Frans M Vos

Delft University of Technology, Delft, South Holland, Netherlands

Are you Frans M Vos?

Claim your profile

Publications (92)116.2 Total impact

  • [Show abstract] [Hide abstract]
    ABSTRACT: In abdominal Dynamic Contrast-Enhanced MRI (DCE-MRI) scans, a series of images are acquired over a period of 20 minutes. The liver is not only subject to quite a displacement by breathing, but also undergoes some deformation and additional motion caused by heartbeat, stomach shrinkage and bowel peristalsis. As a result, accurate liver registration in DCE-MRI is a big challenge. We present a registration framework that incorporates liver segmentation to improve registration accuracy. The segmented liver acts as prior knowledge to our in-house registration method Autocorrelation of local image structure (ALOST). The registration result is more accurate compared with original ALOST.
    No preview · Conference Paper · Feb 2016
  • [Show abstract] [Hide abstract]
    ABSTRACT: Diffusion-weighted magnetic resonance imaging permits assessment of the structural integrity of the brain's white matter. This requires unbiased and precise quantification of diffusion properties. We aim to estimate such properties in simple and complex fiber geometries up to three-way fiber crossings using rank-2 tensor model selection. A maximum a-posteriori (MAP) estimator is employed to determine the parameters of a constrained triple tensor model. A prior is imposed on the parameters to avoid the degeneracy of the model estimation. This prior maximizes the divergence between the three tensor's principal orientations. A new model selection approach quantifies the extent to which the candidate models are appropriate, i.e. a single-, dual- or triple-tensor model. The model selection precludes overfitting to the data. It is based on the goodness of fit and information complexity measured by the total Kullback-Leibler divergence (ICOMP-TKLD). The proposed framework is compared to maximum likelihood estimation on phantom data of three-way fiber crossings. It is also compared to the ball-and-stick approach from the FMRIB Software Library (FSL) on experimental data. The spread in the estimated parameters reduces significantly due to the prior. The fractional anisotropy (FA) could be precisely estimated with MAP down to an angle of approximately 40° between the three fibers. Furthermore, volume fractions between 0.2 and 0.8 could be reliably estimated. The configurations inferred by our method corresponded to the anticipated neuro-anatomy both in single fibers and in three-way fiber crossings. The main difference with FSL was in single fiber regions. Here, ICOMP-TKLD predominantly inferred a single fiber configuration, as preferred, whereas FSL mostly selected dual or triple order ball-and-stick models. The prior of our MAP estimator enhances the precision of the parameter estimation, without introducing a bias. Additionally, our model selection effectively balances the trade-off between the goodness of fit and information complexity. The proposed framework can enhance the sensitivity of statistical analysis of diffusion tensor MRI.
    No preview · Article · Nov 2015 · Physics in Medicine and Biology

  • No preview · Article · Oct 2015 · AIDS

  • No preview · Article · Aug 2015 · Experimental Gerontology
  • [Show abstract] [Hide abstract]
    ABSTRACT: We use a Field of Experts (FoE) model to segment abdominal regions from MRI affected with Crohns Disease (CD). FoE learns a prior model of diseased and normal bowel, and background non-bowel tissues from manually annotated training images. Unlike current approaches, FoE does not rely on hand designed features but learns the most discriminative features (in the form of filters) for different classes. FoE filter responses are integrated into a Random forest (RF) model that outputs probability maps for the test image and finally segments the diseased region. Experimental results show our method achieves significantly better performance than existing methods.
    No preview · Article · Jul 2015
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Registration of images in the presence of intra-image signal fluctuations is a challenging task. The definition of an appropriate objective function measuring the similarity between the images is crucial for accurate registration. This paper introduces an objective function that embeds local phase features derived from the monogenic signal in the modality independent neighborhood descriptor (MIND). The image similarity relies on the autocorrelation of local structure (ALOST) which has two important properties: (1) low sensitivity to space-variant intensity distortions (e.g. differences in contrast enhancement in MRI); (2) high distinctiveness for 'salient' image features such as edges. The ALOST method is quantitatively compared to the MIND approach based on three different datasets: thoracic CT images, synthetic and real abdominal MR images. The proposed method outperformed the NMI and MIND similarity measures on these three datasets. The registration of dynamic contrast enhanced and post-contrast MR images of patients with Crohn's disease led to relative contrast enhancement measures with the highest correlation (r=0.56) to the Crohn's disease endoscopic index of severity.
    Full-text · Article · Jul 2015 · IEEE Transactions on Medical Imaging
  • [Show abstract] [Hide abstract]
    ABSTRACT: We propose a joint registration and segmentation method for cardiac perfusion images using robust PCA (RPCA) to decompose the time series into a low rank and sparse component. Registration maximizes the smoothness of the intensity signal in the low rank component. Segmentation minimizes the sparse component's pixel intensity difference with other pixels having the same label. B-splines are used to combine the registration and segmentation costs. Tests on real patient datasets show the improved registration and segmentation accuracy over conventional methods that perform them separately.
    No preview · Conference Paper · Apr 2015
  • Zhang Li · Lucas J van Vliet · Jaap Stoker · Frans M Vos

