X-Ray Image Processing Using Level Set Segmentation and Filtering Techniques

X-Ray Image Processing Using Level Set Segmentation and Filtering Techniques

1. 1Shweta Gupta

   Research scholar, School of Engineering and Technology Ansal University, Gurgaon-122003, Haryana (India)

   2 Dr. Alok K . K ushwaha

   Professor, School of Engineering and Technology Ansal University, Gurgaon-122003, Haryana (India)

   Abstract

   This paper aims at studying the level set segmentation technique using Variational Level Set Formulation techniques without reintialisation with various filtering methods applied on X-Ray images and analyzing the results obtained after applying various filters to the segmented images. The various steps taken in the development of the program and then the testing of the simulation program with various biomedical are described and the test samples are obtained from set of X- Ray images using MATLAB simulation programs. With the comparison of various filtering techniques on the images sets, it is found that maximum filter provides the best results on the samples of the segmentation of X-ray images.

   Keywords: Level Set Segmentation, Reintialisation, X-Ray, Filtering.

   1. INTRODUCTION – Variational Level Set Formulation of curve evolution without re- initialization

      Re-initialization has been extensively used as a numerical remedy in traditional level set methods. The standard re-initialization method is to solve the following reintialisation equation:

      = sign (0) (1-| |) (1) Where 0 is the function to be re-initialized, and sign is the sign function. But problem is there if 0 is not smooth or 0 is much steeper on one side of the interface than the other, the zero level set of the resulting function 0 can be moved incorrectly from that of the original function. For removing this limitation we use new approach of Variational Level Set Formulation of Curve Evolution without Re-initialization [1], [2]. The evolving level set function can deviate greatly from its value as signed distance in a small number of iteration steps, especially when the time step is not chosen small enough. So far, re-initialization has been extensively used as a numeric remedy for maintaining stable curve evolution and ensuring desirable results but re-initialization process is quite complicated, expensive and has subtle side effects. In Variational level set formulation, the level-set are dynamic curves that move toward the object boundaries. Therefore we define an external energy that can move towards the edges. If I be the image, then edge indicator function (g) is defined by:

      (2)

      Where, G – Gaussian kernel with standard deviation , we define an external energy for a function (x, y) as below:

      Eg, , () = Lg () + Ag () (3)

      Where, > 0 and are constants, and the terms Lg ) and Ag ) are defined by

      Lg () = g () || dx dy (4) Ag () = gH (-) dx dy (5)

      Respectively, where is the unvariate Dirac Function, and H is the Heaviside Function. Now, the following total energy functional.

      E () = µP () + Eg, , () (6)

      The external energy Eg, , drives the zero level set towards the object boundaries, while the internal energy P () penalizes the deviation of from a signed distance function during its evolution which is give in equation given below:

      P () = (||-1) 2 dx dy (7)

      The variational formula derives from the penalize energy equation:

      E () = P () + Em (8)

      Where, > 0 is a parameter controlling the effect of penalizing the deviation of from a signed distance function, and Em ) is a certain energy that would drive the motion of the zero level curve . The energy functional Ag ) introduced to speed up curve evolution. The coefficient of Ag can be positive or negative, depending on the relative position of the initial level-set to the object of interest. If the initial level-sets are placed inside the object, the coefficient should take negative value to speed up the expansion of the level-sets. By calculus of variations, the Gateaux derivative of the functional E in can be written as-

      = -µ [-div )] – () div (g) – g () (9)

      Where, is the Laplacian operator, Therefore, the function that minimizes this functional satisfies the

      Euler-Lagrange equation = 0. The gradient flows of the energy function Lg ) and Ag ), are responsible of driving the zero level curve towards the object boundaries. So this new approach of level-sets is tested on medical images like X-Ray, CT and MRI. It

      shows good result on medical images even on more noisy images. But one problem is there that is we have to make the level-set optimized to the particular image, and if images changes than topology has to change by user itself[5].

      Figure 1.2: Evolution of zero level curve of the corresponding level set function

      Figure1.3: (a) Original X-Ray image

      (b) Segmentation of X-Ray image

      1. Implementation of Algorithm and Simulation

         The algorithm was originally developed by Chumming Li [1] for his MATLAB code for level-set without re- initialisation. However the algorithm is complicated, expensive to implement and images result as obtained are also not smooth. The modified steps include specialized filtering methods used at various levels of image processing.

