M. Kamber, R. Shinghal, D.L. Collins, G.S. Francis, and A.C. Evans. Model-based 3-D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Trans. Medical Imaging, 14: 442-453, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....region growing and constraining its boundary by an elastically registered anatomical atlas. This is used to make the white matter mask containing lesions and white matter that can be separated by a simple intensity based classifier. Similar work can be found for segmenting tumors in Kamber et al. [2]. Another way to classify lesions is using a feature space collected from possible candidate lesions to sort out false positives. Ardizzone [3] uses the fuzzy c means algorithm by first obtaining a set of over segmented regions followed by a reclustering phase. The re clustering uses shape and ....

Kamber M, Shinghal R, Collins L, et al.: Model-based 3D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Trans Med Imaging 14(3): 442-453, 1995.

....However, both authors report failures on atypical (significantly di#erent than the atlas) brain scans, such as child or pathological brains. The use of stereotaxic space tissue probability maps for automating supervised classification algorithms was originally proposed by Kamber [6], and subsequently used by other researchers [1, 7] The maps are used to select training samples from spatial locations that are very likely to contain a given tissue type. This approach s limitations are described in the next section. The main contribution of this paper is a novel method for ....

....for that particular population. Once imaging data is spatially registered (normalized) to the stereotaxic space, TPM s provide an a priori spatial probability distribution for each tissue (Fig. 1) This distribution can be used to automatically produce a training set for the supervised classifier [6]: for example, choose spatial locations that have a TPM value # = 0.99 (99 ) The lower the # , the more qualifying spatial locations there will be. However, this simplistic approach has two limitations: 1. Mis labeled samples ( false positives ) Since the morphology of the human brain is so ....

Kamber, M., et al. : Model-based 3-D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Trans. on Medical Imaging 14 (1995) 442--53

.... Such classi cation techniques are, in fact, not medical imaging speci c an extensive coverage of classi ers is given in [28] and some selected topics are also given in [26, 27] Many researchers have applied to brain MRI classic methods such as the Bayes (maximum likelihood) classi er [19, 40], or non parametric classi ers like kNN (k nearest neighbors) 74] and ANN (arti cial neural network) 77] Expectation Maximization (EM) is a popular statistical classi cation scheme for this segmentation application; originally proposed (in a brain MRI context) by Wells [75] and further ....

....The method requires typical cluster mean and variance values; however, these values are scanner, site, and even pulse sequence, speci c. The use of stereotaxic space tissue probability maps (TPM s) for automating supervised classi cation algorithms was originally proposed by Kamber [40], and subsequently used by other researchers [42, 77] The TPM is used to select training samples from spatial locations that are very likely to contain a given tissue type. However, this approach is very sensitive to any deviations of the subject s anatomy from the statistical model represented ....

[Article contains additional citation context not shown here]

M. Kamber, R. Shinghal, D. L. Collins, G. S. Francis, and A. C. Evans. Modelbased 3-D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Transactions on Medical Imaging, 14(3):442-53, Sept. 1995.

.... to the lack of contrast between normal and pathologic tissue [15, 16] statistical classification may not differentiate between non enhancing tumor and normal tissue [12, 13, 14] Explicit anatomical information derived from a digital atlas has been used to identify normal anatomical structures [9, 10, 11]. We have developed an automated segmentation tool which can identify the skin surface, the ventricles, the brain and tumor in patients with brain neoplasms [18, 19] The purpose of the current study was to compare the accuracy and reproducibility of this automated method with those of manual ....

Kamber M, Shinghal R, Collins DL, et al. Model-based 3-D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE T Med Imaging 1995; 14(3):442-453.

....difficulties in representing possible anatomical variations and voxel classifications are typically derived from a single cadaver. Some methods for segmenting brain MR images have attempted to overcome this by using multiple subjects to create probabilistic spatial distributions of normal anatomy [37,38]. In our system, anatomical knowledge is stored in a declarative model. For each anatomical structure, parametric shape and relational attributes are encapsulated in a frame [39] The frames have a predefined set of slots , corresponding to the attributes, in which parameter values are stored. ....

