MASTER THESIS TOPICS Academic year 2019-2020 MEDISIP 2 MEDISIP 3 - - PowerPoint PPT Presentation

MASTER THESIS TOPICS

Academic year 2019-2020

DEPARTMENT OF ELECTRONICS AND INFORMATION SYSTEMS (ELIS) MEDICAL IMAGE AND SIGNAL PROCESSING (MEDISIP)

MEDISIP

2

MEDISIP

3

RESEARCH GOALS OF MEDISIP

• Make medical imaging more quantitative
• Improve acquisitions/reconstructions
• i. Reduce imaging time
• ii. Improve spatial resolution
• Solve artefacts in multimodal integration
• Additional information from multimodal data
• Application fields: small animal and neuroimaging

4

RESEARCH ACTIVITIES @ MEDISIP

5

6

RESEARCH ACTIVITIES @ MEDISIP

ONGOING PHD PROJECTS

• Radiomics-machine learning-brain tumors (partner nuclear medicine/radiology)
• PET imaging in plants (partner Bioengineering)
• PET-MRI novel isotopes (partner KULeuven)
• Dosimetry in radionuclide therapy (Lutetium, partner Bordet)
• High resolution detectors for Total body PET
• Monolithic Time-of-flight detectors for PET

Collaborations

• EEG/Epilepsy with Neurology dept
• Intraoperative PET/CT lumpectomy margin assessment (R. Van den Broucke)

7

IMAGING

8

F-18 LABELING OF MICROSPHERES TO ENABLE INTERVENTIONAL PET FOR MINIMALLY INVASIVE LIVER RADIO-EMBOLISATION

Supervisor: Marek Beliš, Ken Kersemans (UZ Gent) Promotors: prof. Stefaan Vandenberghe, prof. Christian Vanhove Background Targeted radionuclide therapy (TRT) is an established cancer treatment modality. It relies

• n cancer specific agents that are labeled with radionuclides for internal radiotherapy. By

the use of disease specific carriers linked to radionuclides emitting particle with a short range, a high dose of radiation can be delivered to tumors while sparing the unaffected

• rgans. Imaging the distribution of

these radionuclides is required for individual assessment and planning of TRT. When we would have theranostic F-18 labeled spheres PET imaging could be used to combine diagnostic and therapeutic procedures in one procedure. For this reason we want to study three radiolabelling strategies to introduce PET isotopes (F-18) onto the surface of the microparticles

9

10

Goal

• Investigate the different labeling options.
• Image the labeled microspheres with a high-resolution PET system

(available at Infinity lab). An optional area of research is to investigate with flow simulations the flow

• f the microspheres in a typical hepatic artery and liver.

Tools: Modeling, hotlab, PET … Remark: this project is of direct interest from a pharma company delivering therapeutic microspheres Timeline: literature study, (simulation), labeling, data analysis

More information?! 📪 stefaan.vandenberghe@ugent.be

HIGH SENSITIVITY SPECT USING 12 ROTATING PARALLEL COLLIMATED DETECTORS

Supervisor: Marek Beliš, Dr. Bieke Lamber (UZ Gent) Promotors: prof. Stefaan Vandenberghe, prof. Roel Van Holen Background SPECT is the most frequently used techniques and detects single photon emittors by a mechanical collimator and scintillation detector. The conventional gamma camera, based on a 40-year old design, is composed of 2 large (about 40-50 cm) detector heads equipped with large parallel hole collimators. This limits the sensitivity and spatial resolution of SPECT imaging. To obtain relevant images, relative long acquisition times and/or high doses are required. A totally new design based on 12 detector (CZT) heads has been recently commercialised and first systems are installed at 4 clinical sites (France). Each head has an axial dimension of 35 cm and a smaller axial dimension of about 5 cm. These detectors can be brought very close to any body part of the patient to improve spatial

• resolution. For small objects also a larger sensitivity can be obtained.