    No preview · Conference Paper · Feb 2015
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper studies a novel method to compensate for respiratory and peristaltic motions in abdominal Dynamic Contrast Enhanced MRI. The method consists of two steps: (1) expiration-phase 'template' construction and retrospective gating of the data to the template; (2) non-rigid registration of the gated volumes. Landmarks annotated by three experts were used to directly assess the registration performance. A tri-exponential function fit to time intensity curves from regions of interest was used to indirectly assess the performance. One of the parameters of the tri-exponential fit was used to quantify the contrast enhancement. Our method achieved a mean target registration error (MTRE) of 2.12 mm, 2.27 mm and 2.33mm with respect to annotations by expert, which was close to the average inter-observer variability (2.07mm). A state-of-the-art registration method achieved a MTRE of 2.83-3.10 mm. The correlation coefficient of the contrast enhancement parameter to the Crohn's Disease Endoscopic Index of Severity (r = 0.60, p = 0.004) was higher than the correlation coefficient for the Relative Contrast Enhancement measurements values of two observers (r(Observer 1) = 0.29, p =0.2; r(Observer 2) = 0.45, p = 0.04). Direct and indirect assessments show that the expiration-based gating and a non-rigid registration approach effectively corrects for respiratory motion and peristalsis. The method facilitates improved enhancement measurement in the bowel wall in patients with Crohn's disease.
    Full-text · Article · Dec 2014 · IEEE transactions on bio-medical engineering
  • [Show abstract] [Hide abstract]
    ABSTRACT: The Fast Spoiled Gradient Echo (FSPGR) sequence is often used in MRI to create T1-weighted images. The signal intensity generated by this sequence depends on the applied flip angle. Knowing the correct flip angle is essential for the determination of T1-maps by means of an FSPGR based Variable Flip Angle (VFA) approach. Also, quantitatively determining the concentration of contrast agent in case of Dynamic Contrast Enhanced MRI (DCE-MRI) requires knowledge of the applied flip angle. In both cases, the B1-field (in)homogeneity significantly affects the results. In this paper, we present a new method to obtain both the T1-map and B1-inhomogeneity map using scans that can each be acquired within a breath-hold. We combine two short sequences for T1 quantification: Variable Flip Angle and Look-Locker (LL). The T1-maps obtained from the LL data were used to estimate the B1-inhomogeneity inherently present in the VFA data, which was then used to correct for the VFA method’s inaccurate flip angles. This way, a reliable T1-map could be computed, which was validated using both in vitro and in vivo scans. The in vitro results show that the procedure yields a substantially smaller mean deviation in T1 from the T1 measurement’s gold standard (the Inversion Recovery method), while the in vivo results show both a more accurate estimation of T1 and a reduction of the influence of the B1-inhomogeneity on the signal intensity.
    No preview · Article · Nov 2014 · Magnetic Resonance Imaging
  • [Show abstract] [Hide abstract]
    ABSTRACT: PURPOSE To prospectively compare conventional MRI, dynamic contrast-enhanced (DCE-)MRI and diffusion weighted imaging (DWI) sequences to histopathology of surgical specimens in Crohn’s disease (CD). METHOD AND MATERIALS 3T MR enterography was performed in 25 consecutive CD patients scheduled for surgery within 4 weeks. A total of one to four sections per patient were chosen for detailed image analysis. Evaluated features including mural thickness, T1 signal ratio and T2 signal ratio and on DCE-MRI maximum enhancement (ME), initial slope of increase (ISI) and time to peak (TTP) and on DWI apparent diffusion coefficient (ADC), were compared to location matched-histopathologic grading of acute inflammation score (AIS) and fibrostenosis score (FS) by Spearman correlation, Kruskal Wallis and Mann-Whitney test. RESULTS Twenty patients (mean age 38 years, range 21-73, 12 females) were included and 50 bowel locations (35 terminal ileum, 11 ascending colon, 2 transverse colon, 2 descending colon) were matched to AIS and FS. Median AIS was 3 and median FS 1. Mural thickness, T1 signal ratio, T2 signal ratio, ME and ISI correlated significantly to AIS (r = 0.634, 0.392, 0.485, 0.526, 0.514, respectively; all p<0.05). Mural thickness, T1 and T2 signal ratio differed significantly between the grades of FS (p<0.001, p=0.001, p=0.021, respectively). ME, ISI and ADC values differed significantly between the non-fibrotic sections and the fibrotic sections (p<0.001, p=0.001, p=0.023, respectively). CONCLUSION Quantitative parameters from conventional, DCE-MRI and DWI sequences correlate significantly to histopathologic scores of surgical specimens. DCE-MRI and DWI give comparable results but do not outperform conventional MRI parameters. CLINICAL RELEVANCE/APPLICATION DCE- and DWI-MRI can be used for quantitative evaluation of Crohn's disease activity.