         Step 1: Image acquiring and reading

         Step 2: Processing the image through desired filter Step 3: Processing the image through Gaussian Filter

         Step 4: Select the region of interest from the input image

         Step 5: Finding the gradient of the image Step 6: Set the parameter of level-set Step 7: Set the intensity of the image

         Step 8: Segmentation of image by Level set method

      2. Following changes have been incorporated in the algorithm:

   1. The different filters were used to filter the image. The filters were used before the Gaussian filter. This technique is used to modifying or enhancing an image. It helps to emphasize certain features or remove other features of the image .It smoothness, sharpening and

      enhancement the edge of the image. These filtering techniques include Linear filtering and Non Linear Filtering [3] [4].

   2. To increase the intensity of the image the parameters controlling the intensity has been adjusted.

   3. Now call the selection base program. This program chooses the appropriate values of level-set parameters form its database according to the if-else rules .The output of this control strategy are the input for the segmentation program. The parameters alfa, lamda, sigma, epsilon are the optimized parameter for the level set program.

   4. The height, length and the area of the segmented part can be calculated. Segmented-area equals to the area of the closed curve when it is in anti-clockwise and equals to the negative area when it is in clockwise. Negative area means equal to area in magnitude but negative in sign. It used to judge the direction of a closed curve. C provides the coordinates of the nodes of the curve Area

      = varea (C); Area returns the area of the curve (>0) when it is in anti-clockwise and negative area of the curve (<0) when it is in clockwise.

   5. Next step is to calculate SNR, PSNR, WPSNR and Entropy parameters of the filtered image and the original image.

   2. TEST RESULTS

   X-Ray image shown for Level Segmentation Using different filters for obtaining various parameters –

   Dimension of segmented area when no filter is used: Width= 43.4 pixels; Height= 51.7 pixels;

   Area of closed curve = 4.009e+003.

   Tpe of Filter	Images	Widt h	Hei ght	Area of segmen ted part	Result
   Maxi					The area
   mum					of
   filter					segmentati
   		41.49	49.	3.7256e	on has
   			03	+003	increased
   					and nearly
   					entire
   					desired
   					area is
   					segmented
   					.
   Media					The area
   n filter					of
   					segmentati
   		37.59	49.	2.9039e	on has
   			01	+003	increased
   					but fails to
   					segment
   					the entire
   					desired
   					area.

   Unsha					The area
   rp					of
   filter					segmentati
   		27.52	31.	2.0053e	on is very
   			10	+003	small
   					therefore
   					this filter
   					is not
   					suitable
   					for such
   					type of
   					segmentati
   					on.
   Disk					The area
   filter					of
   					segmentati
   		39.09	44.	3.3242e	on has
   			50	+003	increased
   					but filter
   					fails to
   					segment
   					the desired
   					area.
   Motio					By the use
   n filter					of this
   					filter the
   		28.28	25.	1.5961e	area of
   			50	+003	segmentati
   					on has
   					decreased.
   Gaussi					This filter
   an					fails to
   filter					segment
   		29.73	38.	2.1387e	the entire
   			28	+003	desired
   					part.

   Minim um filter		30.91	39. 16	2.2367e +003	This filter fails to segment the entire desired area.
   Log					This filters
   filter					segment
   					very less
   		30.15	32.	1.7580e	area
   			06	+003	therefore
   					the filter is
   					not
   					suitable
   					for such
   					type of
   					segmentati
   					on
   Avera					The area
   ge					of
   filter					segmentati
   		27.31	38.	2.2625e	on has
   			47	+003	increased
   					but filter
   					fails to
   					segment
   					the desired
   					area.
   Laplac					The filters
   ian					crosses the
   filter					boundaries
   		85.02	109	1.3723e	of
   			.09	+003	segmented
   					area
   					therefore it
   					is not
   					suitable
   					for such
   					type of
   					segmentati
   					on
   Prewit					The area
   t filter					of
   					segmentati
   		25.48	21.	1.3520e	on has
   			36	+003	increased
   					but filter
   					fails to
   					segment
   					the desired
   					area.
   Sobel					The area
   filter					of
   					segmentati
   		22.39	27.	1.3534e	on is very
   			35	+003	small
   					therefore
   					the filter is
   					not
   					suitable
   					for such
   					type of
   					segmentati
   					on.

   Minim um filter		30.91	39. 16	2.2367e +003	This filter fails to segment the entire desired area.
   Log					This filters
   filter					segment
   					very less
   		30.15	32.	1.7580e	area
   			06	+003	therefore
   					the filter is
   					not
   					suitable
   					for such
   					type of
   					segmentati
   					on
   Avera					The area
   ge					of
   filter					segmentati
   		27.31	38.	2.2625e	on has
   			47	+003	increased
   					but filter
   					fails to
   					segment
   					the desired
   					area.
   Laplac					The filters
   ian					crosses the
   filter					boundaries
   		85.02	109	1.3723e	of
   			.09	+003	segmented
   					area
   					therefore it
   					is not
   					suitable
   					for such
   					type of
   					segmentati
   					on
   Prewit					The area
   t filter					of
   					segmentati
   		25.48	21.	1.3520e	on has
   			36	+003	increased
   					but filter
   					fails to
   					segment
   					the desired
   					area.
   Sobel					The area
   filter					of
   					segmentati
   		22.39	27.	1.3534e	on is very
   			35	+003	small
   					therefore
   					the filter is
   					not
   					suitable
   					for such
   					type of
   					segmentati
   					on.

   Table 1.1: Dimensions of segmented area of X-Ray image with various filters

   Images	SNR	PSNR	WPSNR	ENTROPY
   Maximum filter	14.0377	14.5137	8.4931	4.5435
   Median filter	11.2798	11.7558	5.7352	4.4324
   Minimum filter	9.9093	10.3853	4.3647	4.4320
   Log filter	1.8548	4.3814	1.0428	3.2215
   Average filter	10.4712	12.9998	10.2184	4.5206
   Laplacian filter	1.8044	4.3330	1.1826	2.9717
   Prewitt filter	2.4346	4.9632	1.5869	3.0008
   Sobel filter	2.4507	4.9793	1.5670	3.1439
   Unsharp filter	13.1962	15.7248	9.4553	4.4683
   Disk filter	8.5383	11.0669	8.0045	4.6513

   Motion filter	9.6280	12.1566	9.1290	4.5584
   Gaussian filter	11.6451	14.1737	17.8616	4.5320

   Motion filter	9.6280	12.1566	9.1290	4.5584
   Gaussian filter	11.6451	14.1737	17.8616	4.5320

   BIOGRAPHIES

   Table 1.2: Parameters of X Ray image of various filters

   With results in Table 1.1 and Table 1.2, it was found that the Maximum Filter did the best job as far as segmentation of X-Ray Images is concerned as application of Maximum Filter resulted in increasing the area of segmentation which is nearly equal to the desired area of sample with best SNR, PSNR, WPSNR and stable entropy parameter.

   CONCLUSION

   The process of segmentation of biomedical images requires a very high degree of accuracy. A similar effort has been made in this work to analyze the various filtering techniques and to find out the best among the twelve filters. The setup has been tested for a given set of biomedical images such as X-Ray images and can also be used for CT and MRI images. In the process of final evaluation, we found that the results using the variational level set segmentation techniques on X-Ray images are better. Out of the twelve filters used, maximum filter did the best job as far as segmentation of X-Ray images is concerned. It may be concluded that the algorithm applied has been by and far successful.

   REFERENCES

   1. Chunming Li , Chenyang Xu and Changfeng Gui, (2005): Level set evolution without re-initialization: a new variational formulation, IEEE Computer Conference on Computer Vision and Pattern Recognition, Vol:1, pp: 430-436.

   2. Shaojun Liu and Jia Li, (2006): Automatic Medical Image Segmentation Using Gradient and Intensity Combined Level Method, IEEE Annual International Conference on Medical Imaging, pp: 78-82.

   3. Xujia Qin, Jionghui Jiang, Weihong Wang and Fan Zhang, (2007): Canny Operator Based Level Set Segmentation Algorithm for Medical Images, International Conference on Bioinformatics and Biomedical Engineering, pp: 892-895.

   4. Gang Chen, Lixu Gu, Lijun Qian and Jianrong Xu, (2009): An Improved Level Set for Liver Segmentation and Perfusion Analysis in CT, IEEE Transactions on Information Technology in Biomedicine, Vol: 13, pp: 94-103.

   5. Annangi P,Thiruvenkadam S and Raja A, (2010): A region based active contour method for x-ray lung segmentation using prior shape and low level features

,IEEE international symposium on biomedical imaging,pp:892-895

Mrs. Shweta Gupta is working as Assistant Professor in School of Engineering and Technology, Ansal University, Gurgaon. She has been in the field of education since more than 4 years. She has handled diverse set of curriculum in the field of Engineering and technology. She is an Electrical Engineering Graduate from MBS College, Jammu and M.E in Instrumentation & Control from Thapar University. She has presented various papers in National and International conferences and has published papers in several journals. She is currently pursuing her PhD from Ansal University.