11

Goal The aim of this thesis is to characterize in detail how much improvement can be expected from such a design in typical imaging situations Tools: Literature, Monte Carlo simulations, MATLAB, SPECT, … Remark: First 4 systems are installed at sites in France measurements can be performed on these sites Timeline: literature study, simulations, data analysis

12

More information?! 📪 stefaan.vandenberghe@ugent.be

INVESTIGATION OF LYSO BACKGROUND RADIATION IN A TOTAL-BODY PET

Supervisor: Charlotte Thyssen Promotors: prof. Stefaan Vandenberghe, prof. Roel Van Holen Background Positron Emission Tomography (PET) is a molecular imaging modality that uses a radioactive tracer to visualize processes occurring inside the body. However, conventional systems only have a very small length → a lot of the radiation produced inside the patient is lost … For this reason MEDISIP wants to develop a total-body PET with a length of 1 meter → ~20x more radiation is caught!! LYSO, the scintillator crystal of choice, is naturally radioactive → background radiation present during scanning

13

Total-body PET system Conventional PET system

14

More information?! 📪 cathysse.thyssen@ugent.be

Goal

• Mapping out the effect of background radiation in total-body PET
• Monte Carlo simulations of human phantoms with and without

background in total-body PET Software: Gate, XCAT, MATLAB/Python, Root, … Timeline: literature study, Monte Carlo simulations, image reconstruction, data analysis

MEDIUM-SIZE ANIMAL PET SCANNER: INVESTIGATION OF IDEAL SCANNER GEOMETRY

Supervisor: Charlotte Thyssen Promotors: prof. Stefaan Vandenberghe, prof. Roel Van Holen Background Today, rats and mice are mostly used for scientific research, however, translation

• f the obtained results to humans is not straightforward. For this reason there is

an increased interest in larger animals like rabbits. Preclinical imaging modalities for these animals are scarce. The idea is to increase the bore size of the MOLECUBES PET-scanner and to include TOF capabilities.

15

16

Goal

• Comparison of different designs for medium-size animal scanners
• Effect of TOF inclusion in medium size animal scanners
• Comparison of different scintillation crystals to reduce cost

Software: Gate, XCAT, MATLAB/Python, Root, … Timeline: literature study, Monte Carlo simulations, image reconstruction, data analysis

More information?! 📪 cathysse.thyssen@ugent.be

ACCELERATING MONTE CARLO SIMULATIONS FOR MEDICAL SCANNER DATA WITH JULIA

Supervisor: Charlotte Thyssen, Tim Besard Promotors: prof. Bjorn De Sutter, prof. Stefaan Vandenberghe Background Monte Carlo simulations are used for simulation of medical imaging data (to optimize image reconstruction or simulate innovative system designs). The code is based on the computationally intensive Geant 4 package (CERN). Simulation of realistic patient data is a very slow process and needs to be run on multiple CPU or GPU, to

• btain data in an acceptable time frame (days/weeks).

Acceleration of this code would benefit a large community of researchers working on improved medical imaging systems.

17

18

Goal

• Identify the critical parts in the library
• Evaluation of the potential of Julia to make Monte Carlo simulations much more efficient and more easily

accessible Software: Gate, Julia Timeline: literature study, Monte Carlo simulations, analysis of simulation code and optimization using Julia Two different types of simulations will be investigated: the first one relies on voxelized sources for determining patient interactions (e.g., Dosimetry purposes) and the second is the scanner simulation part. To reach these goals, we are looking for students with considerable programming experience and a passion for the latest state-of-the-art programming languages.

More information?! 📪 bjorn.desutter@ugent.be

HIGH-PERFORMANCE YET RAPID IMAGING RECONSTRUCTION WITH JULIA (1 OR 2 STUDENTS)

Supervisor: Charlotte Thyssen, Tim Besard Promotors: prof. Bjorn De Sutter, prof. Stefaan Vandenberghe Background After image acquisition, recorded data are obtained as a list of events or projection data sets. An image reconstruction algorithm uses this output data from the scanner to calculate the 3D image

• f the patient. This step is done in an iterative loop and typically

involves several matrix multiplications resulting in a computationally intensive algorithm. The image reconstruction needs to be run on multiple CPU or GPU to be able to keep it equal to the faster acquisition of the most recent scanners.

19

Reconstruction by back projection

20

Goal

• Migrate the state-of-the-art medical image reconstruction code developed at MEDISIP
• Use Julia to answer the existing open questions

Software: QETIR, Julia Timeline: literature study, image reconstruction, analysis of reconstruction code and optimization using Julia, analysis of a second algorithm algorithm (even-based) and comparison to first Possibility for collaboration with MOLECUBES (UGent Spin-off) To reach these goals, we are looking for students with considerable programming experience and a passion for the latest state-of-the-art programming languages.