    No preview · Conference Paper · Dec 2013
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Objective: The purpose of this article is to assess the interobserver variability for scoring MRI features of Crohn disease activity and to correlate two MRI scoring systems to the Crohn disease endoscopic index of severity (CDEIS). Materials and methods: Thirty-three consecutive patients with Crohn disease undergoing 3-T MRI examinations (T1-weighted with IV contrast medium administration and T2-weighted sequences) and ileocolonoscopy within 1 month were independently evaluated by four readers. Seventeen MRI features were recorded in 143 bowel segments and were used to calculate the MR index of activity and the Crohn disease MRI index (CDMI) score. Multirater analysis was performed for all features and scoring systems using intraclass correlation coefficient (icc) and kappa statistic. Scoring systems were compared with ileocolonoscopy with CDEIS using Spearman rank correlation. Results: Thirty patients (median age, 32 years; 21 women and nine men) were included. MRI features showed fair-to-good interobserver variability (intraclass correlation coefficient or kappa varied from 0.30 to 0.69). Wall thickness in millimeters, presence of edema, enhancement pattern, and length of the disease in each segment showed a good interobserver variability between all readers (icc = 0.69, κ = 0.66, κ = 0.62, and κ = 0.62, respectively). The MR index of activity and CDMI scores showed good reproducibility (icc = 0.74 and icc = 0.78, respectively) and moderate CDEIS correlation (r = 0.51 and r = 0.59, respectively). Conclusion: The reproducibility of individual MRI features overall is fair to good, with good reproducibility for the most commonly used features. When combined into the MR index of activity and CDMI score, overall reproducibility is good. Both scores show moderate agreement with CDEIS.
    Full-text · Article · Dec 2013 · American Journal of Roentgenology
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We propose an information processing pipeline for segmenting parts of the bowel in abdominal magnetic resonance (MR) images that are affected with Crohn's disease (CD). Given a MRI test volume, it is first oversegmented into supervoxels and each supervoxel is analyzed to detect presence of Crohn's disease using random forest (RF) classifiers. The supervoxels identified as containing diseased tissues define the volume of interest (VOI). All voxels within the VOI are further investigated to segment the diseased region. Probability maps are generated for each voxel using a second set of RF classifiers which give the probabilities of each voxel being diseased, normal or background. The negative log-likelihood of these maps are used as penalty costs in a graph cut segmentation framework. Low level features like intensity statistics, texture anisotropy and curvature asymmetry, and high level context features are used at different stages. Smoothness constraints are imposed based on semantic information (importance of each feature to the classification task) derived from the second set of learned RF classifiers. Experimental results show that our method achieves high segmentation accuracy with Dice metric values of 0.90±0.04 and Hausdorff distance of 7.3±0.8 mm. Semantic information and context features are an integral part of our method and are robust to different levels of added noise.
    Full-text · Article · Sep 2013
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: To prospectively compare conventional MRI sequences, dynamic contrast enhanced (DCE) MRI and diffusion weighted imaging (DWI) with histopathology of surgical specimens in Crohn's disease. 3-T MR enterography was performed in consecutive Crohn's disease patients scheduled for surgery within 4 weeks. One to four sections of interest per patient were chosen for analysis. Evaluated parameters included mural thickness, T1 ratio, T2 ratio; on DCE-MRI maximum enhancement (ME), initial slope of increase (ISI), time-to-peak (TTP); and on DWI apparent diffusion coefficient (ADC). These were compared with location-matched histopathological grading of inflammation (AIS) and fibrosis (FS) using Spearman correlation, Kruskal-Wallis and Chi-squared tests. Twenty patients (mean age 38 years, 12 female) were included and 50 sections (35 terminal ileum, 11 ascending colon, 2 transverse colon, 2 descending colon) were matched to AIS and FS. Mural thickness, T1 ratio, T2 ratio, ME and ISI correlated significantly with AIS, with moderate correlation (r = 0.634, 0.392, 0.485, 0.509, 0.525, respectively; all P < 0.05). Mural thickness, T1 ratio, T2 ratio, ME, ISI and ADC correlated significantly with FS (all P < 0.05). Quantitative parameters from conventional, DCE-MRI and DWI sequences correlate with histopathological scores of surgical specimens. DCE-MRI and DWI parameters provide additional information. • Conventional MR enterography can be used to assess Crohn's disease activity. • Several MRI parameters correlate with inflammation and fibrosis scores from histopathology. • Dynamic contrast enhanced imaging and diffusion weighted imaging give additional information. • Quantitative MRI parameters can be used as biomarkers to evaluate Crohn's disease activity.
