Medical Visionday Abstracts
Return to the Medical Visionday.
| Session I |
| 09.10 |
Knowledge driven analysis of cardiovascular images
Boudewijn P. F. Lelieveldt, Leiden University Medical Center, The Netherlands
Keynote address
Nowadays, a variety of 2D and 3D imaging modalities provide a wealth of
information to the
cardiologists to diagnose heart disease. However, the manual extraction
of diagnostic information from these images often relies on measurements,
which are prone to subjective errors and are time-consuming to perform.
Currently, there is a great demand for
software tools that automate(a part of) the analysis procedure to
retrieve quantitative information from cardiac images. Due to the
differences in image quality and image content in various types of
cardiovascular images, low-level and data-driven methods are generally
not sufficiently robust to perform this task. In many cases,
integration of prior knowledge about "normal" and "pathological"
anatomy is necessary to automatically extract the required information
from these images. At the Division of Image Processing at the Leiden
University Medical Center, different methods are explored to integrate
anatomical knowledge into the segmentation of cardiovascular images. In
this presentation, an overview is given of this knowledge-guided image
processing research: from statistical shape models to virtual autonomous
robots. Furthermore, a number of concrete clinical application examples to
various imaging modalities will be discussed.
|
| 10.00 |
On the M.Sc.(Eng) in medicine and technology
Jørgen Arendt Jensen, Ørsted, DTU
The education in Medicine and Technology is a new
biomedical engineering degree jointly offered
by the Faculty of Health Science at the University of Copenhagen
and the Technical University of Denmark. It consists
of a Bachelor degree (3 years) and Master degree
(2 years). The programme spans a broad range of topics
within medicine and engineering, where University
of Copenhagen teaches 30% of the curriculum and
DTU 70%. The bachelor degree consists of three
themes: The Human Biological Universe, Mechanics of
Man, and Methods in Medical Diagnosis. Each theme
spans courses at both institutions and include anatomy,
physiology, mathematics, biomedical engineering,
biomechanics, physics, chemistry, signal processing, and
medical imaging. The bachelor degree is completed
with a bachelor project carried out at a hospital
or in a biomedical company in the Copenhagen area.
The students can for the Master's programme
choose between three specializations:
Signal Processing and Diagnostics, Medical Imaging,
and Biomaterials and Biomechanics. The degree
is again completed with a project carried out in a hospital
or a biomedical company. The degree may lead to
the PhD programme (3 years) offered by the two universities.
The programme has been designed by the faculties at
Ørsted*DTU and KU, industry (Novo Nordisk, Radiometer,
and others) and Rigshospitalet. They are all
involved in planning and giving the education. Industry has
also sponsored equipment and positions exceeding DKK 6 mill.
The education started September 2003 and
219 people applied for a position, whereof 66 students were
accepted. The minimum high school grade average was 8.7,
making it one of the most sought after engineering
degrees in Denmark.
More information about the education can
be found on the Danish web-site.
|
| 10.15 |
Poster teasers |
| 10.25 |
Øresund IT - vision in the human-tech region
Teresia Nilsson, Øresund IT Academy
Øresund IT Academy is a network organisation for Danish and Swedish IT actors in the Øresund region.
Our primary goal is to make the Øresund region more appealing to students, researchers, companies and venture capitalists in the IT sector. Knowledge is created through the cross-fertilisation of Swedish and Danish cultures with world-class business practices and higher education whereby generating growth in both new and established IT organisations. This will place the Øresund region at the top of Europe's IT growth regions.
|
| Session II |
| 11.00 |
Segmentation of medical images using active contours, level-sets and variational methods
Anders Heyden, Maths, Lund University
Using active contours, especially snakes, to segment images has
proven to be a powerful tool in many different applications.
The snake is usually propagated by minimizing an energy function.
The standard way of updating the snake from the energy function
is however time consuming. We will present a fast snake evolution
algorithm, based on a more efficient numeric scheme for updating
the snake. Instead of inverting a matrix derived from approximating
derivatives in a sampled snake, an analytical expressions is obtained.
The expression takes the form of a convolution with a filter given by
an explicit formula. The filter function can then be sampled and
used to propagating snakes in a fast and straightforward manner.
The proposed method will be demonstrated on medical images of white
blood cells, i.e. leukocytes.
During the last few years level-sets have gainded a lot in popularity
for segmenting medical images. It is a powerful and mathematically
well-founded tool. We will present a level-set method for segmenting
phase-contrast MR images of the human heart. The level-set function
is based on a edge measure that drives the level-set to the desired
boundary between the flowing liquid and the surrounding muscles etc.