More information?! 📪 bjorn.desutter@ugent.be

21

22

23

24

25

26

Deep Learning for Computer-Aided Detection and Diagnosis

• f Breast Cancer

Background

̶ Breast cancer is the second leading cause of cancer-related death among women ̶ Early detection increases the chance of full recovery ̶ Screening mammography is associated with a high risk of false positive testing ̶ Computer-aided detection and diagnosis (CAD) systems: Supervisor: Milan Decuyper Promotor: prof. Roel Van Holen

workload Accuracy +

27

Goal Design and train algorithms for computer-aided detection or diagnosis of abnormalities in mammograms such as calcification and masses. Data: CBIS-DDSM database @ The Cancer Imaging Archive. Software: Python (PyTorch/Tensorflow/Keras/...) Different tasks possible such as:

• Detection of breast cancer
• Segmentation of masses and calcifications
• Diagnosis of masses and calcifications as benign or malignant

Deep Learning for Computer-Aided Lung Nodule Detection

Background

̶ Lung cancer is the leading cause of cancer-related death worldwide ̶ Early detection reduces lung cancer mortality ̶ Manual interpretation of lung CT scans is error-prone and time intensive. ̶ Computer-aided detection and diagnosis (CAD) systems: Supervisor: Milan Decuyper Promotor: prof. Roel Van Holen

workload Accuracy +

29

Goal Design and train algorithms for computer-aided lung nodule annotation in CT scans. Data: LIDC-IDRI, NSCLC-Radiomics and NSCLC-Radiogenomics @ The Cancer Imaging Archive. Different tasks possible such as:

• Lung Nodule Detection
• Lung Cancer Diagnosis, survival prediction, prediction
• f genomic mutations etc.

Software: Python (PyTorch/Tensorflow/Keras/...)

FAST AND EFFICIENT RECONSTRUCTION ALGORITHM FOR MAGNETIC RESONANCE ELECTRICAL PROPERTIES TOMOGRAPHY (MREPT)

Supervisors: Prakash Parappurath Vasudevan Promotors: prof. Roel Van Holen, prof. Wout Joseph Background MREPT is a technique used to obtain the admittivity (both conductivity and permittivity) of tissues ̶ Electrical properties (EP) can be used for Cancer diagnosis, Staging and Grading ̶ EPs are critical in applications utilizing EM stimulation for treatment ̶ Accurate assessment of EPs are necessary for subject Specific Absorption Rate (SAR) measurements

30

B1 Mapping Reconstruction Conductivity/Permittivity Image

Goal

• Improve the existing reconstruction algorithm of MREPT
• Test the algorithm using Electromagnetic (EM) field simulation
• Optimize the algorithm for different measurement set-up
• Investigate different B1 mapping methods and compare their performance

Data: Simulated B1 maps, MRI data of Phantoms and Mouse tumour models Software: MATLAB/Python, Sim4Life (optional)

31

For more information: Prakash.ParappuarthVasudevan@UGent.be

RHENIUM-188 SPECTRA SIMULATION FOR SPECT

Supervisor: Marek Beliš Promotors: prof. Stefaan Vandenberghe Background Rhenium-188 (188Re)

• theranostic agent => β- and γ-emissions
• 155 keV γ-ray (15 %) suitable for SPECT => single-photon emission

computed tomography

• similar to 99mTc
• several high-energy γ-rays in emission spectrum & Bremsstrahlung

may complicate quantitative imaging Collimation is necessary to ensure good reconstruction, therefore parameters

• f the collimator affect the quality of images, but also sensitivity etc.

32

33

More information?! 📪 Marek.Belis@UGent.be

Goal

• Simulation of 188Re spectra and comparison to 99mTc
• Search for improvement by changing the parameters of the

collimator

• Possible upgrade to more radionuclides

Software: Gate, MATLAB, … Timeline: literature study, Monte Carlo simulations, data analysis, 3D-modelling

DEVELOPMENT OF LIGANDS FOR COMPLEXES WITH RHENIUM

Supervisor: Marek Beliš Promotors: prof. Stefaan Vandenberghe Background Rhenium-188 (188Re)

• β- and γ-emissions => theranostic agent (suitable both for therapy and

imaging)

• chemically similar to Tc, but with much more complicated redox

chemistry Stability of the radiopharmaceutical is the key aspect for success of targeted radionuclide therapy (TRNT). Therefore development of ligands stabilizing the Re is necessary.