    Full-text · Article · Sep 2013 · European Radiology
  • Zhang Li · Lucas J. van Vliet · Frans M. Vos
    [Show abstract] [Hide abstract]
    ABSTRACT: The modality independent neighbourhood descriptor (MIND) is a local registration metric that is based on the principle of self-similarity. However, the metric requires recalculation of the self similarity during registration as this inherently changes during image deformation. We propose a self similarity registration method based on the Hessian (HE) that efficiently deals with the recalculation issue. The representation of the local self-similarity via the Hessian enables keeping it up to date during deformation. As such, the registration procedure is efficient and not prone to fall in local minima. We have shown that reorienting the hessian gives a significant improvement (p<0.05) over leaving the reorientation out. Our technique also has a better performance over the existing MIND method on the DIR-Lab dataset as well as on abdominal MRI datasets albeit not significant. Ultimately, we will use the technique to quantify Crohn’s disease severity based on the relative contrast enhancement in registered images.
    No preview · Conference Paper · Sep 2013
  • [Show abstract] [Hide abstract]
    ABSTRACT: CT colonography (CTC) is one of the recommended methods for colorectal cancer screening. The subject's preparation is one of the most burdensome aspects of CTC with a cathartic bowel preparation. Tagging of the bowel content with an oral contrast medium facilitates CTC with limited bowel preparation. Unfortunately, such preparations adversely affect the 3D image quality. Thus far, data acquired after very limited bowel preparation were evaluated with a 2D reading strategy only. Existing cleansing algorithms do not work sufficiently well to allow a primary 3D reading strategy. We developed an electronic cleansing algorithm, aimed to realize optimal 3D image quality for low-dose CTC with 24-hour limited bowel preparation. The method employs a principal curvature flow algorithm to remove heterogeneities within poorly tagged fecal residue. In addition, a pattern recognition based approach is used to prevent polyp-like protrusions on the colon surface from being removed by the method. Two experts independently evaluated 40 CT colonography cases by means of a primary 2D approach without involvement of electronic cleansing as well as by a primary 3D method after electronic cleansing. The data contained four variations of 24-hour limited bowel preparation and was based on a low radiation dose scanning protocol. The sensitivity for lesions 6mm was significantly higher for the primary 3D reading strategy (84%) than for the primary 2D reading strategy (68%) (p = 0.031). The reading time was increased from 5:39min (2D) to 7:09min (3D) (p = 0.005); the readers' confidence was reduced from 2.3 (2D) to 2.1 (3D) (p = 0.013) on a 3-point Likert scale. Polyp conspicuity for cleansed submerged lesions was similar to not submerged lesions (p = 0.06). To our knowledge this study is the first to describe and clinically validate an electronic cleansing algorithm that facilitates low-dose CTC with 24-hour limited bowel preparation.
    No preview · Article · May 2013 · IEEE transactions on bio-medical engineering
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Anatomical alignment in neuroimaging studies is of such importance that considerable effort is put into improving the registration used to establish spatial correspondence. Tract-based spatial statistics (TBSS) is a popular method for comparing diffusion characteristics across subjects. TBSS establishes spatial correspondence using a combination of nonlinear registration and a "skeleton projection" that may break topological consistency of the transformed brain images. We therefore investigated feasibility of replacing the two-stage registration-projection procedure in TBSS with a single, regularized, high-dimensional registration. To optimize registration parameters and to evaluate registration performance in diffusion MRI, we designed an evaluation framework that uses native space probabilistic tractography for 23 white matter tracts, and quantifies tract similarity across subjects in standard space. We optimized