR, v. v. i., remunerated in line with the EU rules valid for this project type. The student is required to have appropriate knowledge of signal theory, mathematical calculation and object-oriented programming and experience in programming (Python and Java, or C++, Matlab). Coordination and tutorship from the side of the project will be provided by Ing. Jana Starčuková, Dr.

   Tutor: Jiřík Radovan, doc. Ing., Ph.D.

8. New methods of rational drug design for finding novel HDAC inhibitors as anticancer agents

   Carcinogenesis cannot be explained only by genetic alterations, but also involves epigenetic processes (DNA methylation, histone modifications and non-coding RNA deregulation). Modification of histones by acetylation plays a key role in epigenetic regulation of gene expression and is controlled by the balance between histone deacetylases (HDAC) and histone acetyltransferases (HAT). HDAC inhibitors induce cancer cell cycle arrest, differentiation and cell death, reduce angiogenesis and modulate immune response. HDAC inhibitors seem to be promising anti-cancer drugs particularly in the combination with other anti-cancer drugs and/or radiotherapy. The objective of the current study is to establish structure activity relationships using virtual screening, docking, energetic based pharmacophore modelling, atom based 3D QSAR models and their validation. The outcome of these studies could be further employed for the design of novel HDAC inhibitors for anticancer activity. PhD students will complete a six-month internship at attractive partner universities abroad.

   Tutor: Provazník Valentine, prof. Ing., Ph.D.

1. round (applications submitted from 01.04.2019 to 15.05.2019)

1. Analysis of movement stereotypes in sports performance

   Motion analysis in sports is irreplaceable. A detailed analysis of movement stereotypes leads to improved quality of training plans and improve sports results of training individuals. Analysis can also be used for diagnostic purposes - monitoring faulty movement patterns after injuries in order to clarify procedures for rehabilitation treatments. This work will be focused on monitoring specific movement stereotypes, selection of appropriate parameters and subsequent analysis of data, which will be performed in order to describe the motion stereotypes in sports performance. The study will be carried out in cooperation with the Centre of Sports Activities of BUT. PhD students will complete a six-month internship at attractive partner universities abroad. UBMI provides doctoral students with a stipend and/or a part-time contract beyond the state stipend when joining a grant project or engaging in teaching.

   Tutor: Kolářová Jana, doc. Ing., Ph.D.

2. Efficient computer simulation of MR excitation of brain metabolites based on quantum mechanics

   The work will be based on the current technique of calculation of model spectroscopic signals, as implemented in program NMRScopeB, which is an independent part of software jMRUI, used at many MR sites worldwide for the quantitation of metabolite concentrations from MR spectra. This calculation utilizes the metabolite-molecule spin-systems by a density operator, subjected to evolution by the Schrödinger equation and the Redfield model of relaxation; MR signal separation is achieved for specific sequences by stratification of coherence transfer pathway classes in the k-space. The implementation enables the application of any pulse sequence and observation of spatial and other functional dependences. The aim of this work is to develop a division of the simulation task into calculations performed in parallel on a multiprocessor CPU or GPU controlled by tools of Python language and CUDA environment. This reconstruction will involve the analysis of weak points and the revision of theoretical background from the point of view of numerical mathematics and the ability to reflect unavoidable specific imperfections of each experiment. The new efficient simulation technique will be used in the development of prototypes of 3 applications: 1. Calculation of the dictionary for the application of the fingerprinting principles in MR spectroscopy (such a technique will be developed by the consortium under the INSPiRE-MED project). 2. Embedding of the fast simulation into a pulse sequence parameter optimizer for achieving specific excitation patterns (e.g. discriminatioin of selected metabolites, for multidimensional spatial selectivity, for the compensation of the inhomogeneity of the RF field B1). 3. Embedding in an iterative optimization of model parameters, for the quantification of metabolite concentrations, optionally undersampled, or for the identification of spin system parameters. All results will be integrated into a new version of the jMRUI software that will support spectroscopic metabolite quantitation techniques characterized by higher robustness and ability to process undersampled data. The work will be part of Marie-Curie ITN project INSPiRE-MED. The student will participate in the workshops and review meetings within this project, he/she will be seconded for 3 months to the University of Bern, to the company Bruker PCI, and will visit EPFL Lausanne. He/she will be employed for 36 months by ÚPT AV ČR, v. v. i., remunerated in line with the EU rules. The student is required to have appropriate knowledge of linear algebra, signal theory and magnetic resonance a experience in programming (Python and Java, or C++, Matlab). Coordination and tutorship will be provided by Ing. Zenon Starčuk jr., CSc.

   Tutor: Jiřík Radovan, doc. Ing., Ph.D.

3. Methods for analysis of long-term clinical recordings of intracranial EEG in epileptic patients

   The aim of this work will be development of new methods of determination of EEG features for diagnosis and treatment of epileptic disorders. This mainly consists of description and identification of epileptic sources appropriate for resective or stimulation treatment and determination of areas and methods for prediction and elimination of epileptic seizures. The source will be extensive clinical datasets acquired from deep brain structures of epileptic patients measured with high acquisition dynamics and sampling frequency. These datasets include vigilant and sleep stages, epileptic seizures as well as research protocols. The work will be handled in cooperation with the St. Anne’s University Hospital in Brno.

   Tutor: Jurák Pavel, Ing., CSc.