34

Radionuclide – 188Re, 99mTc Biomolecule Cancer cell

Goal

• Synthetical modification of macrocyclic ligands
• Formation of complexes with cold Re and later with 188Re
• Radiolabelling of biomolecules, stability testing

Fields: Organic synthesis, coordination chemistry Timeline: literature study, synthesis, coordination chemistry, data analysis Cooperation with SCK•CEN (Belgian Nuclear Research Centre)

35

More information?! 📪 Marek.Belis@UGent.be

NEUROENGINEERING

36

CALIBRATING EEG SOURCE IMAGING USING EVOKED RESPONSES

37

Supervisor: Jolan Heyse Promotors: prof. Pieter van Mierlo Background In EEG source imaging, the brain activity underlying the measured EEG is estimated by modelling the spreading of electrical activity in MR-based electromagnetic head models. Despite very accurate models that are available these days, the spatial resolution of EEG source imaging is in the order of cm. New MRI sequences (ultra-short echo time, UTE) could help to further improve the head models by refining the tissue segmentation. Evoked potentials (e.g. finger tapping) can be used to evaluate the performance of ESI with the new head model.

38

Goal

• To evaluate the spatial resolution of existing EEG source imaging

methods and improve it through refinement of the head model. Parametrization of the head model and assessing the spatial resolution of the EEG source imaging will be done using evoked potentials as a ground truth.

• A lot of the work will be practical. The student will obtain his/her
• wn data (MRI and EEG) for further analysis.

Software: MATLAB/Python Timeline: literature study, MRI/EEG experiments, data analysis

More information?! 📪 Pieter.vanMierlo@ugent.be

EEG SIGNALS TREATED AS SOUND

Supervisor: Jolan Heyse Promotors: prof. dr. ir. Pieter van Mierlo, prof. dr. ir. Nilesh Madhu Background Because signal transmission occurs instantaneously in the brain, each EEG electrode measures the sum of the individual activities of different brain regions. Individual contributions of the different sources can be obtained by applying a de- mixing procedure. The EEG signals are further corrupted by different artifacts of environmental (e.g. 50Hz hum from power supplies) and biological (e.g. eye blinks, muscle activity etc.) nature. Similar problems have been well-studied for the multi-microphone recording and processing of audio signals and robust solutions have been developed for these use-cases.

39

40

Goal

• To

use algorithms developed for speech/audio processing (e.g. beamforming and spectral corrections) to get more information from the EEG signals.

• These techniques will be applied to the problem of localizing the

epileptic focus in epilepsy patients. Seizure recordings often involve activity from many brain regions and contain many artifacts because of muscle contraction and movement of the patient. Software: MATLAB/Python Timeline: literature study, algorithm implementation, data analysis

More information?! 📪 Pieter.vanMierlo@ugent.be

EEG SOURCE IMAGING AND FUNCTIONAL CONNECTIVITY ANALYSIS OF MONKEY EEG

Supervisor: Jolan Heyse Promotors: prof. dr. ir. Pieter van Mierlo Background In EEG source imaging, the brain activity underlying the measured EEG signals is

• estimated. Looking at the activity patterns from different brain regions, functional

connectivity methods can be applied to reconstruct the functional network of the brain (i.e. how do the brain regions interact with each other?). Many methods exist for assessing functional connectivity, but they are hard to validate as the ground truth communicating network is rarely known. A dataset of simultaneous recordings with scalp and intracranial EEG with electrodes placed inside a monkey's brain can serve as a validation tool.

41

42

Goal

• To validate different functional connectivity methods, plus the added

value of time lag information, based on the monkey dataset. Software: MATLAB/Python Timeline: literature study, data analysis, functional connectivity evaluation

More information?! 📪 Jolan.Heyse@UGent.be

EEG-NEUROFEEDBACK FOR IMPROVED BCI PERFORMANCE

Supervisor: Jolan Heyse Promotors: prof. Pieter van Mierlo Background Brain computer interfaces (BCI) involve direct communication between the brain and an external device (e.g. a neuro-prosthetic limb). As EEG provides a direct measurement of brain activity, it poses a viable candidate as communicating interface in BCI. However classification of brain signals into the intended tasks is hampered by the complexity and variability of the underlying activity. Neurofeedback uses real-time displays of brain activity to teach self-regulation of brain function and could help to improve BCI performance by teaching the subject to steer brain activity towards the desired classification area.

