Article

Segmenting Articular Cartilage Automatically Using a Voxel Classification Approach

IT Univ. of Copenhagen
IEEE Transactions on Medical Imaging (Impact Factor: 3.39). 02/2007; 26(1):106 - 115. DOI: 10.1109/TMI.2006.886808
Source: DBLP

ABSTRACT

We present a fully automatic method for articular cartilage segmentation from magnetic resonance imaging (MRI) which we use as the foundation of a quantitative cartilage assessment. We evaluate our method by comparisons to manual segmentations by a radiologist and by examining the interscan reproducibility of the volume and area estimates. Training and evaluation of the method is performed on a data set consisting of 139 scans of knees with a status ranging from healthy to severely osteoarthritic. This is, to our knowledge, the only fully automatic cartilage segmentation method that has good agreement with manual segmentations, an interscan reproducibility as good as that of a human expert, and enables the separation between healthy and osteoarthritic populations. While high-field scanners offer high-quality imaging from which the articular cartilage have been evaluated extensively using manual and automated image analysis techniques, low-field scanners on the other hand produce lower quality images but to a fraction of the cost of their high-field counterpart. For low-field MRI, there is no well-established accuracy validation for quantitative cartilage estimates, but we show that differences between healthy and osteoarthritic populations are statistically significant using our cartilage volume and surface area estimates, which suggests that low-field MRI analysis can become a useful, affordable tool in clinical studies