parameters for two registration algorithms on two diffusion datasets of different quality. We investigated reproducibility of the evaluation framework, and of the optimized registration algorithms. Next, we compared registration performance of the regularized registration methods and TBSS. Finally, feasibility and effect of incorporating the improved registration in TBSS were evaluated in an example study. The evaluation framework was highly reproducible for both algorithms (R(2) 0.993; 0.931). The optimal registration parameters depended on the quality of the dataset in a graded and predictable manner. At optimal parameters, both algorithms outperformed the registration of TBSS, showing feasibility of adopting such approaches in TBSS. This was further confirmed in the example experiment.
    Full-text · Article · Mar 2013 · NeuroImage
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We address the problem of weakly supervised segmentation (WSS) of medical images which is more challenging and has potentially greater applications in the medical imaging community. Training images are labeled only by the classes they contain, and not by the pixel labels. We make use of the Multi Image Model (MIM) for weakly supervised segmentation which exploits superpixel features and assigns labels to every pixel. MIM connects superpixels from all training images in a data driven fashion. Test images are integrated into the MIM for predicting their labels, thus making full use of the training samples. Experimental results on abdominal magnetic resonance (MR) images of patients with Crohn's disease show that WSS performs close to fully supervised methods and given sufficient samples can perform on par with fully supervised methods.
    Full-text · Conference Paper · Jan 2013
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: We propose a graph cut based method to segment regions in abdominal magnetic resonance (MR) images affected with Crohn's disease (CD). Intensity, texture, curvature and context information are used with random forest (RF) classifiers to calculate probability maps for graph cut segmentation. The RF classifiers also provide semantic information used to design a novel smoothness cost. Experimental results on 26 real patient data shows our method accurately segments the diseased areas. Inclusion of semantic information significantly improves segmentation accuracy and its importance is reflected in quantitative measures and visual results.
    Full-text · Conference Paper · Jan 2013
  • [Show abstract] [Hide abstract]
    ABSTRACT: Scientists, engineers and physicians are used to analyze 3D data with slice-based visualizations. Radiologists for example are trained to read slices of medical imaging data. Despite the numerous examples of sophisticated 3D rendering techniques, domain experts, who still prefer slice-based visualization do not consider these to be very useful. Since 3D renderings have the advantage of providing an overview at a glance, while 2D depictions better serve detailed analyses, it is of general interest to better combine these methods. Recently there have been attempts to bridge this gap between 2D and 3D renderings. These attempts include specialized techniques for volume picking in medical imaging data that result in repositioning slices. In this paper, we present a new volume picking technique called WYSIWYP (“what you see is what you pick”) that, in contrast to previous work, does not require pre-segmented data or metadata and thus is more generally applicable. The positions picked by our method are solely based on the data itself, the transfer function, and the way the volumetric rendering is perceived by the user. To demonstrate the utility of the proposed method, we apply it to automated positioning of slices in volumetric scalar fields from various application areas. Finally, we present results of a user study in which 3D locations selected by users are compared to those resulting from WYSIWYP. The user study confirms our claim that the resulting positions correlate well with those perceived by the user.
    No preview · Article · Dec 2012 · IEEE Transactions on Visualization and Computer Graphics

Publication Stats

1k Citations
116.20 Total Impact Points

Institutions

  • 2000-2015
    • Delft University of Technology
      • • Faculty of Applied Sciences (AS)
      • • Department of Imaging Science and Technology
      • • Applied Geophysics and Petrophysics
      Delft, South Holland, Netherlands
  • 2007-2013
    • University of Amsterdam
      • Department of Radiology
      Amsterdamo, North Holland, Netherlands
  • 2003
    • Academisch Medisch Centrum Universiteit van Amsterdam
      • Department of Radiology
      Amsterdamo, North Holland, Netherlands
  • 2001
    • Erasmus Universiteit Rotterdam
      • Department of Radiology
      Rotterdam, South Holland, Netherlands