43

44

Goal

• To establish an EEG-based BCI and evaluate the added value of

neurofeedback on the performance, or to learn new tasks using neurofeedback Software: MATLAB/Python Timeline: literature study, experiment design, neurofeedback and BCI implementation, experiments, data analysis

More information?! 📪 Jolan.Heyse@UGent.be

AGE RELATED CHANGES IN CORTICO-CORTICAL CONNECTIONS IN PHONEME DISCRIMINATION

Supervisor: Jolan Heyse Promotors: prof. dr. ir. Pieter van Mierlo, prof. Miet De Letter Background Phonemes are perceptually distinct units of sound and can be considered fundamental building blocks for speech. Discrimination of these phonemes is important for speech comprehension and has been investigated in an EEG-study conducted at Ghent University. In this study, aging was associated with increased latencies and decreased amplitude with age during phonemic discrimination tasks. However, why this difference was observed is not yet explained.

45

46

Goal

• To investigate the connection between brain regions during phoneme

discrimination tasks. Functional connectivity analysis will be used to reveal information flow in several frequency bands.

• We will investigate these interactions during phoneme discrimination and

study age-related differences. This will shed light on why elderly have more difficulties discriminating phonemes. Software: MATLAB/Python Timeline: literature study, data analysis, clinical interpretation

More information?! 📪 Pieter.vanMierlo@ugent.be

INTERHEMISPHERIC CONNECTIVITY OF SUBCORTICAL NUCLEI DURING WORD TASKS

Supervisor: Jolan Heyse Promotors: prof. dr. ir. Pieter van Mierlo en prof. dr. Patrick Santens Background Deep brain stimulation is an established treatment for patients with Parkinson’s

• disease. Here depth electrodes are bilaterally implanted in the subthalamic

nucleus (STN). In literature it has been shown that the stimulation of the STN has an impact on speech. However, the exact role of the STN during speech and the coupling between the STNs from both hemispheres remains to be elucidated.

47

48

Goal

• To investigate the communication between the left and right STN during

word tasks. We will work with intracranial EEG data from patients with implanted electrodes. The stimulator is only implanted a couple of days after the depth electrodes, which provides us a time frame to measure intracranial EEG activity.

• Several word tasks have been recorded in a number of patients, where

action and non-action words were visually shown to the patients. Software: MATLAB/Python Timeline: literature study, data analysis

More information?! 📪 Pieter.vanMierlo@ugent.be

SPIKE SORTING OF SUBTHALAMIC SINGLE NEURON RECORDINGS IN PARKINSON PATIENTS

Supervisor: Jolan Heyse Promotors: prof. dr. ir. Pieter van Mierlo en prof. dr. Patrick Santens Background Deep brain stimulation of the subthalamic nucleus (STN) is an established treatment to reduce motor tremors in patients with Parkinson’s disease. A depth electrode is implanted into the subthalamic nuclei to stimulate the neurons. First multiple micro-electrodes are inserted into the STN, that are capable to record multiple single neurons. Based on these recordings, the location is chosen to implant the macro-electrode that is used for current stimulation. Because the micro-electrode records the activity of multiple neurons simultaneously, spike sorting algorithms are used to separate the activity of the neurons.

49

50

Goal

• To investigate the micro-electrode recordings in patients with

Parkinson’s disease. This requires the implementation of spike sorting algorithms which allow separating activity from the different recorded neurons. Different algorithms will be implemented and their performance will be assessed.

