Limits...
Multiple sclerosis lesion detection using constrained GMM and curve evolution.

Freifeld O, Greenspan H, Goldberger J - Int J Biomed Imaging (2009)

Bottom Line: The proposed CGMM-CE algorithm is used to segment 3D MRI brain images with an arbitrary number of channels.The CGMM-CE algorithm is automated and does not require an atlas for initialization or parameter learning.Experimental results on both standard brain MRI simulation data and real data indicate that the proposed method outperforms previously suggested approaches, especially for highly noisy data.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Engineering, Tel-Aviv University, Tel Aviv 69978, Israel. orenf@eng.tau.ac.il

ABSTRACT
This paper focuses on the detection and segmentation of Multiple Sclerosis (MS) lesions in magnetic resonance (MRI) brain images. To capture the complex tissue spatial layout, a probabilistic model termed Constrained Gaussian Mixture Model (CGMM) is proposed based on a mixture of multiple spatially oriented Gaussians per tissue. The intensity of a tissue is considered a global parameter and is constrained, by a parameter-tying scheme, to be the same value for the entire set of Gaussians that are related to the same tissue. MS lesions are identified as outlier Gaussian components and are grouped to form a new class in addition to the healthy tissue classes. A probability-based curve evolution technique is used to refine the delineation of lesion boundaries. The proposed CGMM-CE algorithm is used to segment 3D MRI brain images with an arbitrary number of channels. The CGMM-CE algorithm is automated and does not require an atlas for initialization or parameter learning. Experimental results on both standard brain MRI simulation data and real data indicate that the proposed method outperforms previously suggested approaches, especially for highly noisy data.

No MeSH data available.


Related in: MedlinePlus

Real brain images, example 2. (a) T1,  (b) T2,  (c) PD,  (d) FF,  (e) Manual segmentation of MS lesions (red) overlayed on T1,  (f) CGMM-CE segmentation. Blue: CSF; Green: GM; Yellow: WM; Red: MSL.
© Copyright Policy - open-access
Related In: Results  -  Collection


getmorefigures.php?uid=PMC2742654&req=5

fig9: Real brain images, example 2. (a) T1, (b) T2, (c) PD, (d) FF, (e) Manual segmentation of MS lesions (red) overlayed on T1, (f) CGMM-CE segmentation. Blue: CSF; Green: GM; Yellow: WM; Red: MSL.

Mentions: In what is still a work in progress, the algorithm was also tested on real MRI data. These experiments were performed in a joint effort with the MS Center at Sheba Medical Center, Israel. Here, ground truth segmentation of the healthy tissues is not available. However, we have manual MS lesion segmentation, provided by a human expert. In these experiments, the Fast Flair (FF) modality, was added to T1, T2, and PD which were previously used. Note that the BrainWeb data [33] does not offer this modality. The FF provides a good contrast between CSF and the other healthy tissues (GM and WM). To some extent, it also provides a good contrast between MS lesions and the three healthy tissues. The main purpose of FF in MS is to suppress false positive from CSF signals. Preprocessing steps include conventional procedures such as extraction of the brain region (also know as skull removal); coregistration of the images from the different channels; and bias filed correction. Once the preprocessing is finalized, we extract from each voxel both intensity and spatial features. Unlike the isotropic voxel used in the BrainWeb data experiment, here the voxel size was 1 mm × 1 mm × 3 mm. The set of decision rules for Gaussian detection was similar to the one used in the BrainWeb dataset, with the addition of rules that take the FF into account (e.g., lesions are brighter than all the other tissues). Results on real data are shown in Figures 8 and 9. In both examples it is possible to see a smooth and visually plausible segmentation of the healthy tissues. With regard to the lesion detection and delineation, the main lesions are in fact detected (except for one in Figure 9).


Multiple sclerosis lesion detection using constrained GMM and curve evolution.

Freifeld O, Greenspan H, Goldberger J - Int J Biomed Imaging (2009)

Real brain images, example 2. (a) T1,  (b) T2,  (c) PD,  (d) FF,  (e) Manual segmentation of MS lesions (red) overlayed on T1,  (f) CGMM-CE segmentation. Blue: CSF; Green: GM; Yellow: WM; Red: MSL.
© Copyright Policy - open-access
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC2742654&req=5

fig9: Real brain images, example 2. (a) T1, (b) T2, (c) PD, (d) FF, (e) Manual segmentation of MS lesions (red) overlayed on T1, (f) CGMM-CE segmentation. Blue: CSF; Green: GM; Yellow: WM; Red: MSL.
Mentions: In what is still a work in progress, the algorithm was also tested on real MRI data. These experiments were performed in a joint effort with the MS Center at Sheba Medical Center, Israel. Here, ground truth segmentation of the healthy tissues is not available. However, we have manual MS lesion segmentation, provided by a human expert. In these experiments, the Fast Flair (FF) modality, was added to T1, T2, and PD which were previously used. Note that the BrainWeb data [33] does not offer this modality. The FF provides a good contrast between CSF and the other healthy tissues (GM and WM). To some extent, it also provides a good contrast between MS lesions and the three healthy tissues. The main purpose of FF in MS is to suppress false positive from CSF signals. Preprocessing steps include conventional procedures such as extraction of the brain region (also know as skull removal); coregistration of the images from the different channels; and bias filed correction. Once the preprocessing is finalized, we extract from each voxel both intensity and spatial features. Unlike the isotropic voxel used in the BrainWeb data experiment, here the voxel size was 1 mm × 1 mm × 3 mm. The set of decision rules for Gaussian detection was similar to the one used in the BrainWeb dataset, with the addition of rules that take the FF into account (e.g., lesions are brighter than all the other tissues). Results on real data are shown in Figures 8 and 9. In both examples it is possible to see a smooth and visually plausible segmentation of the healthy tissues. With regard to the lesion detection and delineation, the main lesions are in fact detected (except for one in Figure 9).

Bottom Line: The proposed CGMM-CE algorithm is used to segment 3D MRI brain images with an arbitrary number of channels.The CGMM-CE algorithm is automated and does not require an atlas for initialization or parameter learning.Experimental results on both standard brain MRI simulation data and real data indicate that the proposed method outperforms previously suggested approaches, especially for highly noisy data.

View Article: PubMed Central - PubMed

Affiliation: Department of Biomedical Engineering, Tel-Aviv University, Tel Aviv 69978, Israel. orenf@eng.tau.ac.il

ABSTRACT
This paper focuses on the detection and segmentation of Multiple Sclerosis (MS) lesions in magnetic resonance (MRI) brain images. To capture the complex tissue spatial layout, a probabilistic model termed Constrained Gaussian Mixture Model (CGMM) is proposed based on a mixture of multiple spatially oriented Gaussians per tissue. The intensity of a tissue is considered a global parameter and is constrained, by a parameter-tying scheme, to be the same value for the entire set of Gaussians that are related to the same tissue. MS lesions are identified as outlier Gaussian components and are grouped to form a new class in addition to the healthy tissue classes. A probability-based curve evolution technique is used to refine the delineation of lesion boundaries. The proposed CGMM-CE algorithm is used to segment 3D MRI brain images with an arbitrary number of channels. The CGMM-CE algorithm is automated and does not require an atlas for initialization or parameter learning. Experimental results on both standard brain MRI simulation data and real data indicate that the proposed method outperforms previously suggested approaches, especially for highly noisy data.

No MeSH data available.


Related in: MedlinePlus