WordCloud
A wordcloud created from the words in abstracts of our research papers

In our research, we develop innovative ways to acquire, represent, and process heterogeneous information about human health with the purpose of achieving a better personalised and predictive healthcare, and so consequently enhancing health. Our research activities, which are co-funded by the institutional support, various national grants (e.g., MZ IGA – NT 13326, GAČR 407/12/1525, GAČR 406/09/0150, CZ.1.05/3.1.00/14.0298), European grants (e.g., VPHOP – FP7-ICT-223865), but also by private health companies, can be divided into the following directions.

Digital Patient

As creating an accurate patient-specific virtual model is currently not feasible (due to technological limitations, etical issues, cost and human effort involved), a more efficient solution lies in constructing an atlas – a generic model (created from medical images and other data available for a cadaver) that is then deformed to fit a particular patient using specific individual features captured from data related to this patient, such as height, weight, x-ray images, and enhanced by various heteregenous data and meta-data available  for the particular patient (e.g.,µCT images of femur bone, genetics information, etc.). The resulting digital model of the patient is then subject to various simulations, e.g., deformation of muscles in the dependence on the movement of bones, changing the shape of heart by the ECG of the patient, the flow of blood in the blood vessels by reference to the measured blood pressure of the patient, etc. All this require automatic or semi-automatic segmentation of medical images (using probabilistic approaches, neural networks, or other approaches of machine learning), followed by the extraction of 3D surface or volume meshes of organs (e.g., skin, muscles, bones, blood vessels) or tissues (e.g., bone structure). Rigid or non-rigid transformations of these meshes have to be performed in order to register their precise location and orientation in the body, or in order to detect changes in their shape in time caused by external or internal forces. These transformations are based on PCA and other methods for data analysis.The virtual model of the patient, whether as a whole or just part of it, is then visualized by computer graphics techniques.

Our most important projects concerning the development of digital patient

DP creation process
DP creation process

Population Modelling and Medical Information Systems

The data for a particular patient, for example, segmented medical images, 3D meshes of organs or tissues, EEG, ECG, and other bio-signals, are analyzed and parameterized (e.g., by their size, orientation, ratio of the lengths of their sides, the predominant frequencies) and along with other biometric data (e.g., temperature, blood pressure) and additional information from the medical records, DICOM headers, etc., are saved to the database (or data warehouse) designed on the basis of given ontology. Queries to this database then provides, for the given patient, the reconstruction of their data from the data of other patients with similar related parameters (e.g., for the patient, there is no 3D model of their femur, but its length is known, so it is possible to retrieve the most similar model available in the database of the patients with the similar length). One may also use queries that, based on the symptoms of disease of a particular patient, return comprehensive data of patients in the database with the most similar symptoms, including also information about their confirmed diagnosis and treatments, and thus help the doctors with the diagnosis and prediction of the further development of the disease of this patient depending on the chosen treatment. All this requires Fourier or wavelet transformations of signals, their filtering, e.g., by Gabor filter, exploitation of neural networks, graph databases, some natural language processing, anonymization and encryption processes (to ensure privacy of the data), etc.

Our most important projects concerning the population modelling

Neuroinformatics

EEG/ERP measurement
EEG/ERP measurement

Neuroinformatics deals with a measurement of electrical activity in the human brain in various situations, especially those that are stressful and repetitive, and whith an evaluation of such data (typically, EEG/ERP = brain event related potentials) in order to assist with diagnosis of neurological disorders or to enable control of a machine by a thought (i.e., brain-computer interface, BCI), which is very useful particularly for improving the treatment of the disabled patients. The evaluation requires an explotation of methods for 1D signal pattern recognition and classification based on, e.g., Fourier transformation, probabilistic approaches and neural networks. All the data, after being evaluated and annotated, are published and shared in a data store according to international standards. For more information, see this page and, please, see also web pages at: http://neuroinformatics.kiv.zcu.cz/

Our most important projects in neuroinformatics

Diabetes & The Modelling of Glucose Dynamics

Blood Glucose Chart
Blood Glucose Chart

Diabetes is the 8th most common cause of death. Technology plays a vital role in managing diabetes and educating patient about importance of the treatment. The patient must be able to manage his blood glucose level. Due to an important discomfort to the patient, blood glucose level is measured sporadically, while subcutaneous tissue glucose level is measured continuously. Measuring glucose level in subcutaneous tissue is minimally invasive and thus preferred, but this level differs from blood glucose level. This highly specialized activity focuses on the development of a method for a more accurate real-time prediction of glucose concentration in blood based on the known glucose concentration in subcutaneous tissue (obtained from a CGM system). The glucose dynamics mathematical model, we have developed, is composed of two inter-dependent differential equations whose solving requires numerical methods, genetic algorithms, and/or parallel processing. For more information, see this page and see also: http://diabetes.zcu.cz/

Diagnostic Approaches for Otorhinolaryngology

ORL data
ORL data

The aim of this highly specialized activity is the development of methods for early diagnosis of serious diseases (including cancer) of  larynx and inner ear. These methods combine analysis of acoustic signal and analysis of images from a high-frequency camera, which requires exploitation of various signal and image filtering techniques, algorithms of pattern recognition, etc.