• Furthermore the relation between the micro-recordings of neuronal

activity and the macro-recordings of local field potentials will be studied. Software: MATLAB/Python Timeline: literature study, data analysis, algorithm evaluation

More information?! 📪 Pieter.vanMierlo@ugent.be

AUTOMATED EPILEPSY DIAGNOSIS FROM ROUTINE EEG USING MACHINE LEARNING

Supervisor: ir. Tom Van Steenkiste, prof. dr. Dirk Deschrijver Promotors: prof. dr. ir. Tom Dhaene, prof. dr. ir. Pieter van Mierlo Background Epilepsy is a neurological disorder that affects approximately 0.5-1% of the world’s population. The most important technique to diagnose epilepsy is electroencephalography (EEG). In the EEG, the occurrence of epileptic spikes, i.e. brief electrical discharges in the brain, are a hallmark to diagnose epilepsy. The occurrence of epileptic spikes differs from patient to patient and even within a patient from time to time. In clinical practice, a routine EEG of 20min duration is recorded to diagnose epilepsy. Unfortunately, many patients with epilepsy do not have frequent spikes; therefore the sensitivity of routine EEG to confirm the diagnosis of epilepsy is only 25-56% and the specificity is 78-98%.

51

52

Goal

• To increase the sensitivity and specificity of routine EEG to

diagnose epilepsy. This will be done by using and developing state-

• f-the-art machine learning techniques to classify routine EEGs

recorded in Ghent and Geneva University Hospital as epileptic or non-epileptic.

• In addition to the detection of epilepsy, classification into subtypes

can be performed. In a first step, classification in temporal vs extra- temporal lobe epilepsy can be done.

• This master thesis is in close collaboration with Epilog, a startup

company specialized in EEG

• analysis. The student has the
• pportunity to do an internship at Epilog before the master thesis.

Software: MATLAB/Python Timeline: literature study, data analysis, machine learning

More information?! 📪 Pieter.vanMierlo@ugent.be

DETECTING THE CAUSE OF DEMENTIA USING EEG MEASUREMENTS AND MACHINE LEARNING

Supervisor: ir. Tom Van Steenkiste, prof. dr. Dirk Deschrijver Promotors: prof. dr. ir. Tom Dhaene and prof. dr. ir. Pieter van Mierlo Background Dementia is a syndrome of several diseases: Alzheimer’s Disease (AD), Frontotemporal lobe degeneration (FTD), creutzfeldt-jakob disease (CJD) or Lewy body disease (LBD). Up to now, there is no medical test to diagnose which disease is causing the dementia. Some pilot studies have indicated that electroencephalography (EEG) could be a useful neuroimaging technique to diagnose the cause of dementia. At the same time, recent advancements in machine learning and deep learning have resulted in powerful analysis techniques for medical time-series data. The application of machine learning to EEG data for detecting the cause of dementia could lead to valuable insights and models and could optimize patient treatment.

53

54

Goal

• To use machine learning to classify EEGs from patients that have

dementia into AD, FTD, CJD and LBD. A post-mortem confirmed database from Antwerp University Hospital is available to address this

• question. The student can explore and develop state-of-the-art machine

learning algorithms for time-series analysis and can develop custom algorithms for EEG data analysis.

• This master thesis is in close collaboration with Epilog, a startup

company specialized in analyzing EEG data. The student has the

• pportunity to do an internship at Epilog before the master thesis.

Software: MATLAB/Python Timeline: literature study, data analysis, machine learning

More information?! 📪 Pieter.vanMierlo@ugent.be

EFFECT OF ANTI-EPILEPTIC DRUGS ON FUNCTIONAL BRAIN CONNECTIONS

Supervisor: Jolan Heyse Promotors: prof. dr. ir. Pieter van Mierlo, dr. Gregor Strobbe Background The first line treatment of epilepsy is antiepileptic drugs (AEDs). In approximately 60-70% of patients AED mono- or polytherapy have the desired outcome, namely the patient is seizure-free. Most of the AEDs go hand in hand with many side-effects such as drowsiness, dizziness, fatigue, nausea and vomiting. In all patients an AED is tested without knowing whether the AED will lead to seizure freedom

• r not. Furthermore, the side effects cannot be predicted.

55

56

Goal

• To assess how antiepileptic drugs affect the functional connectivity

and whether these alterations are indicative for the side-effects of the AEDs.

