Image processing and visualization topics Image processing and
visualization topics
Cardiovascular medical imaging, in particular echocardiography
and cardiac magnetic resonance imaging, has reached a level that
provides significant added value for cardiologists in diagnosing
cardiovascular diseases.At present most clinical tools are modality
specific and relevant information is merged retrospectively by the
cardiologist.Assuming that combining multi-modal anatomical and
functional information allows for a quicker assessment of a given
case, the goal is to provide improved diagnostic tools that enhance
both the qualitative (e.g. data visualization and fusion) and
quantitative (e.g. more accurate functional measures based on
several data sources) assessment process.This would benefit the
quantification process in the daily clinic and would enable the
creation of improved anatomic and functional cardiac models that
enable the physiological understanding of the heart in healthy and
diseased cases.The proposed projects are suitable for
project/master work.1. Augmented reality for live ultrasound
scanningCurrently the successful employment of ultrasound is highly
dependent on the experience of the examiner. During acquisition, in
addition to the cognitive load associated with interpreting the
image, the examiner has to control the correct positioning and
orientation of the transducer in order to ensure that the correct
anatomical area is imaged and that the image quality is
satisfactory.By providing visual guidance through overlays, which
communicate the anatomical structures being imaged, one can greatly
improve the understanding of the structural arrangement of tissues
during the scanning procedure. Thus it is highly desirable to
assist the ultrasound examiner and as such partially alleviate the
burden of image interpretation.The aim of the project is to develop
2D/3D visualization methods that enable the clinician to find a
visual correspondence between the ultrasound data being acquired
and a generic anatomical mesh-model of a human heart.
Figure: Standard ultrasound slice augmented with anatomical
information from a highly descriptive cardiac mesh
model.Objectives: Real-time generation of augmented views on the
tablet, investigate the possibilities for generating advanced
visualizations and implement them directly on the tablet. Non-rigid
mesh deformation of the heart model based on existing anatomic
landmarks Testing of the system during live scanningSkills:
Knowledge of C++ programming Desire to learn Open GLSL and Android
programmingContact: Gabriel Kiss or Hans Torp2. Real-time fusion of
multi-modal cardiac data on the GPUFull 3D datasets of the heart
can be acquired using different imaging modalities e.g. 3D
echocardiography (3D echo) and cardiac magnetic resonance imaging
(CMR).Using image fusion the strengths of different image
modalities can be combined. As such, by co-registering multi-modal
datasets, a direct spatial relationship between anatomical and
functional information in the underlying data is established and
visualized.Furthermore in the case of 3D echo the image quality has
large variations between subjects, the spatial resolution varies
with depth and the tissue contrast is angle dependent. One
possibility of improving image to noise ratio and to extend the
field of view is to acquire multiple datasets from different
angles, detect corresponding features in each view and fuse the
recordings in order to generate a combined volume.The aim of the
project is to develop data registration techniques for fusing
multiple 3D cardiac acquisitions with the goal of improving
signal-to-noise ratio, spatial resolution and field of view (3D
echo to 3D echo volumetric fusion) or to improve the diagnostic
process (3D echo to CMR alignment) and to demonstrate their
applicability during live scanning.
Figure: Top row: fusion of 3D echocardiographic and CMR data.
Bottom row: registration of 3D echocardiographic data acquired from
different views (e.g. parasternal, apical)Objectives: Extend and
customize the current registration methods for multimodal 3D
cardiac data Improve the registration method relying on data from
an optical tracking system Validate the developed methods during
live scanningSkills: Knowledge of C++ programming (existing code
implemented in C++) Desire to learn parallel computing (CUDA or
OpenCL)Contact: Gabriel Kiss or Hans TorpLink to this Back
Ultrasound Mediated Drug Delivery
To minimize undesired side effects of cancer drugs on normal
tissue, several groups are working on encapsulating the cancer
drugs into particles of diameter ~ 100 nm. The delivery of both the
particles and the drugs can be enhanced by ultrasound.The
capillaries in a tumor grow aggressively with an imperfect wall, so
that these particles leak into the space between tumor cells (the
interstitium), while they are maintained in the blood stream in
normal tissue, which has well developed capillary walls.
Encapsulating the smaller anti-cancer drug molecules into particles
opens for selective pharmacological treatment of tumor tissue,
while normal tissue is not exposed to the drug.Ultrasound radiation
force can increase transport of the particles deeper into the
interstitium. Increase of temperature produced by ultrasound
absorption will also increase diffusion of the particles into the
interstitium.Ultrasound can also be used to break the particles.