Kamber M, Shinghal R, Collins DL, Francis GS, Evans AC. Modelbased 3-D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Transactions on Medical Imaging 1995;14(3):442--453.

....Several approaches have been proposed in the last years: Zijdenbos [15] proposes a neural approach to segment brain in five classes: background, white matter, gray matter, CSF and white matter lesions (WML) by means of a multi layer perceptron trained with the back propagation algorithm. Kamber [11] and Johnston [10] propose stochastic approaches to the 3D segmentation of the brain. In [11] a probabilistic model providing the a priori probability of the tissues distribution is derived from the analysis of several volunteers slices, and Bayesian classification is performed on 3D data sets ....

....approach to segment brain in five classes: background, white matter, gray matter, CSF and white matter lesions (WML) by means of a multi layer perceptron trained with the back propagation algorithm. Kamber [11] and Johnston [10] propose stochastic approaches to the 3D segmentation of the brain. In [11] a probabilistic model providing the a priori probability of the tissues distribution is derived from the analysis of several volunteers slices, and Bayesian classification is performed on 3D data sets that have been previously affine transformed in order to normalize them. In [10] a stochastic ....

Kamber, M., Shinghal, R., Collins, D.L., Francis, G.S., Evans, A.C., Model-Based 3-D Segmentation of Multiple Sclerosis Lesions in Magnetic Resonance Brain Images, IEEE Transactions on Medical Imaging, vol. 14, no. 3, pp. 442/453, September 1995.

....3 D segmentation in [10] employs a 3 D connectivity algorithm, after appropriate thresholding. Mathematical morphology [11] was used in 3 D domain for segmenting pulmonary trees [12] and brain tissue [13] Various model based supervised classi ers have been tested in segmenting 3 D brain images in [14, 15]. Each region is associated with a multivariate Gaussian mixture density in [15] 3 D modeling from range images using self organizing maps [16] was employed in [17] Radial Basis Functions (RBF s) were used for 3 D iterative image reconstruction from projection data in [18] and in 3 D shape from ....

M. Kamber, R. Shinghal, D. L. Collins, G. S. Francis, A. C. Evans, \Model-based 3-D segmentation of multiple sclerosis lesions in magnetic resonance brain images," IEEE Trans. on Medical Imaging, vol. 14, no. 3, pp. 442-453, Sep. 1995.

....lesions in large number of series of 2 D slices is not only time consuming but also tedious and error prone. Errors for the segmentation of small structures are often in the range of the volume of the observed structures. Automated image segmentation systems have been proposed by several groups [4 7]. They consist of well designed sequences of processing steps, including preprocessing, biasfield correction, feature space clustering of multi echo MRI data [8] and a matching of a statistical anatomical atlas [9, 10] to solve ambiguities of statistical classification. As a result, they present ....

M. Kamber, R. Shinghal, D.L. Collins, G.S. Francis, and A.C. Evans, Model-based 3-D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Transactions in Medical Imaging, 14(3):442-453, Sept. 1995

....[16] or statistical classification methods [2] work well in some cases but may not differentiate between active tumor, associated pathology and normal tissue. Template based segmentation methods solve the segmentation problem by aligning a digital atlas of a normal brain to the individual [3, 9, 12]. The anatomical knowledge represented in the atlas then may serve as a lookup map. However, these methods rely on the correctness of the alignment, and, by definition, normal digital brain atlases don t include pathologic structures, which poses a problem if used for the segmentation of ....

M. Kamber, R. Shinghal, D.L. Collins, G.S. Francis and A.C. Evans: ModelBased 3-D Segmentation of Multiple Sclerosis Lesions in Magnetic Resonance Brain Images. IEEE Transactions on Magnetic Resonance Imaging, 14(3):442-- 453, 1995.