• Furthermore we will investigate the possibility of predicting who will

be a drug responder (i.e. seizure-free) or not, based on the functional connectivity. Software: MATLAB/Python Timeline: literature study, data analysis, machine learning

More information?! 📪 Pieter.vanMierlo@ugent.be

SMALL ANIMALS

57

USING FUNCTIONAL MRI AND GRAPH THEORY TO INVESTIGATE ABNORMAL FUNCTIONAL BRAIN NETWORKS IN A RAT MODEL OF TEMPORAL LOBE EPILEPSY

Supervisor: Emma Christiaen Promotors: prof. Chris Vanhove, prof. Robrecht Raedt Background

• Epilepsy is a disease characterized by recurrent seizures
• More insight into the functional brain networks involved can lead to new therapies
• Resting state functional magnetic resonance imaging (fMRI) can be used to identify functionally connected

brain regions and construct functional networks

• These networks can be analysed and compared using graph theory

58

Goal Use graph theory to investigate abnormal functional brain networks in a rat model of temporal lobe epilepsy

• use previously acquired resting-state fMRI images
• preprocess images and do global signal regression
• construct networks of functionally connected brain regions
• analyse networks using graph theory

Software: Matlab

59

More information?! 📪 emma.christiaen@ugent.be

USING RESTING STATE FUNCTIONAL MRI TO IDENTIFY QUASI- PERIODIC PATTERNS OF FUNCTIONAL CONNECTIVITY IN A RAT MODEL OF TEMPORAL LOBE EPILEPSY

Supervisor: Emma Christiaen Promotors: prof. Chris Vanhove, dr. Benedicte Descamps Background

• Functional magnetic resonance imaging (fMRI) is a functional imaging technique that allows the visualization of

whole-brain activity

• Resting state functional magnetic resonance imaging (fMRI) can be used to identify functionally connected

brain regions

• Functional connectivity is usually assumed to be stationary
• In reality it varies over time and recurring patterns can be found (=quasi-periodic patterns)

60

Goal Identify quasi-periodic patterns of functional connectivity using resting state fMRI data of the rat brain and investigate how these patterns differ in healthy and epileptic animals

• use previously acquired resting-state fMRI images
• identify quasi-periodic patterns
• compare patterns between healthy and

epileptic animals Software: Matlab

61

More information?! 📪 emma.christiaen@ugent.be

DYNAMIC PET IMAGING OF CHEMOGENETIC MODULATION OF THE HIPPOCAMPUS

Supervisor: Emma Christiaen Promotors: prof. Chris Vanhove, dr. Benedicte Descamps Background

• Chemogenetics is a neuromodulation technique that allows very specific activation or inhibition of neurons
• Neuronal activity can be modulated by injecting a drug-like ligand (clozapine)
• Dynamic PET imaging allows monitoring of radioactive tracer uptake over time
• Changing concentration of radioactivity in tissue gives information about underlying mechanisms of diseases or

interventions

62

Goal Use dynamic PET imaging to investigate the effects of chemogenetic modulation of the hippocampus

• acquire dynamic 18F-FDG PET images of animals while clozapine is administered -> inhibition of hippocampus
• visualize changing concentration of radioactivity in brain regions
• visualize the effect of inhibition of hippocampus

Software: MATLAB, Amide, Amira

63

More information?! 📪 emma.christiaen@ugent.be

USING DIFFUSION MRI AND TRACTOGRAPHY TO INVESTIGATE CHANGES IN WHITE MATTER TRACTS IN A RAT MODEL OF TEMPORAL LOBE EPILEPSY

Supervisor: Emma Christiaen Promotors: prof. Chris Vanhove, prof. Robrecht Raedt Background ̶ Epilepsy is a disease characterized by recurrent seizures ̶ Diffusion magnetic resonance imaging (dMRI) can be used to identify epileptogenic abnormalities ̶ White matter tracts can be mapped using tractography ̶ More insight into changes in white matter tracts during the development of epilepsy can lead to new biomarkers or therapies

64

Goal Use dMRI and tractography to investigate abnormal white matter tracts in a rat model of temporal lobe epilepsy

• use previously acquired diffusion MRI images
• preprocess images and do tractography
• investigate changes in known white matter tracts

Software: MATLAB, ExploreDTI, MRtrix3

65

More information?! 📪 emma.christiaen@ugent.be

MACHINE LEARNING FOR DISEASE DIAGNOSIS AND PROGNOSIS IN A RAT MODEL OF TEMPORAL LOBE EPILEPSY

Supervisors: Emma Christiaen, Milan Decuyper Promotors: prof. Chris Vanhove, prof. Robrecht Raedt Background