Finally, we will demonstrate how variational methods can be used for
segmenting medical image. These methods can be seen as an extension
to the level-set methods and they are capable of handling very
complex situations.
|
| 11.20 |
Informatics in proteomics
Ola Forsstrom-Olsson, Ludesi AB
In the pharmaceutical industry, every day that a new drug can be put on the market earlier is worth 1 million USD. One of the most mission critical applications to cut development times, and save time and money in future drug development, is proteome informatics.
Proteome informatics is a combination of proteomics and bioinformatics. It is believed that proteomics data is now overtaking genomics data in volume terms. The proteomics field has been developing gradually over the last several years, and is establishing itself as proven tool kit for performing experiments. The traditional approach consists of 2D gel electrophoresis (2DGE) followed by mass spectrometry (MS).
One of the main bottlenecks of 2DGE is image analysis. The 2DGE experiment results in very complex and irregular images of protein separation maps. The image analysis consists of spot detection, segmentation, quantification and protein spot matching between 2DGE samples. The resulting data needs to be statistically analyzed and put into an understandable context to aid biological interpretation.
Graphical presentation of proteome data is crucial to validate results. Validation in the field of proteomics is today most often done by eye-sight in combination with statistics. However, intelligent imaging and data visualization tools have the potential to significantly aid and accelerate the validation process.
Furthermore, in the biological context, the complexity of protein signaling pathways, and protein-protein interaction is tremendous. There are known to be about 30 000 genes in the human organism, but an unknown number of different proteins. This almost unlimited complexity in proteomics calls for intelligent informatics tools, with focus on data visualization, biostatistics and image analysis.
Using modern computer graphics, it is today possible to start to model an entire cell. In a relatively near future, it will be possible for pharmaceutical R&D-divisions to have a graphical representation of a cell on their desktop computers, a virtual cell, representing the protein signaling pathways in a model system cell. This will considerably aid biological interpretation and hypothesis driven R&D.
|
| 11.40 |
Where are the opportunities for successful innovation in medical technology/image analysis today?
Morten Bro Nielsen, CapMan Invest
Securing venture capital funding for a seed/early-stage life science company can be a daunting task. But approached in the right way it can also be a learning experience, which at best can lead to financing for the entrepreneurial company.
In many cases it will the first time that the entrepreneur faces the venture capital community. Understanding how venture capitalists work and what they look for in a successful company is therefore of paramount importance, if the entrepreneur is to be successful in gaining funding for his or her venture.
Venture capitalists are essentially merchants that buy and sell shares of companies. While they own shares of a company they work to enhance the value of their investment by sitting on company boards and contributing industry knowledge and connections.
The appetite of venture capitalists for new investments is of course always influenced by the perceived opportunities to sell the investments. The so-called exit-markets (the stock exchange, trade buyers etc) drive the venture capital pipeline. When exit-opportunities are few, venture capitalists tend to invest less and demand lower prices and vice-versa.
But there are some fundamentals that venture capitalists always look for when they review a company, which I will try to draw out in this presentation.
In addition, I will try to describe the best approach to developing a business strategy for an entrepreneurial company. Engineers tend to focus on technology. I will stress that focus on markets is the essential thing from the beginning. In addition, I will show the importance of addressing the combined "system" that the product will part of, rather than focusing narrowly on a technology-derived product.
Predicting what the next "new thing" is going to be has always been difficult. But I will try to outline some examples of good applications of medical vision technology that have been able to attract venture capital.
|
| Session III |
| 13.30 |
The Ladis Project - analysis of MR images in a European multicenter study
Egill Rostrup, Danish Research Centre for Magnetic Resonance, H:S Hvidovre Hospital |
| 13.50 |
Characterization and recognition of brain structure in MRI for diagnostic support
Karl Skoglund, IMM, DTU
Modern hospitals gather overwhelming amounts of data due to an increase in examination frequency and higher imaging resolution. In addition, most imaging devices, such as MR and PET scanners, acquire three-dimensional data which can be difficult to investigate and interpret. This has resulted in an increased interest in computer vision methods for diagnostic support.
Two important areas of medical image analysis are segmentation (recognition) and characterization. Segmentation deals with automatic or semi-automatic methods for locating relevant image structures. One such structure is the bundle of nervous fibres (corpus callosum) connecting the hemispheres of the brain. Measurements on such structures are used for detecting and measuring the effect of diseases such as dementia and multiple sclerosis, as well as different types of drugs. These measurements are traditionally limited to e.g. volume. It may be more relevant to look at bending/flattening, thickness and local changes. This is the goal of characterization in this context.
This talk will present an overview of seminal work within these areas, as well as later concepts and developments.
|
| 14.10 |
Bi-temporal 3D active appearance modelling with applications to unsupervised ejection fraction estimation from 4D cardiac MRI
Mikkel B. Stegmann, IMM, DTU
Rapid and unsupervised quantitative analysis is of utmost importance to ensure clinical acceptance of many examinations using cardiac magnetic resonance imaging (MRI). We present a framework that aims at fulfilling these goals for the application of left ventricular ejection fraction estimation in four-dimensional MRI. The theoretical foundation of our work is the generative two-dimensional Active Appearance Models by Cootes et al., here extended to bi-temporal, three-dimensional models. Further issues treated include correction of respiratory induced slice displacements, systole detection, and a texture model pruning strategy. Cross-validation carried out on clinical-quality scans of twelve volunteers indicates that ejection fraction and cardiac blood pool volumes can be estimated automatically and rapidly with an accuracy on par with typical inter-observer variability.
(more)
|
| 14.30 |
Analysis of multi-slice myocardial perfusion MRI
Hildur Ólafsdóttir, IMM, DTU
A novel method for registration of single and multi-slice cardiac
perfusion MRI is presented. Utilising computer intensive analyses
of variance and clustering in an annotated training set off-line,
the presented method is capable of providing registration without
any manual interaction in less than a second per frame. Changes in
image intensity during the bolus passage are modelled by a
slice-coupled active appearance model, which is augmented with a
cluster analysis of the training set. Landmark correspondences are
optimised using the minimum description length framework due to Davies et al. Image
search is verified and stabilised using perfusion specific prior
models of pose and shape estimated from training data. Given the
automatically registered sequences, a semi-quantitative perfusion
assessment is carried out. This includes signal-intensity curves
and perfusion maps for three parameters; maximum up-slope, peak
and time-to-peak. Qualitative and quantitative validation of the
method is carried out using 2000 clinical quality, short-axis,
perfusion MR slice images, acquired from ten freely breathing
patients with acute myocardial infarction. Despite evident
perfusion deficits and varying image quality in the limited
training set, a leave-one-out cross validation of the registration
method showed a mean point to curve distance of 1.25 ± 0.36
pixels for the left and right ventricle combined. Regarding the
perfusion assessment, good consistency was obtained with respect
to results obtained from the ground-truth registration. We conclude that this
learning-based registration method holds great promise for the
automation of cardiac perfusion investigations, due to its
accuracy, robustness and generalisation ability.
|
| 14.50 |
Medical Image Computing and Computer-Assisted Intervention - MICCAI 2006, Copenhagen, Denmark
Rasmus Larsen, IMM, DTU
MICCAI is the premier international conference with in-depth papers on medical image computing, computer-assisted intervention and medical robotics. The multidisciplinary nature of these emerging fields brings together clinicians, bioscientists, computer scientists, engineers, physicists & other researchers who are contributing to, and need to keep abreast of, advances in the methodology and applications.
(more)
|
| Session IV |
| 15.30 |
Modelling articular cartilage
Erik Dam, IT-University Copenhagen
The Osteo-Arthritis Project is a recently started collaboration between
the Image Analysis group at ITU and the Center for Clinical and Basic
Research in Ballerup. The focus of the project is osteo-arthritis (in
Danish: "slid-gigt") and in particular measuring and modelling the
articular cartilage ("led-brusk") in the knee.
The primary end goal of the project is to be able to accurately measure
the effect of drugs on the development of the knee cartilage. This will
enable the pharmaceutical companies to develop drugs that can effectively
help people suffering from OA. The work is still in its infancy but looks
promising.
Our approach is to first model the entire articular cartilage from MR
scans of the knee (see figure 3). From this 3D model of the cartilage, we
will design appropriate measures that accurately describe the condition of
the cartilage in terms of thickness/volume, smoothness, and softness.
The talk will introduce the medical background and the proposed
methodology that we plan to pursue.
|
| 15.50 |
Quantizing calcification in the lumbar aorta on 2D lateral x-ray images using inpainting techniques
Lars Arne Conrad-Hansen, IT-University Copenhagen
We seek to improve upon the standard method of assessing the degree of calcification in the lumbar aorta, which is commonly used on lateral 2-D x-rays. The necessity for improvement arises from the fact that the existing method can not measure subtle progressions in the plaque development; neither is it possible to express the density of individual plaques. Both of these qualities would be desireable to assess, since they are the key for making progression studies as well as for testing the effect of drugs in longitudinal studies.
Our approach is based on inpainting, a technique used in image restoration as well as postprocessing of film. In this study we discuss the potential implications of total variation inpainting for characterizing aortic calcification.
|
Return to the Medical Visionday.
Visionday frontpage
|
Further information:
Secretary Eina Boeck,
info@visionday.dk,
Ph. +45 4535 3424
|