This seems to be stimulated by cavitation of small gas-bubble
nuclei in the tissue. It is therefore a very interesting strategy
to combine small gas bubbles (diam ~ 2m) and the drug encapsulating
nano-particles with ultrasound.There are several interesting Master
and PhD topics in this field, ranging from multi-frequency
ultrasound acoustics and transducer arrays for imaging of particles
and stimulated transport and breakage of the particles signal
processing for multi-frequency ultrasound imaging (SURF Imaging) of
the particles combined optical imaging of particles with ultrasound
mediated drug delivery experimental studies of ultrasound mediated
transport and breakage of gas micro-bubbles and drug encapsulating
nano-particles in lab models and small animal tumor modelsContact:
Professor Bjrn Angelsen Researcher Rune HansenLink to this Back
Beamforming topics Beamforming topics
Ultrasound beamforming is about controlling the interference
pattern of the acoustical waves emitted by several small
piezo-electric elements at the tip of a transducer. If you apply an
alternating voltage signal to an piezo-electric element, it will
start vibrating and emit sound. If you select the spacing between
your elements and the delay in the elements'signals just right, you
can create an interference pattern that's to your benefit, in
particular one in which the majority of the signal energy all goes
out in one angular direction.When using the transducer to receive
sound, the principles are the same. Received sound vibrations at
the elements will be converted to an electric signal. Adjust the
amplitude anddelays of the received signal on each element before
summing, and you'll be able to receive from a chosen angular
direction.1. New beamforming techniques based on spatial
coherenceConventional ultrasound images are formed by delay-and-sum
beamforming of the backscattered echoes received by the transducer
elements. Such an interferential process can however be challenged
in the presence of phase aberrations, acoustic reverberation
clutters, strong off-axis targets or electronic noise. These
phenomenons will all contribute in decreasing the spatial coherence
of the received ultrasound signal across the aperture of the
transducer, and will result in blurring artifacts in the
delay-and-sum ultrasound image.Modern ultrasound scanners allow for
software processing of the data received by all the transducer
elements. We can now test new beamforming techniques that can cope
better with in-vivo acoustic perturbations, resulting therefore in
a better contrast and signal-to-noise ratio. Several beamforming
techniques based on the coherence of the received data have been
proposed lately, promising for better ultrasound images using a
non-linear beamforming scheme.Aim: Implement and test new
beamforming algorithms based on spatial coherence Apply them on
simulated and in-vivo collected channel data, and compare with
B-mode imagesQualifications: Interested in ultrasound medical
imaging Signal processing Matlab programming skillsContact persons:
Bastien Denarie Hans Torp2. Estimation and correction of speed of
soundBackground:Ultrasound imaging of tissue always assumes a sound
velocity of 1540 m/s. This is close to correct on average, but not
always. In extreme cases, with a high percentage of fatty tissue,
the actual average velocity can be as low as 1400 m/s. This causes
the ultrasound image to completely collapse. Estimating and
correcting the actual speed of sound in these cases can bring back
the ultrasound image.Including such correction in modern ultrasound
scanners will have a large clinical importance and will enable
better/more accurate diagnosis on difficult patients.Our Ultrasound
lab now has excellent availability of raw data for doing such
estimation on in-vivo data.Aim: Investigate methods for automatic
estimation of sound velocity. Test methods in simulations Test
methods on in-vitro and in-vivo data.Qualifications: Medical
ultrasound theory/acoustics Signal processing Mathematics Matlab
programmingContact persons: Hans Torp Bastien Denarie Tore
BjaastadExample of impact of speed of sound correction - from
Phillips (pdf)3. Detection and compensation for blocked/noisy
channelsBackground:The ultrasound scanners of today typically
blindly utilize data from all probe elements without any form of
analysis. In the future, raw channel data will be available for
real time processing on a CPU inside a ultrasound machine. This
allows analysis of the data prior to beamforming. A simple, yet
potentially very usefull analysis, is to check whether an element
is contributing positively to the beamformer sum. In some cases
elements might be blocked. For example by ribs. Elements could also
be broken, or electronics for a channel could be broken. In all
these cases it would be beneficial to exclude data from these
channels/elements.Our Ultrasound lab provides excellent
availability of raw data for testing out the impact of such errors
and the benefits of correcting for them.Aim: Investigate/develop
techniques for detecting and compensating for blocked/broken
channels. Test techniques in simulations Test techniques on in-vivo
and in-vitro data.Qualifications: Signal processing Matlab
programmingContactpersons: Hans Torp Bastien Denarie Tore
BjaastadLink to this Back Doppler imaging of blood flow Doppler
imaging of blood flow
Blood gives very weak echoes compared to the surrounding tissue,
and it is usually not visible in ultrasound images. But since the
blood is moving, it creates a Doppler-shift in the returning
echoes. This makes it possible to filter out just the signal from
moving blood and detect the velocity of the blood from the Doppler