• Epilepsy is a disease characterized by recurrent seizures
• Not clear which patients will develop epilepsy after head trauma
• Need for biomarkers: functional brain networks involved in development of epilepsy
• Resting state functional magnetic resonance imaging (fMRI) can be used to identify functionally connected

brain regions and construct functional networks

66

67

Goal Use machine learning to identify epileptic rats and to predict their eventual seizure frequency

• rat model of temporal lobe epilepsy
• use previously acquired resting-state fMRI images
• construct networks of functionally connected brain regions
• extract features and build a classifier

Software: MATLAB, Python

More information?! 📪 emma.christiaen@ugent.be

ARTIFICIAL INTELLIGENCE FOR AUTOMATIC SEIZURE DETECTION IN EPILEPSY

Supervisor: dr. Lars Emil Larsen Promotors: dr. Lars Emil Larsen and prof. dr. ir. Pieter van Mierlo Background Automatic seizure detection algorithms

• preclinical experiments: save experiments countless hours
• assist clinicians inspecting electroencephalographic data from epilepsy patients
• feedback driven closed-loop neurostimulation techniques for epilepsy

69

70

Goal

• Electroencephalographic data will be available from several rodent epilepsy models, which will

be used to build seizure detection algorithms and compare performance.

• The project will revolve around testing the robustness of select machine learning techniques

such as random forest classification, support vector machines or neural networks. Software: MATLAB/Python Timeline: literature study,

More information?! 📪 Pieter.vanMierlo@ugent.be

ARTIFICIAL INTELLIGENCE FOR AUTOMATIC DETECTION OF HIGH FREQUENCY OSCILLATIONS IN EPILEPSY

Supervisor: dr. Lars Emil Larsen Promotors: dr. Lars Emil Larsen and prof. dr. ir. Pieter van Mierlo Background Pathological high frequency oscillations (pHFOs)

• hallmark of epileptogenic brain regions
• reflect activity of a diseased brain predisposed to generate epileptic seizures
• exact mechanisms underlying pHFOs are unknown
• their frequency is generally correlated to seizure frequency

pHFOs are more frequent than seizures -> useful surrogate biomarker of disease severity Quantification of pHFOs can be very labor intensive -> need for automatic detection tool

72

73

Goal

• Electroencephalographic data will be available from

several rodent epilepsy models, which will be used to build pHFO detection algorithms and compare performance.

• The project will revolve around testing the robustness of

select machine learning techniques such as random forest classification, support vector machines or neural networks. Software: MATLAB/Python Timeline: literature study,

More information?! 📪 Pieter.vanMierlo@ugent.be

Example of HFO recorded in the epileptic hippocampus: a) raw signal, b) high-pass filtered signal, c) relative time frequency plot

EEG CAP DEVELOPMENT FOR SMALL ANIMALS

Supervisor: dr. Lars Emil Larsen Promotors: prof. dr. ir. Pieter van Mierlo and prof. dr. Robrecht Raedt Background

• Preclinical validation of medical research on laboratory animals: rat brain model for human brain
• Humans: electroencephalography (EEG) using scalp electrodes
• Rats: small scalp area to place electrodes on => intracranial EEG (electrodes implanted in brain)
• Interesting to capture brain signals of rats from scalp electrodes -> a means to validate methods developed

for human scalp EEG

• Up to now, examples of scalp EEG for rats in literature are limited

75

76

Goal

• Improve this prototype and to eliminate some existing problems. You can work on different aspects, depending
• n what you prefer:

✓ How can we make the setup more practical? What is an easy, safe and fast way to fasten the cap to the rat’s head? ✓ Skin-electrode impedance could be lowered in order to better pick up the brain signals. ✓ What is the best technique and design for the electrodes? ✓ Can we improve the impedance with conductive gel? ✓ Techniques could be designed and implemented to shield the electrodes and cables from interfering signals, especially in MR room. ✓ Different materials for the electrodes to improve on MRI compatibility (making the artifact on the MR images smaller).

• Design and implement your improvements or make your own prototype. Finally, the system should be tested

and you will be able to register EEG data of rats Software: MATLAB/Python Timeline: literature study, experiments, data analysis

More information?! 📪 Pieter.vanMierlo@ugent.be

77

Don’t hesitate to contact us for more information!

The presentation will be made available

• n medisip.ugent.be