Where is Wally - user study af Bettina Ridler og Helle Mortensen, GN Netcom
Binarization in Magnetic Resonance Images (MRI) …T using Ridler calvard method. It is an iterative...
4
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. CCSEIT-12, October 26-28, 2012, Coimbatore [Tamil nadu, India] Copyright © 2012 ACM 978-1-4503-1310-0/12/10…$10.00. Binarization in Magnetic Resonance Images (MRI) of Human Head Scans with Intensity Inhomogeneity K. Somasundaram 1 Image Processing Lab, Department of Computer Science and Applications, Gandhigram Rural Institute, Gandhigram, Tamil Nadu, India. 91-9865005751 [email protected] T. Genish 2 Image Processing Lab, Department of Computer Science and Applications, Gandhigram Rural Institute, Gandhigram, Tamil Nadu, India. 91-8754246272 [email protected] ABSTRACT Medical image segmentation plays a crucial role in identifying the shape and structure of human anatomy. The most widely used image segmentation algorithms are edge-based and typically rely on the intensity homogeneity of the image at the edges, which often fail to provide accurate segmentation results due to the intensity inhomogeneity. This paper proposes a boundary detection technique for segmenting the hippocampus (the subcortical structure in medial temporal lobe) from MRI with intensity inhomogeneity without ruining its boundary and structure. The image is pre-processed using a noise filter and morphology based operations. An optimal intensity threshold is then computed. We have used mean, top-hat and bottom hat filters for noise removal and Ridler Calvard method to compute the threshold value. Our method has been validated on human brain sagittal MRI, with desirable performance in the presence of intensity inhomogeneity. The proposed method works well even for weak edge. Experimental results show that our method can be used to detect boundary for accurate segmentation of hippocampus. The proposed method takes no more than 2 seconds for boundary detection. Keywords Segmentation, intensity inhomogeneity, hippocampus, morphological operations, thresholding, MRI. 1. INTRODUCTION Intensity inhomogeneity [1] often occurs in real-world images due to various factors, such as spatial variations in illumination and imperfections of imaging devices, which complicates many problems in image processing and computer vision. In particular, image segmentation may become considerably difficult for images with intensity inhomogeneity overlapping with the ranges of the intensities in the regions to be segmented. This makes it impossible to identify these regions based on the pixel intensity. The widely used image segmentation algorithms [2] [3] usually rely on intensity homogeneity, and therefore are not applicable to images with intensity inhomogeneity. In general, intensity inhomogeneity has been a challenging task in image segmentation. One of the important stages in medical image analysis is segmentation of objects or identification of their contours. Medical images are relatively hard to segment due to several undesirable properties such as low signal-to-noise and contrast-to- noise ratios, and discontinuous edges. Since segmentation of medical images is a significant task, it is essential for medical image analysis. The segmentation of the hippocampus (critical structure in human limbic system responsible for memory forming, organizing, and storing) is a difficult task [4] [5] due to its small size, low contrast, and apparent discontinuity of the edges. Some previous work that has utilised manual thresholding [6] [7], trilinear interpolation with thresholding [8] [9], image intensity standardization for correcting acquisition-to-acquisition signal intensity variations [10] [11] [12], homomorphic methods [13] [14], curvelet and wavelet based methods [15] [16] [17] [18] are fast and easy adaptable methods. However such methods sometimes sensitive to small data variations and it is difficult to find the optimum threshold value for segmenting the hippocampus from brain regions. In order to overcome this problem, the image is first pre-processed by mean filter followed by grey-scale morphological operations to remove or subdue the background noise. Since mean filter is consistently more effective than median-based algorithms [11] for removing intensity inhomogeneity in MR images, it is used in our proposed scheme. The filtered image is then binarized using Ridler’s method. Application of our method on few MRI of human head scans show that our approach gives improved results. The remaining paper is organized as follows. In Section 2, the proposed scheme is explained. In Section 3, imaging protocol is given. In section 4, we discuss the benefits and limitations of our approach, as well as possibilities for future improvements. In Section 5, the conclusion is given. 2. PROPOSED METHOD The first step in the proposed method is to remove the noise. We apply mean filter to the input image to preserve edges while removing noise. We consider a small mask of size 3x3 pixels and the computation is made from top-left to bottom-right of an image to remove the background noise in the MRI. Then grey-scale morphology based top-hat and bottom-hat filtering is used to smooth the pixel intensities of an image to correct uneven background and gray-level feature extraction. Generally 244
Transcript of Binarization in Magnetic Resonance Images (MRI) …T using Ridler calvard method. It is an iterative...