....perform good quantitative studies, the regions of interest must be well defined. In traditional methods, a radiologist manually outlines the region of interest using a mouse or cursor. More recently, computer assisted methods have been used [6] 32] some of these methods are surveyed recently in [14]. We describe here our new fully automatic method for segmenting the brain from the head in MR images. The key to any automatic method is that it must be robust, so that it produces reliable results on every image acquired from any particular MR scanner using any echo sequence. Our method is so ....

Micheline Kamber, Rajjan Shinghal, D. Louis Collins, Gordon S. Francis, and Alan C. Evans. Model-based 3D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Transactions on Medical Imaging, pages 442--453, 1995.

....Rule based systems allow expert knowledge about image processing be explicitly captured in rules [43] However, parameters in these rules often need to be manually adjusted due to the wide variations of the intensity distribution among different images. Model based system [3] 13] 21] [22] are similar to the rule based system. The external knowledge about the brain structure or physiology are incorporated. But the tissue models are not as explicit as the rules in the rule based system [28] 38] 4 It makes the users difficult to develop the new tissue models for the new. In this ....

Micheline Kamber, Rajjan Shinghal, Louis Collins, Gordon S. Francis, and Alan C. Evans; Model-Based 3-D Segmentation of Multiple Sclerosis Lesions in Magnetic Resonance Brain Images, IEEE Transactions on Medical Imaging, Vol. 14, No. 3, pp 442-453, September, 1995.

....is not required for human analysis of MR scans, phantoms are not always available and other correction methods must be pursued. 3.3.2 Approximate Phantom Correction Axel et al. 5] suggest low pass filtering the MR image to approximate a phantom. Similar methods have been implemented by others [20, 6, 18, 31, 33, 26]. Kamber et al. have shown that this method improves the quality of their MS lesion segmentations [26] However, a quick glance at any signal or image processing text book (see [19] for example) will confirm that, in general: I RF (x; y; z) 6 lpf[I RF (x; y; z) Delta I MR (x; y; z) 3.2) where ....

....3.3.2 Approximate Phantom Correction Axel et al. 5] suggest low pass filtering the MR image to approximate a phantom. Similar methods have been implemented by others [20, 6, 18, 31, 33, 26] Kamber et al. have shown that this method improves the quality of their MS lesion segmentations [26]. However, a quick glance at any signal or image processing text book (see [19] for example) will confirm that, in general: I RF (x; y; z) 6 lpf[I RF (x; y; z) Delta I MR (x; y; z) 3.2) where lpf( denotes the low pass filter operation. Thus, artifacts are injected into the corrected ....

[Article contains additional citation context not shown here]

Micheline Kamber, Rajjan Shinghal, D. Louis Collins, Gordon S. Francis, and Alan C. Evans. Model-based 3D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Transactions on Medical Imaging, 14(3), September 1995. BIBLIOGRAPHY 147

....on which MedGen and MedGenAdjust are based is C4.5 [26] an earlier version of which is known as ID3 [25] C4.5 was chosen because it is generally accepted as a standard for decision tree algorithms, and has been extensively tested. It has been used in many application areas ranging from medicine [18] to game theory [24] and is the basis of several commercial rule induction systems [25] Furthermore, C4.5 allows the use of an attribute selection measure known as information gain [25] In a comparative study of selection measures, information gain was found to produce accurate and small trees ....

M. Kamber, R. Shinghal, D. L. Collins, G. S. Francis, and A. C. Evans. Model-based 3D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Transactions on Medical Imaging, 14(3):442--453, 1995.

No context found.

M. Kamber, R. Shinghal, D.L. Collins, G.S. Francis, and A.C. Evans. Model-based 3-D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Trans. Medical Imaging, 14: 442-453, 1995.

No context found.

Kamber M, Shinghal R, Collins DL, et al. Model-based 3-D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Trans Med Imaging 1995; 14:442--453.

No context found.

Kamber M, Shinghal R, Collins DL, Francis GS, Evans AC. Modelbased 3-D segmentation of multiple sclerosis lesions in magnetic resonance brain images. IEEE Trans Med Imaging 1995;14:442--453.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC