During the weekend 8-10th of May we have presented work about fetal magnetic resonance imaging as it is done at Imaging Sciences & Biomedical Engineering in the course of Imperial festival. I recently moved to King’s college London to work on the iFind project (http://www.ifindproject.com/). Since I have been working at Imperial on motion related fetal imaging problems and high-performance computing solutions for the last few year, Imperial festival was an ideal opportunity to conclude and present this stage of my work.
The festival started with a VIP opening event on Friday evening, continued on Saturday with an exclusive session for Imperial Alumni, and was then open for the general public until Sunday evening. The organisers stated in their debriefing email that we had approximately 15,000 visitors this year, which are about 3,000 more than in the previous years. For the first time the festival ran over the whole weekend, which required to form a team of four presenters. At this point I would like to thank Emma Robinson, Kerstin Pannek, and Amir Alansary for their input and their help coping with the many interested visitors and I hope the “Grab me, I’m a Geek” badges were not too embarrassing. We started planning the event in February. Overall it was not a lot of preparation overhead for us, since our marketing staff did most of the organisation and advertising autonomously. However, that’s also how the original activity title for our presentation area changed from “Automatic motion correction for fetal MRI” to “Baby’s first home video.” The new title was ok but also shows what and how outsiders predominantly perceive our work.
We also organised a model of a fetus in utero from a midwifery equipment online shop. Thanks to the Wellcome Trust and EPSRC for sponsoring it! This model in combination with a MRI cine sequence from a moving fetus, turned out to be a great children attractor, since we let them take out the fetus and play with it. Children were then also the best in asking questions. To mention only a few examples: “… so, how does the baby get in there?”, usually followed by: “… and how does it come out when it is ready?”, which is again followed by unbelieving head-shaking if you explain them the cervix at the model. Sighing parents were not uncommon in these situations since we were careful not to pre-empt sex education. Other questions were about mind-reading, mind-programming, congenital heart conditions, down-syndrome, the safety of MRI, the price of a MRI device, and of course also about the unavoidable ethical issues when it comes to early diagnostics of untreatable conditions. Being a Computer Scientist is then sometimes a good excuse to avoid lengthy discussions or a (potentially wrong) answer at all. However, lengthy discussions are eventually unavoidable, latest when fellow high-performance computing Geeks arrive at the booth. Showing some hardware helped to attract this audience as well, which completed our visitors’ age range from 0 to 99 years.
Overall the festival was a great success and everybody of the team enjoyed this scientifically rather unusual presentation setting. The hurting back and feet from standing around all day and the sore throat from telling 1000 times the same story are all forgotten by now. After all we got free lunch and water for the whole weekend and since not everybody used their food-voucher we could even acquire a free pint from the Haemoglobin-Bar for each team-member at the end of the event on Sunday evening.
Would I do it again: Probably. I am confident talking to the general public and as long as I am not approached by a dirty, smelly or aggressive person I am enjoying being admired as the scientist who made ‘that’ possible.
Would I recommend doing it: Yes, definitely. This is great training in giving public talks, especially for shy persons. The audience is not shy to give immediate feedback and future employers love proven outreach experience in a CV.
https://kingsimaging.wordpress.com/2015/05/14/babys-first-home-video/
Donnerstag, 14. Mai 2015
Samstag, 21. März 2015
IEEE TMI paper accepted!
Our IEEE Transactions on Medical Imaging has been accepted! The tentative reference is:
Bernhard Kainz, Markus Steinberger, Wolfgang Wein, Maria Kuklisova-Murgasova, Christina Malamateniou, Kevin Keraudren, Thomas Torsney-Weir, Mary Rutherford, Paul Aljabar, Joseph V. Hajnal, and Daniel Rueckert: Fast Volume Reconstruction from Motion Corrupted Stacks of 2D Slices. IEEE Transactions on Medical Imaging, to appear, 2015. doi:10.1109/TMI.2015.2415453
check out the repository with the source code for this paper:
https://github.com/bkainz/fetalReconstruction.git
The paper will be available here as open access article.
Bernhard Kainz, Markus Steinberger, Wolfgang Wein, Maria Kuklisova-Murgasova, Christina Malamateniou, Kevin Keraudren, Thomas Torsney-Weir, Mary Rutherford, Paul Aljabar, Joseph V. Hajnal, and Daniel Rueckert: Fast Volume Reconstruction from Motion Corrupted Stacks of 2D Slices. IEEE Transactions on Medical Imaging, to appear, 2015. doi:10.1109/TMI.2015.2415453
check out the repository with the source code for this paper:
https://github.com/bkainz/fetalReconstruction.git
The paper will be available here as open access article.