Download full-text

Full-text

Available from: Jenny Folkesson, Mar 14, 2013
  • Source
    • "To the best of our knowledge, no fully automated algorithm for segmenting individual femoral and acetabular cartilage layers has been reported for MR images of the hip joint without the use of joint distraction or contrast agents. In comparison, a number of MR studies on the knee joint have reported automatic segmentation approaches for morphometric analyses of femoral, tibial and patellar cartilage plates (Folkesson et al 2007, Fripp et al 2010, Dodin et al 2010, Yin et al 2010, Lee et al 2011, Tamez-Pena et al 2012). Amongst these approaches, the BCI has been used regularly as an underlying frame of reference for subsequent cartilage segmentation (Dodin et al 2010, Fripp et al 2010, Lee et al 2011, Yin et al 2010). "
    [Show abstract] [Hide abstract]
    ABSTRACT: To develop an automated approach for 3D quantitative assessment and measurement of alpha angles from the femoral head-neck (FHN) junction using bone models derived from magnetic resonance (MR) images of the hip joint.Bilateral MR images of the hip joints were acquired from 30 male volunteers (healthy active individuals and high-performance athletes, aged 18-49 years) using a water-excited 3D dual echo steady state (DESS) sequence. In a subset of these subjects (18 water-polo players), additional True Fast Imaging with Steady-state Precession (TrueFISP) images were acquired from the right hip joint. For both MR image sets, an active shape model based algorithm was used to generate automated 3D bone reconstructions of the proximal femur. Subsequently, a local coordinate system of the femur was constructed to compute a 2D shape map to project femoral head sphericity for calculation of alpha angles around the FHN junction. To evaluate automated alpha angle measures, manual analyses were performed on anterosuperior and anterior radial MR slices from the FHN junction that were automatically reformatted using the constructed coordinate system.High intra- and inter-rater reliability (intra-class correlation coefficients > 0.95) was found for manual alpha angle measurements from the auto-extracted anterosuperior and anterior radial slices. Strong correlations were observed between manual and automatic measures of alpha angles for anterosuperior (r = 0.84) and anterior (r = 0.92) FHN positions. For matched DESS and TrueFISP images, there were no significant differences between automated alpha angle measures obtained from the upper anterior quadrant of the FHN junction (two-way repeated measures ANOVA, F < 0.01, p = 0.98).Our automatic 3D method analysed MR images of the hip joints to generate alpha angle measures around the FHN junction circumference with very good reliability and reproducibility. This work has the potential to improve analyses of cam-type lesions of the FHN junction for large-scale morphometric and clinical MR investigations of the human hip region.
    Full-text · Article · Sep 2015 · Physics in Medicine and Biology
  • Source
    • "Since performing such segmentations manually is very time consuming and prone to inter-and intra-observer variabilities, a large variety of algorithms has been developed to perform automated segmentation . Many methods for automated image segmentation are based on supervised learning (Anbeek et al., 2005; Folkesson et al., 2007; Geremia et al., 2011; Van der Lijn et al., 2008; Liu et al., 2006; Van Opbroek et al., 2013b), where a segmentation model is trained on a manually annotated set of training images. To train a successful model, these supervised-learning techniques require a training set that is both sufficiently large to capture a large variation of appearances and representative of the target data. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Many automatic segmentation methods are based on supervised machine learning. Such methods have proven to perform well, on the condition that they are trained on a sufficiently large manually labeled training set that is representative of the images to segment. However, due to differences between scanners, scanning parameters, and patients such a training set may be difficult to obtain. We present a transfer-learning approach to segmentation by multi-feature voxelwise classification. The presented method can be trained using a heterogeneous set of training images that may be obtained with different scanners than the target image. In our approach each training image is given a weight based on the distribution of its voxels in the feature space. These image weights are chosen as to minimize the difference between the weighted probability density function (PDF) of the voxels of the training images and the PDF of the voxels of the target image. The voxels and weights of the training images are then used to train a weighted classifier. We tested our method on three segmentation tasks: brain-tissue segmentation, skull stripping, and white-matter-lesion segmentation. For all three applications, the proposed weighted classifier significantly outperformed an unweighted classifier on all training images, reducing classification errors by up to 42%. For brain-tissue segmentation and skull stripping our method even significantly outperformed the traditional approach of training on representative training images from the same study as the target image. Copyright © 2015 Elsevier B.V. All rights reserved.
    Full-text · Article · Jul 2015 · Medical image analysis
  • Source
    • "To the best of our knowledge, no fully automated algorithm for segmenting individual femoral and acetabular cartilage layers has been reported for MR images of the hip joint without the use of joint distraction or contrast agents. In comparison, a number of MR studies on the knee joint have reported automatic segmentation approaches for morphometric analyses of femoral, tibial and patellar cartilage plates (Folkesson et al 2007, Fripp et al 2010, Dodin et al 2010, Yin et al 2010, Lee et al 2011, Tamez-Pena et al 2012). Amongst these approaches, the BCI has been used regularly as an underlying frame of reference for subsequent cartilage segmentation (Dodin et al 2010, Fripp et al 2010, Lee et al 2011, Yin et al 2010). "
    [Show abstract] [Hide abstract]
    ABSTRACT: Accurate segmentation of hip joint cartilage from magnetic resonance (MR) images offers opportunities for quantitative investigations of pathoanatomical conditions such as osteoarthritis. In this paper, we present a fully automatic scheme for the segmentation of the individual femoral and acetabular cartilage plates in the human hip joint from high-resolution 3D MR images. The developed scheme uses an improved optimal multi-object multi-surface graph search framework with an arc-weighted graph representation that incorporates prior morphological knowledge as a basis for segmentation of the individual femoral and acetabular cartilage plates despite weak or incomplete boundary interfaces. This automated scheme was validated against manual segmentations from 3D true fast imaging with steady-state precession (TrueFISP) MR examinations of the right hip joints in 52 asymptomatic volunteers. Compared with expert manual segmentations of the combined, femoral and acetabular cartilage volumes, the automatic scheme obtained mean (± standard deviation) Dice's similarity coefficients of 0.81 (± 0.03), 0.79 (± 0.03) and 0.72 (± 0.05). The corresponding mean absolute volume difference errors were 8.44% (± 6.36), 9.44% (± 7.19) and 9.05% (± 8.02). The mean absolute differences between manual and automated measures of cartilage thickness for femoral and acetabular cartilage plates were 0.13 mm (± 0.12) and 0.11 mm (± 0.11), respectively.
    Full-text · Article · Nov 2014 · Physics in Medicine and Biology
Show more