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. CCSEIT-12, October 26-28, 2012, Coimbatore [Tamil nadu, India]
Copyright © 2012 ACM 978-1-4503-1310-0/12/10…$10.00.
Binarization in Magnetic Resonance Images (MRI) of Human Head Scans with Intensity Inhomogeneity
K. Somasundaram1
Image Processing Lab, Department of Computer
Science and Applications, Gandhigram Rural Institute,
Gandhigram, Tamil Nadu, India.
91-9865005751
T. Genish2
Image Processing Lab, Department of Computer
Science and Applications, Gandhigram Rural Institute,
Gandhigram, Tamil Nadu, India.
91-8754246272
ABSTRACT
Medical image segmentation plays a crucial role in identifying the shape and structure of human anatomy. The most widely used image segmentation algorithms are edge-based and typically rely on the intensity homogeneity of the image at the edges, which often fail to provide accurate segmentation results due to the intensity inhomogeneity. This paper proposes a boundary detection technique for segmenting the hippocampus (the subcortical structure in medial temporal lobe) from MRI with intensity inhomogeneity without ruining its boundary and structure. The image is pre-processed using a noise filter and morphology based operations. An optimal intensity threshold is then computed. We have used mean, top-hat and bottom hat filters for noise removal and Ridler Calvard method to compute the threshold value. Our method has been validated on human brain sagittal MRI, with desirable performance in the presence of intensity inhomogeneity. The proposed method works well even for weak edge. Experimental results show that our method can be used to detect boundary for accurate segmentation of hippocampus. The proposed method takes no more than 2 seconds for boundary detection.
Keywords Segmentation, intensity inhomogeneity, hippocampus, morphological operations, thresholding, MRI.
1. INTRODUCTION Intensity inhomogeneity [1] often occurs in real-world images due to various factors, such as spatial variations in illumination and imperfections of imaging devices, which complicates many problems in image processing and computer vision. In particular, image segmentation may become considerably difficult for images with intensity inhomogeneity overlapping with the ranges of the intensities in the regions to be segmented. This makes it impossible to identify these regions based on the pixel intensity.
The widely used image segmentation algorithms [2] [3] usually rely on intensity homogeneity, and therefore are not applicable to
images with intensity inhomogeneity. In general, intensity
inhomogeneity has been a challenging task in image segmentation.
One of the important stages in medical image analysis is segmentation of objects or identification of their contours. Medical images are relatively hard to segment due to several undesirable properties such as low signal-to-noise and contrast-to-noise ratios, and discontinuous edges. Since segmentation of medical images is a significant task, it is essential for medical image analysis. The segmentation of the hippocampus (critical structure in human limbic system responsible for memory forming, organizing, and storing) is a difficult task [4] [5] due to its small size, low contrast, and apparent discontinuity of the edges. Some previous work that has utilised manual thresholding [6] [7], trilinear interpolation with thresholding [8] [9], image intensity standardization for correcting acquisition-to-acquisition signal intensity variations [10] [11] [12], homomorphic methods [13] [14], curvelet and wavelet based methods [15] [16] [17] [18] are fast and easy adaptable methods. However such methods sometimes sensitive to small data variations and it is difficult to find the optimum threshold value for segmenting the hippocampus from brain regions. In order to overcome this problem, the image is first pre-processed by mean filter followed by grey-scale morphological operations to remove or subdue the background noise. Since mean filter is consistently more effective than median-based algorithms [11] for removing intensity inhomogeneity in MR images, it is used in our proposed scheme. The filtered image is then binarized using Ridler’s method. Application of our method on few MRI of human head scans show that our approach gives improved results. The remaining paper is organized as follows. In Section 2, the proposed scheme is explained. In Section 3, imaging protocol is given. In section 4, we discuss the benefits and limitations of our approach, as well as possibilities for future improvements. In Section 5, the conclusion is given.
