Research activities

I work on fundamental image processing issues in the Quantitative Imaging Group at TU Delft and apply the results to solve clinical problems in the Department of Radiology at AMC Amsterdam.

As such I act as the bridge between the two institutions.

My research focusses on three topics in medical image analysis:

All my work is characterized by inventions of new image measurement principles through a combination of my knowledge on imaging physics and clinical applications.

 


Assesment of Crohn's disease severity

Inflammatory bowel diseases such as Crohn's disease are a large healthcare problem in the Western World. Grading of Crohn’s disease severity is important to determine the proper treatment strategy and to quantify the response to the treatment.

Currently, Magnetic Resonance Imaging is widely used for diagnosing and grading luminal Crohn’s disease (CD). Unfortunately, MRI has been shown to be accurate for severe disease cases (91% accuracy), but mediocre for mild disease activity or remission (62% accuracy).

We aim to create a suite of fundamental tools for accurate assessment of CD severity. It will involve image analysis, classification and visualization algorithms to measure disease severity from MRI.

For example, we have taken a combined levelset/graph cuts approach to segment the bowel wall. Here is a nice visualization result:

This research project is carried out in conjunction with leading clinical groups (University College London Hospital, AMC), technical institutes (ETH Zurich, Zuse Institute Berlin) and industrial partners (Biotronics3D Inc, Vodera Inc). In fact, there are 4 Phd Students, 2 postdocs, 2 research fellows, and a scientific programmer working on the project. It is funded from the European Community’s Seventh Framework Programme as the VIGOR++ Project. I am coordinating the project, which scored 15/15 and ended first of 640 research proposals!

These are some papers by us on this topic:

  1. J.A.W. Tielbeek, F.M. Vos, J. Stoker, A computer-assisted model for detection of MRI signs of Crohn’s disease activity: future or fiction?, accepted with Abdominal Imaging, 2012.
  2. A. Wiebel, F.M. Vos, H.C. Hege, WYSIWYP: What you see is what you pick, accepted with IEEE Transactions on Visualization and Computer Graphics, 2012.
  3. F.M. Vos, J.A.W. Tielbeek, R.E. Naziroglu, Z. Li, P. Schueffler, D. Mahapatra, A. Wiebel, C. Lavini, J.M. Buhmann, H.C. Hege, J. Stoker, L.J. van Vliet, Computational Modeling for Assessment of IBD: to be or not to be?, accepted with IEEE Engineering in Medicine and Biology Conference 2012.

Back to top.

 


Diffusion tensor imaging

Diffusion Tensor Imaging (DTI) is a modern MRI technique that can supply information on the integrity of white matter tracts. It measures the ability of water molecules to move freely in the surrounding tissue. Normal white matter tracts have high diffusion along and slow across axons. Such anisotropy is measured by DTI.

A state-of-the-art post-processing method is to trace the direction of high diffusion (click on the ‘tractography’ image):

Tractography

We developed several pattern recognition procedurse to identify deviating structures in DTI data (in comparison to normals). One such procedure uses linear discriminant analysis during which non-distinghuishing areas are iteratively discarded ('shaving').

The next image shows significant, deviating areas thus identified in a group of schizofrenia patients (the arrows point at the corpus callosum and the uncinate fasciculus):

Shaving

Additionally, we proposed an optimization framework for the estimation of diffusion parameters in fiber crossings based on inverse problem solving theory.

The next images show the mean FA-profiles with standard deviations calculated via dual (red) and single (blue) tensor fits in the commissural fibers (left) and the corticospinal tract (middle). Also shown is an an overview image (right).

Notice that the dual tensor FA-profiles are more 'consistent' (i.e. do not show a dip) in the crossing region.

We closely collaborate with Liesbeth Reneman and Charles Majoie from the Radiology Department.

These are some papers by us on this topic:

  1. M.W.A. Caan, K.A. Vermeer, L.J. van Vliet, C.B.L.M. Majoie, B.D. Peters, G.J. den Heeten, and F.M. Vos, Shaving diffusion tensor images in discriminant analysis: A study into schizophrenia, Medical Image Analysis, vol. 10, no. 6, 2006, 841-849.
  2. E.J. Aukema, M.W.A. Caan, N. Oudhuis, C.B.L.M. Majoie,  F.M. Vos, L. Reneman, B.F. Last, M.A. Grootenhuis, A.Y.N. Schouten-van Meeteren, White Matter Fractional Anisotropy Correlates with Speed of Processing and Motor Speed in Young Childhood Cancer Survivors, International Journal of Radiation Oncology Biology Physics, Vol. 74, Issue 3, pp. 837-843, 2009.
  3. M.W.A. Caan, H.G. Khedoe, D.H.J. Poot, A.J. den Dekker, L.J. van Vliet, F.M. Vos, Adaptive noise filtering for accurate and precise diffusion estimation in fiber crossings, in: T. Jiang, N. Navab, J.P.W. Pluim, M.A. Viergever (Eds.), Medical Image Computing and Computer-Assisted Intervention - MICCAI 2010 (13th International Conference, Beijing, China, September 20-24, 2009, Proceedings, Part I), Lecture Notes in Computer Science, vol. 6361, Springer-Verlag Berlin Heidelberg, 2010, 167-174.
  4. M.W.A. Caan, H.G. Khedoe, D.H.J. Poot, A.J. den Dekker, S.D. Olabarriaga, C.A. Grimbergen, L.J. van Vliet, F.M. Vos, Estimation of diffusion properties in crossing fiber bundles, IEEE Transactions on Medical Imaging, Vol. 29, No. 8, 2010, 1504-1515.
  5. M.M. van der Graaff , M.W.A. Caan, E.M. Akkerman, C.A. Sage, C. Lavini, C.B. Majoie, A.J. Nederveen, A.H. Zwinderman, S. Sunaert, F. Brugman, L.H. van den Berg, M.C. de Rijk, P.A. van Doorn, J.M.B. Vianney de Jong, F.M. Vos, M. de Visser, Diffusion tensor imaging of the corticospinal tract in recent onset motor neuron disease: a longitudinal fibre tracking study, Brain, vol. 134, no. 4,  2011, 1211-28.

Back to top.

 


Virtual colonoscopy

Introduction

Virtual colonoscopy is a minimally invasive method to screen for colonic polyps (a benign tumor growing on the colon wall).

I worked on several issues in virtual colonoscopy,

The unfolded cube display

The unfolded cubic projection offered a new way for comprehensive visualizion of the surface (click on the image to play the movie):

UnfoldingCube

Here's another spectacular example:

UnfoldedExample

We demonstrated that the unfolded cube facilitates omniscent visibility, making the 3D exploration very time-efficient.

Electronic cleansing

Fecal remains may mimmick or obscure polyps due to its tissue equivalent intensity. An oral contrast agent ('fecal tagging') makes it distinguishable from the physiologic bowel wall. Over the past four years, we have been working on electronic cleansing. Electronic cleansing automatically segments the colon wall in the presence of fecal tagging.

Click on the next image to see the colon surface with and without electronic cleansing:

CleansedvsUncleansed

Automatic polyp detection

The interpretation of VC data by physicians is still rather time consuming (approximately 15 minutes per patient). More important, large polyps are sometimes missed. Therefore, methods were proposed to support the inspection by way of automatic polyp detection: computer aided diagnosis, CAD.

We designed a CAD system that works by flattening the colon wall in order to remove the protrusions. That is, the colon wall is locally deformed until it looks as if protrusions were never there. The amount of deformation needed for the flattening is a measure of the 'protrudedness'. Objects for which this measure has a high value are considered as candidate polyps.

This image illustrates what the method does:

DeformedObject

Here are some figures on the performance of the system:

CADperformance

Current work

Wenow aim at making CAD work with patients taking a low fiber diet and scanning at a low radiation dose. This would make the whole procedure more patient friendly than it currently is. However, the tagged material may become inhomogeneously mixed with fecal matter and it will form complex structures such as thin layers. Also, there is much more noise in the images due to the low radiation dose. Current electronic cleansing and CAD systems are not designed to handle such data.

All this work is done in close collaboration with Jaap Stoker from the Radiology Department at the Academic Medical Center Amsterdam.

These are some papers by us on this topic:

  1. F.M. Vos, R.E. van Gelder, I.W.O. Serlie, J. Florie, C.Y. Nio, A.S. Glas, F.H. Post, R. Truyen, F.A. Gerritsen, and J. Stoker, Three-dimensional display modes for CT colonography: conventional 3D virtual colonoscopy versus unfolded cube projection, Radiology, vol. 228, 2003, 878-885.
  2. I.W.O. Serlie, F.M. Vos, R. Truyen, F.H. Post, L.J. van Vliet, Classifying CT Image Data Into Material Fractions by a Scale and Rotaion Invariant Edge Model, IEEE transactions on image processing, 16(12), 2007, 2891-2904.
  3. C. van Wijk, V.F. van Ravesteijn, F.M. Vos, L. J. van Vliet, Detection and Segmentation of Colonic Polyps on Implicit Isosurfaces by Second Principal Curvature Flow, IEEE Transaction on Medical Imaging, Vol. 29, No. 3, 2010, 688-698.
  4. V.F. van Ravesteijn, C. van Wijk, F.M. Vos, R. Truyen, J.F. Peters, J. Stoker, L.J. van Vliet, Computer Aided Detection of Polyps in CT Colonography using Logistic Regression, IEEE Transaction on Medical Imaging, Vol 29, No. 1, 2010, 120-131.

Back to top.