Sonntag, 28. September 2014
This year's student projects
If you are Imperial Computing student and interested in one of the following projects, please feel free to write me an email (b.kainz@imperial.ac.uk).
--------------------------------------------------------------------
Title: Evaluation of lung volume estimation methods from foetal MRI data
Offered to:
401 - Individual Project MEng
401I - Individual Project MEng - EIE
401J - Computing Individual Project MEng - JMC
301 - Individual Project BEng
301I - Individual Project BEng - EIE
301J - Individual Project BEng - JMC
Recently, several novel Magnetic Resonance Imaging (MRI) sequences have been developed to scan a foetuses inside the womb. We have developed a method to estimate the lung volume of foetuses in-utero exactly. However, the importance of this value for the clinical practise is still unknown. The aim of this project is to compare the state-of-the-art methods to estimate the foetal lung volume with our novel methods and to perform statistical analysis of the results. The question to be answered is which of the investigated methods is most reliable for a clinical use as a biomarker for the baby's health.
The key objectives of this project are therefore:
- to perform a professional literature research,
- to facilitate the currently available foetal lung evaluation pipelines,
- to generate measured lung volume data for a database of foetal datasets,
- to derive statistical analysis of the variation of the results and to compare the outcome with expected values for a given foetal age.
The project should be implemented in Matlab, C/C++, or R running on a Desktop PC. Reasonable good programming and scripting skills and experience in image processing and machine learning are desirable.
--------------------------------------------------------------------
Title: Uncertainty visualization for foetal MRI 3D super-resolution
Offered to:
301 - Individual Project BEng
301I - Individual Project BEng - EIE
301J - Individual Project BEng - JMC
401 - Individual Project MEng
401I - Individual Project MEng - EIE
401J - Computing Individual Project MEng - JMC
Description:
Super-resolution reconstruction of the foetal brain got recently possible through the development of novel Magnetic Resonance Imaging (MRI) sequences for the womb. Thereby, several low resolution image stacks of the foetus are acquired in overlapping but different directions. These image stacks can subsequently be used to reconstruct a high resolution volume of the desired region, usually the foetal brain. However, the resulting images show errors that are not directly noticeable. Therefore, a useful clinical extension of this approach would be to develop a tool that is able to visualize the errors, which occur during the reconstruction process.
The key objectives of this project are therefore:
- to perform a professional literature research,
- to extend an existing framework for the reconstruction of foetal MRI data, so that additional information about the uncertainty of the calculated intensity values are generated.
- to develop a tool for the visualization of these errors. Thereby, the uncertainty information should not distract the user from the anatomical information and state-of-the-art uncertainty visualization methods should be used.
The project should be implemented in C/C++ running on a Desktop PC. The visualization can be integrated into existing visualization software. Good programming skills and experience in image processing and machine learning are desirable.
--------------------------------------------------------------------
Title: The MRI 3D foetus detector
Offered to:
301 - Individual Project BEng
301I - Individual Project BEng - EIE
301J - Individual Project BEng - JMC
401 - Individual Project MEng
401I - Individual Project MEng - EIE
401J - Computing Individual Project MEng - JMC
Description:
Recently, several novel Magnetic Resonance Imaging (MRI) sequences have been developed to scan a foetuses inside the womb. The resulting 3D scans show, besides the foetus, a significant amount of maternal tissue and are also subject to movement artefacts caused by the foetus. An automatic evaluation of the foetal organs would be desirable but is currently difficult because of the large amount of background information. The aim of this project is to suppress the maternal background information and to easy subsequent processing of pre-natal foetal MRI datasets.
The key objectives of this project are therefore:
- to perform a professional literature research,
- to implement state-of-the-art and novel object detection algorithms and to train and evaluate their performance on the individual slices of a 3D datasets,
- to compare the performance of different image analysis approaches for this specific task,
- to use machine learning for an automatically generated probability map and to visualize the likelihood of foetal tissue.
The project should be implemented in Matlab or C/C++ running on a Desktop PC. Good programming skills and experience in image processing and machine learning are desirable.
--------------------------------------------------------------------
Title: Machine learning for the automatic evaluation of 4D foetal in-utero data sets
Offered to:
401 - Individual Project MEng
401I - Individual Project MEng - EIE
401J - Computing Individual Project MEng - JMC
301 - Individual Project BEng
301I - Individual Project BEng - EIE
301J - Individual Project BEng - JMC
Description:
Currently, the face of prenatal diagnostics is changing rapidly. Novel Magnetic Resonance Imaging (MRI) sequences allow making videos of the foetal movements inside the womb with a large field of view. Thereby, the assessment of motor function is an essential component of neurology examinations. Recent research shows evidence that it is possible to assess the healthiness of a foetus by observing the kind and extend of movements. An automatic pre-classification in healthy and abnormal foetal behaviour would form a valuable tool for the clinical practise.
The key objectives of this project are therefore:
- to perform a professional literature research,
- to extend or re-implement state-of-the-art motion estimation algorithms for fully automatic limb detection using machine learning methods,
- to compare the performance of different image analysis approaches for the specific task of foetal 4D imaging,
- to deploy the developed method as a useful tool for bioengineering research at Imperial.
The project should be implemented in Matlab or C/C++ running on a Desktop PC. Good programming skills and experience in image processing and machine learning are desirable.
This project is especially rewarding because useful approaches are quickly integrated into the biological/clinical research practise. Therefore, it is likely that the successful candidate will see her or his name mentioned together with ground-breaking results about the human development!
--------------------------------------------------------------------
These projects are especially rewarding because useful approaches are quickly integrated into the biological/clinical research practise. Therefore, it is likely that the successful candidate will see her or his name mentioned together with ground-breaking results about the human development!
--------------------------------------------------------------------
Title: Evaluation of lung volume estimation methods from foetal MRI data
Offered to:
401 - Individual Project MEng
401I - Individual Project MEng - EIE
401J - Computing Individual Project MEng - JMC
301 - Individual Project BEng
301I - Individual Project BEng - EIE
301J - Individual Project BEng - JMC
Recently, several novel Magnetic Resonance Imaging (MRI) sequences have been developed to scan a foetuses inside the womb. We have developed a method to estimate the lung volume of foetuses in-utero exactly. However, the importance of this value for the clinical practise is still unknown. The aim of this project is to compare the state-of-the-art methods to estimate the foetal lung volume with our novel methods and to perform statistical analysis of the results. The question to be answered is which of the investigated methods is most reliable for a clinical use as a biomarker for the baby's health.
The key objectives of this project are therefore:
- to perform a professional literature research,
- to facilitate the currently available foetal lung evaluation pipelines,
- to generate measured lung volume data for a database of foetal datasets,
- to derive statistical analysis of the variation of the results and to compare the outcome with expected values for a given foetal age.
The project should be implemented in Matlab, C/C++, or R running on a Desktop PC. Reasonable good programming and scripting skills and experience in image processing and machine learning are desirable.
--------------------------------------------------------------------
Title: Uncertainty visualization for foetal MRI 3D super-resolution
Offered to:
301 - Individual Project BEng
301I - Individual Project BEng - EIE
301J - Individual Project BEng - JMC
401 - Individual Project MEng
401I - Individual Project MEng - EIE
401J - Computing Individual Project MEng - JMC
Description:
Super-resolution reconstruction of the foetal brain got recently possible through the development of novel Magnetic Resonance Imaging (MRI) sequences for the womb. Thereby, several low resolution image stacks of the foetus are acquired in overlapping but different directions. These image stacks can subsequently be used to reconstruct a high resolution volume of the desired region, usually the foetal brain. However, the resulting images show errors that are not directly noticeable. Therefore, a useful clinical extension of this approach would be to develop a tool that is able to visualize the errors, which occur during the reconstruction process.
The key objectives of this project are therefore:
- to perform a professional literature research,
- to extend an existing framework for the reconstruction of foetal MRI data, so that additional information about the uncertainty of the calculated intensity values are generated.
- to develop a tool for the visualization of these errors. Thereby, the uncertainty information should not distract the user from the anatomical information and state-of-the-art uncertainty visualization methods should be used.
The project should be implemented in C/C++ running on a Desktop PC. The visualization can be integrated into existing visualization software. Good programming skills and experience in image processing and machine learning are desirable.
--------------------------------------------------------------------
Title: The MRI 3D foetus detector
Offered to:
301 - Individual Project BEng
301I - Individual Project BEng - EIE
301J - Individual Project BEng - JMC
401 - Individual Project MEng
401I - Individual Project MEng - EIE
401J - Computing Individual Project MEng - JMC
Description:
Recently, several novel Magnetic Resonance Imaging (MRI) sequences have been developed to scan a foetuses inside the womb. The resulting 3D scans show, besides the foetus, a significant amount of maternal tissue and are also subject to movement artefacts caused by the foetus. An automatic evaluation of the foetal organs would be desirable but is currently difficult because of the large amount of background information. The aim of this project is to suppress the maternal background information and to easy subsequent processing of pre-natal foetal MRI datasets.
The key objectives of this project are therefore:
- to perform a professional literature research,
- to implement state-of-the-art and novel object detection algorithms and to train and evaluate their performance on the individual slices of a 3D datasets,
- to compare the performance of different image analysis approaches for this specific task,
- to use machine learning for an automatically generated probability map and to visualize the likelihood of foetal tissue.
The project should be implemented in Matlab or C/C++ running on a Desktop PC. Good programming skills and experience in image processing and machine learning are desirable.
--------------------------------------------------------------------
Title: Machine learning for the automatic evaluation of 4D foetal in-utero data sets
Offered to:
401 - Individual Project MEng
401I - Individual Project MEng - EIE
401J - Computing Individual Project MEng - JMC
301 - Individual Project BEng
301I - Individual Project BEng - EIE
301J - Individual Project BEng - JMC
Description:
Currently, the face of prenatal diagnostics is changing rapidly. Novel Magnetic Resonance Imaging (MRI) sequences allow making videos of the foetal movements inside the womb with a large field of view. Thereby, the assessment of motor function is an essential component of neurology examinations. Recent research shows evidence that it is possible to assess the healthiness of a foetus by observing the kind and extend of movements. An automatic pre-classification in healthy and abnormal foetal behaviour would form a valuable tool for the clinical practise.
The key objectives of this project are therefore:
- to perform a professional literature research,
- to extend or re-implement state-of-the-art motion estimation algorithms for fully automatic limb detection using machine learning methods,
- to compare the performance of different image analysis approaches for the specific task of foetal 4D imaging,
- to deploy the developed method as a useful tool for bioengineering research at Imperial.
The project should be implemented in Matlab or C/C++ running on a Desktop PC. Good programming skills and experience in image processing and machine learning are desirable.
This project is especially rewarding because useful approaches are quickly integrated into the biological/clinical research practise. Therefore, it is likely that the successful candidate will see her or his name mentioned together with ground-breaking results about the human development!
--------------------------------------------------------------------
These projects are especially rewarding because useful approaches are quickly integrated into the biological/clinical research practise. Therefore, it is likely that the successful candidate will see her or his name mentioned together with ground-breaking results about the human development!
Freitag, 1. August 2014
Motion Corrected 3D Reconstruction of the Fetal Thorax from Prenatal MRI
This year I'll present the following work at MICCAI:
Bernhard Kainz, Christina Malamateniou, Maria Murgasova, Kevin Keraudren, Mary Rutherford, Joseph V Hajnal, Daniel Rueckert: \emph{Motion Corrected 3D Reconstruction of the Fetal Thorax from Prenatal MRI.} to appear in: P. Golland et al. (Eds.): Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2014, Part II, LNCS 8674, pp. 284--291. Springer International Publishing Switzerland (2014)
In this paper we present a semi-automatic method for analysis of the fetal thorax in genuine three-dimensional volumes. After one initial click we localize the spine and accurately determine the volume of the fetal lung from high resolution volumetric images reconstructed from motion corrupted prenatal Magnetic Resonance Imaging (MRI). We compare the current state-of-the-art method of segmenting the lung in a slice-by-slice manner with the most recent multi-scan reconstruction methods.
We use fast rotation invariant spherical harmonics image descriptors with Classification Forest ensemble learning methods
to extract the spinal cord and show an efficient way to generate a segmentation prior for the fetal lung from this information for two different MRI field strengths. The spinal cord can be segmented with a DICE coefficient of 0.89 and the automatic lung segmentation has been evaluated with a DICE coefficient of 0.87.
We evaluate our method on 29 fetuses with a gestational age (GA) between 20 and 38 weeks and show that our computed segmentations and the manual ground truth correlate well with the recorded values in literature.
Bernhard Kainz, Christina Malamateniou, Maria Murgasova, Kevin Keraudren, Mary Rutherford, Joseph V Hajnal, Daniel Rueckert: \emph{Motion Corrected 3D Reconstruction of the Fetal Thorax from Prenatal MRI.} to appear in: P. Golland et al. (Eds.): Medical Image Computing and Computer-Assisted Intervention (MICCAI) 2014, Part II, LNCS 8674, pp. 284--291. Springer International Publishing Switzerland (2014)
In this paper we present a semi-automatic method for analysis of the fetal thorax in genuine three-dimensional volumes. After one initial click we localize the spine and accurately determine the volume of the fetal lung from high resolution volumetric images reconstructed from motion corrupted prenatal Magnetic Resonance Imaging (MRI). We compare the current state-of-the-art method of segmenting the lung in a slice-by-slice manner with the most recent multi-scan reconstruction methods.
We use fast rotation invariant spherical harmonics image descriptors with Classification Forest ensemble learning methods
to extract the spinal cord and show an efficient way to generate a segmentation prior for the fetal lung from this information for two different MRI field strengths. The spinal cord can be segmented with a DICE coefficient of 0.89 and the automatic lung segmentation has been evaluated with a DICE coefficient of 0.87.
We evaluate our method on 29 fetuses with a gestational age (GA) between 20 and 38 weeks and show that our computed segmentations and the manual ground truth correlate well with the recorded values in literature.
Dienstag, 27. Mai 2014
MICCAI paper conditionally accepted!
My work on "Motion Corrected 3D Reconstruction of the Fetal Thorax from Prenatal MRI" has been conditionally accepted for MICCAI 2014! I'm excited! This year's acceptance rate was again somewhere between 20-30%. As soon as it's accepted without "conditionally" I'll post some details on this work here.
Donnerstag, 13. Februar 2014
Ph.D. student position in medical visualization @ TU-Graz, ICG, Austria
|
Objective
|
Work on simulation mesh refinement and unstructured
grid visualization
The Institute for Computer Graphics and Vision at
Graz University of Technology, Austria, announces one new PhD student
position in the research field of medical visualization, which is granted for
three years. The start-date for the associated EU-funded project was January,
1st, 2013.
The focus of the project is on the visualization of
the results from simulations of minimally invasive interventions, such as
cancer treatment by Radio-Frequency-Ablation (RFA) or Cryoablation. The
primary task will be to conduct world-class research in visualization
technology for combined medical scanner and simulation data.
|
|
|
|
|
Description
|
The applicants must have relevant background, which
also must be sufficiently documented in the application, and their research
interest should fit within existing activities in visualization. With this
opening, we are primarily searching for a candidate in illustrative medical
visualization. If applicants can report experiences with related research,
they should verbosely document this. Candidates are also encouraged to
provide a short description of their research vision as well as letters of recommendation
from their previous employers or university teachers. The project will also
require excellent programming skills in C++ and CUDA. Knowledge of VTK and
ITK libraries is also preferred. Furthermore, very good spoken and written
English is essential for the position.
We offer a full-time work contract, including
specified time to finish a Ph.D. Thesis, 48000 € gross salary per year, in
accordance with the collective employment agreement of
Graz University of Technology. The research will be carried out under Prof.
Dieter Schmalstieg, who is full professor and head of the Institute for
Computer Graphics and Vision at Graz University of Technology (TUG), Austria.
His current research interests are augmented reality, virtual reality,
real-time graphics, medical visualization and 3D user interfaces. The
Institute for Computer Graphics and Vision at Graz University of Technology
is the only Austrian academic group with the charter to address both Computer
Vision and Computer Graphics, and is carefully nurturing a culture of Digital
Visual Information Processing to resolve the artificial boundaries between
computer graphics and computer vision. The research at ICG is focused on
Computer Graphics, Visualization, Medical Computer Vision, Object Recognition,
Object Reconstruction, Robotics, Virtual Reality and Augmented Reality.
|
|
|
|
|
|
If interested, please send an introductory letter
and CV to Bernhard Kainz, kainz@icg.tugraz.at
or Jan Egger, egger@icg.tugraz.at
|
|
|
|
|
Team
|
Computer Graphics
|
|
Contact
|
|
|
Offered as
|
Dissertation
|
|
Duration
|
Immediately
-
2015-Dec
|
Donnerstag, 9. Januar 2014
Two papers at Eurographics 2014
At Eurographics 2014: Parallel Generation of Architecture on the GPU
https://www.youtube.com/watch?v=DFtCyaBpxCk
At Eurographics 2014: On-the-fly Generation and Rendering of Infinite Cities on the GPU
https://www.youtube.com/watch?v=u-J7JokvHyw
https://www.youtube.com/watch?v=DFtCyaBpxCk
At Eurographics 2014: On-the-fly Generation and Rendering of Infinite Cities on the GPU
https://www.youtube.com/watch?v=u-J7JokvHyw
Abonnieren
Posts (Atom)