shift.Some established methods for presenting this information is
Color Flow Doppler and Pulsed Wave Doppler, as shown in the figures
on the right: PW Doppler (upper panel) and Color Flow (lower panel)
are well established methods for imaging blood flow1: Simulation
models for Doppler imaging based on computational fluid dynamics
(patient specific models)There is increasing interest in using
advanced computational models for flow based on computation fluid
dynamics (CFD) as input to ultrasound imaging simulations. This
gives the possibility to develop and compare new imaging algorithms
towards a realistic ground truth where all information of scatterer
movement is available. We have previously developed a framework for
these simulations in cooperation with the University in Ghent,
Belgium. However, the simulations can take a long time to finish
(~days), and it is critical to find approaches to reduce this time
during development. In this project you will work with the
trade-off between simulation accuracy and time for producing
realistic Doppler signals from both patient specific CFD-models and
a more ideal jet-flow. In addition to the simulations, a new flow
imaging algorithm will be investigated based on ultrafast
acquisition scheme, providing an image frame rate > 1000
fps.Preferred qualifications: Programming in
Matlab.Contact:Professor Hans Torp, Researcher Lasse Lvstakken2:
Tracking of complex blood flow in congenital heart disease
(babies)Cardiac flow patterns may reveal several kinds of
cardiovascular disease. Well known examples include the detection
and quantification of leaky heart valves and poor systolic and
diastolic function. Conventional flow imaging with ultrasound is
however limited to only measuring the velocity component along the
ultrasound beam, i.e. it is a one-dimensional and angle-dependent
measurement. This discrepancy limits the usefulness of Doppler
ultrasound in diagnostic settings. In this work we will focus on
further developing multi-dimensional flow velocity estimators based
on speckle tracking, i.e. image pattern matching techniques. The
main clinical application will be pediatric cardiology, with the
aim to improve the depiction of complex flow patterns such as
vortex and shunt flow.The proposed multidimensional approaches
proposed are however not as robust as conventional methods. Thus,
the aims of this student project will be to further develop and
optimize tracking algorithms within a robust framework based on the
predicted motion of flow, for example using a Kalman filter.Aims:
Further develop and validate robust tracking algorithms that
optimally weight measurement and modelling errors Test the proposed
methods on simplified simulations as well as in vivo data from
pediatric cardiologyQualifications:Knowledge of digital signal
processing and preferably Matlab.Contact:Researcher Lasse
Lvstakken, PhD student Solveig Alnes3: Navigated ultrasound imaging
3-D reconstruction of (pulsatile) artery geometry and
flowConventional ultrasound imaging of blood flow in central and
peripheral arteries is today based on 2-D imaging, while pathology
related to atherosclerosis is inherently three-dimensional. While
real-time 3-D ultrasound is available for cardiac imaging,
transducers for vascular imaging are not yet available. However, by
utilizing highly accurate position sensors during scanning, it is
possible to reconstruct the 3-D geometry of arteries based on
multiple 2-D flow and B-mode images. In this project we will
utilize recently installed navigation system based on optical and
magnetic sensors to reconstruct 3-D flow in the carotid artery.
This flow is highly pulsatile, and we will also incorporate
information from ECG (electro-cardiogram) to also get timing
information. The imaging approach will follow a recent plane-wave
imaging scheme, where a high frame rate and high image quality can
be achieved. Investigations will first be done using in vitro setup
of known stationary and pulsatile flow. In vivo imaging in healthy
volunteers will further be tries to show the potential of mapping
arterial geometry and pulsatile 3-D flow patterns.Preferred
qualifications:Programming in Matalb and C++Contact:Researcher
Lasse Lvstakken, PhD student Daniel H. IversenLink to this Back
Cardiac ultrasound Cardiac ultrasoundCardiac ultrasound, also known
as echocardiography, concerns the ultrasound imaging of a very fast
moving complex organ positioned deep within the body - the heart.
In Trondheim, a group of engineers and medical doctors have a more
than 30 year history for collaborative efforts on improving the
methods for imaging and analysis of the function of the
heart.Topic: Myocardial deformationThe heart is a muscle that both
pushes blood out and sucks new blood in. In some cardiac diseases,
the ability to suck is more reduced than the ability to push
because the heart muscle is stiff, and this creates problems for
the filling of the heart. The consequence is that the heart pumps
less efficiently, and that the patient's exercise capacity is
reduced. It is difficult to measure cardiac muscle stiffness
directly, but with new ultrasound technology, we can achieve
extreme time resolution (> 1000 images per second). Thus we can
see mechanical phenomena we have not seen before, such as fast
deformation waves.Our hypothesis is that the velocity of these
waves is related to cardiac muscle stiffness.The task of this
thesis is to develop a finite element model where the propagation
of such waves can be simulated. The work will be based on
FEM-models already developed at the Dept. of structural
engineering.Supervisors:Leif Rune HellevikHans TorpBrage H
Amundsen