2. PROPOSED METHOD The first step in the proposed method is to remove the noise. We apply mean filter to the input image to preserve edges while removing noise. We consider a small mask of size 3x3 pixels and the computation is made from top-left to bottom-right of an image to remove the background noise in the MRI. Then grey-scale morphology based top-hat and bottom-hat filtering is used to smooth the pixel intensities of an image to correct uneven background and gray-level feature extraction. Generally
244
morphological operations are used for binary images to expand or to shrink. But, here we consider morphological operations on grey-scale images to show the darker regions as lighter and vice-versa. After performing morphological top-hat and bottom-hat filtering, the region of interest (hippocampus) appeared as a darker part. The boundary of the region does not get affected. The flow chart of the proposed scheme is shown in Fig. 1. The top-hat filtering is given as:
Itop = f - ( A ° B ) (1)
Similarly, the bottom-hat filtering is given as:
Ibot = f - ( A • B ) (2)
where, f is an original image, A ° B and A • B are the morphological opening and morphological closing of an image A by a structuring element B. Since the appearance of the hippocampus is small, structuring element B of size 3x3 is used. The filtered image is then obtained by:
IF = ( f + Itop ) - Ibot (3)
Fig. 1. Flow chart of the proposed method
The image IF is then binarized by computing a threshold value T using Ridler calvard method. It is an iterative method where, the initial threshold T is computed as:
(4)
where, is the intensity of ith pixel and varies from 0 to 255 in the input image and N is the total number of pixels in the image IF
.By using eqn. (4), the pixel intensities of the image is separated in to two modes M1 and M2 as:
IF (xi) = M1 if xi >= T1
= M2 otherwise (5)
Now, the new threshold T is computed as follows:
(6)
where, C1 is the number of pixels in M1 and C2 is the number of pixels in M2. From the eqn. (6), the given input image IF (x ,y) is converted to a binary image g as:
g ( x , y) = 1 if IF ( x , y ) > = T
= 0 otherwise (7)
3. RESULT AND DISCUSSIONS We have used T1 sagittal dataset collected from the Whole Brain Atlas (WBA) [19] under the volume of Normal Anatomy in 3-D with MRI/PET. This volume constitutes 127 slices of thickness 3 mm each. Among them, we have used 8 slices numbered from 43-50 for the research because the hippocampus is appeared only on that mid frontal slices.
To evaluate the performance of our method, we carried out experiments by applying our method on the material pool. The result obtained using the proposed method is shown in Fig. 2.
By comparing the Column 1 and Column 3, we observe that the pre-processed images can be made use to identify true edges when the intensity inhomogeneity image is concerned. Since the edge plays a crucial role in image segmentation, the proposed scheme can be used as a pre-processing technique to detect the edges precisely. This pre-processing result would also be helpful for detecting the edges of tedious subcortical structures such as hippocampus, amygdala and dentate gyrus but, the proposed method focuses only on hippocampus. After applying top-hat and bottom-hat filtering, the hippocampal structure and its corresponding boundary become darker and hence, we can trace out its edge easily.
The 3 x 3 mask is considered in the morphological filtering since the shape of the hippocampus resembles an oval structure. The binarized image produces satisfactory result and shows that hippocampus region does not merge or join with unwanted adjacent regions in most of the cases.
End
Start
Input slice
Apply mean filter
Morphological top-hat and bottom-hat filtering
Extracted binary image
Image binarization using Ridler Calvard method
245
(a) (b) (c) (d)
Fig.2. Image binarization (a) Original image (b) Image obtained after mean filtering (c) Image obtained after morphological filtering (d) Binarized image using Ridler Calvard method.
Our experiment also shows that, the Ridler Calvard method is better than most efficient binarization techniques like mean thresholding, Otsu’s thresholding techniques.
The experiment has been done only on T1 sagittal images. The proposed method can be improved to process coronal and axial images also. The edges identified by our method can be used as a shape prior to develop an automatic tool for extracting hippocampus from human MRI in future.
4. CONCLUSION In this paper, we have proposed a binarization technique for correcting intensity inhomogeneity in MR images to remove the artifacts. The combination of morphological top-hat and bottom-hat filtering operations help to find edges sharply.
5. ACKNOWLEDGMENTS This work is supported by a research grant by University Grants Commission, New Delhi, India. Grant No: M.R.P, F.No-37- 154/2009(SR).
6. REFERENCES
[1] Chunming Li, Huang, R. Ding, Z. Chris Gatenby, J. and Metaxas, D.N. A Level Set Method for Image Segmentation in the Presence of Intensity In homogeneities With Application to MRI. IEEE Transactions on Image
Processing. 20 (2011), 2007-2016.
[2] Chan, T. and Vese, L. Active contours without edges. IEEE
Transactions on Image Processing. 10 (2001), 266–277.
[3] Ronfard R. Region-based strategies for active contour models. International Journal of Computer Vision.13 (1994), 229–251.
[4] Shen, D. Susan, M.S. Resnick, M. and Davatzikos, C. Measuring Size and Shape of the Hippocampus in MR Images Using a Deformable Shape model. NeuroImage. 15 (2002), 422-434.
[5] Ghanei, A. Zadeh, H.S. and Windham, J.P. Segmentation of the hippocampus from brain MRI using deformable contours. Computerized Medical Imaging and Graphics. 22 (1998), 203-216.
[6] Freeborough, P.A. Fox, N.C. and Kitney, R.I. Interactive algorithms for the segmentation and quantitation of 3-D MRI brain scans. Computer Methods and Programs in
Biomedicine. 53 (1997), 15-25.
[7] Boyes, B.J. Lewis, R.G. Schott, E.B. Frost, J.M.C. Scahill, R.I and Fox, N.C. Automatic calculation of hippocampal atrophy rates using a hippocampal template and the boundary shift integral. Neurobiology of Aging. 28 (2007), 1657-1663.
[8] Sun, T. and Neuvo, Y. Detail-preserving median based filters in image processing. Pattern Recognition Letters. 15 (1994), 341-347.
[9] Brinkmann, B.H. Manduca, A. and Robb, R.A. Optimized homomorphic unsharp masking for MR gray scale inhomogeneity correction. IEEE Transactions on Medical
Imaging. 17 (1998), 161-171.
[10] Somasundaram, K. and Kalaiselvi, T. Fully automatic brain extraction algorithm for axial T2-weighted magnetic resonance images. Computers in Biology and Medicine. 40 (2010), 811-822.
[11] Heckemann, R.A. Hajnal, J.V. Aljabar, P. Rueckert, D. and Hammer, A. Automatic anatomical brain MRI segmentation combining label propagation and decision fusion. NeuroImage. 33 (2006), 115-126.
[12] Thillou, C. and Gosselin, B. Robust Thresholding Based on Wavelets and Thinning Algorithms for Degraded Camera Images. Proceedings of ACIVS. 2004 - tcts.fpms.ac.be.
[13] Sonka, M. Hlavac, Boyle. Digital Image Processing and
Computer Vision. Thomson Learning Inc., Second Edition, 2007.
[14] Somasundaram, K. and Kalavathi, P. Medical Image Binarization Using Square Wave Representation.
Communications in Computer and Information Science.
140 (2011), 152-158.
246
[15] Starck, J.L. Candes, E.J. and Donoho, D.L. The curvelet transform for image denoising. IEEE Transactions on Image
Processing.11 (2002), 670-684.
[16] Portilla, J. Strela, V. Wainwright, M.J. and Simoncelli, E.P.
Image denoising using scale mixtures of Gaussians in the wavelet domain. IEEE Transactions on Image Processing.
12 (2003), 1338-1351.
[17] Saeedi, J. Moradi, M.H. and Faez, K. A new wavelet-based fuzzy single and multi-channel image denoising. Image and
Vision Computing. 28 (2010), 1611-1623.
[18] Huang, K. Wu, Z.Y. Fung, S.K. and Chan, H.Y. Color image denoising with wavelet thresholding based on human visual system model. Signal Processing: Image Communication. 20 (2005), 115-127.
[19] http://www.med.harvard.edu/aanlib/cases/caseNA/pb9.htm.
247
