Diagnosis of Liver Tumors Using Image Processing

Diagnosis of Liver Tumors Using Image Processing

Wassem Abdulrahman

Dept. Control and Instrumentation Brno University of Technology Brno, Czech Republic

AbstractIn this paper, a computer-aided diagnostic (CAD) system for the diagnosis of benign and malignant liver tumors from computed tomography (CT) images is presented. The research aims to develop an automated way to extract suspicious regions of liver area in the scanner image to the abdominal area, this research uses a new method for extracting the liver area of the scanned image for abdominal area automatically, depending on the use of connected compounds numbering algorithm CCL, in addition to the use of data collection algorithm FCM to extract suspicious regions of the extracted area of the liver.

Keywordsliver; Image processing; Tumors; regions.

1. INTRODUCTION

   The liver has vital has importance to human beings, so liver diseases are considered seriously. The liver is made up of various cell types, so several distinct types of tumors can de- velop in it. These tumors can be benign or malignant (cancer- ous). The benign tumors of the liver seen most frequently include cavernous hemangioma, liver cell adenomas and focal nodular hyperplasia (FNH) .The malignant tumors of the liver seen most frequently are hepatocellular carcinoma, intrahepatic cholangiocarcinoma, bile duct cystadenocarcinoma, and hepatoblastoma [1].

   Computers are used widely in medical research, where there is a vital need forbetter microelectronic sensors for data acquisition. Imaging modalities like Ultrasound, MRI (Magnetic ResonanceImaging), CT (Computed Tomography) and PET (PositronEmission Tomography) are widely used techniques for livercancer tumor diagnosis [2].

   Liver cancer tumor is sixth dangerous diseases in the world. Because of the liver's vital importance to human beings, Liver diseases are considered seriously. There are two classes ofliver tumors: benign and malignant [3].

   Computer-assisted liver tumor classification which is based on the image analysis techniques provides more useful information. The conventional methods for the liver tissue classification consist of three-step process. The first step involves the segmentation of liver and tumor from CT abdominal image. The second step is the feature extraction and the third step is classification using a classifier. The Characterization of liver images based on texture analysis techniques have been developed over the years [4]-[7].

2. DEVELOPED METHOD

   The research aims to develop an automated way to extract suspicious regions of liver area in the scanner image to the

   abdominal area, this research uses a new method for extracting the liver area of the scanned image for abdominal area automatically, depending on the use of connected compounds numbering algorithm CCL, in additionto the use of data collection algorithm FCM to extract suspicious regions of the extracted area of the liver.

   This method is characterized as follows:

   1. Based on fractionation through two phases:

      • Extracting the liver area of the abdominal area scanned image.

      • Extraction of suspicious areas of the liver area.

   2. Use of one cross-sectional image to the abdomen to give a two-dimensional representation of the suspicious blocks.

   3. Depends on a particular situation of scanned images to the abdominal area, where Liver area details are fully clear.

   4. Automated method at all stages and does not require any user intervention.

   The developed method consists of two phases:

   • First Phase :

     Input Image: Scanned Image of the abdomen area as following:

     1. Giving image thresholds values of 90 & 200.

     2. CCL Algorithm applying to number connected areas which form the image.

     3. Specification and extraction the largest region which contains Liver.

     4. Getting rid of other organs areas which appear along with the Liver area.

     5. Retrieval of the original values of the elements of Liver area.

   • Second Phase:

   Getting Output Image(Suspicious Regions image in the Liver area) after:

   1. Applying FCM Algorithm to group elements of Liver image into three sets.

   2. Retrieval of original values of the elements of three sets.

   3. Specification of image which represents the suspicious regions in the Liver area.

   4. Getting rid of regions of small area comparing with the suspicious regions areas.

   5. Retrieval of original values of the suspicious regions elements.

   (A)

   in the search. Figure2 shows one of the images used in the research, medical diagnostics has indicated to the presence of two suspicious regions in left and right lobes, these suspicious regions are indistinguishable by eye due to the difference in grey levels between cells of these regions and normal liver cells.

   Fig. 2. Computerized Scanned Image of Abdomen Area

   First Phase consists of five steps as follow:

   1 First Step:

   In this step, each image element that belongs to the range of (90-200) is given the value of zero, so it will appear in black. On the other hand each image element that doesnt belong to the range will be given the value of one, which will appear it in white.

   . = 0, 90 . 200 1,

   (B)

   (C)

   (D)

   Fig. 1. Images for examples of the basic cases for medical used: (A) Liver Tumor, (B) cyst in liver, (C) secondary Liver Tumor, (D) liver healthy.

   First Phase (Extracting the liver region of the computerized scanned image to the liver area):

   In this paragraph, first stage steps of the developed method are explained in details, and applied to one of the images used

   These values ( 90 & 200 ) have been chosen as the threshold values because of that the used medical image grey analysis has shown that most of the Liver image elements belong to the range in between these two thresholds ( Grey levels ) 90 & 200.

   Fig. 3. Resulted Image of giving elements thresholds values ( 90 & 200 ).

   1. Second Step:

      In this step, algorithm numbering of related compounds CCL has been used, to number areas or related compounds in image, so that each set of connected elements form a region. Because of the resulted image from the previous step is a binary image, so numbering of related compounds has been performed on the levels of 1 & 0.

      As a result of CCL algorithm application, a set of connected regions will appear. Applying of this algorithm has resulted into 339 connected regions figure 3.

   2. Third Step:

      By this step, anapproximate specification of the areaof liver is being performed, basedon: The anatomicaldescription,whichrefers to it asthe region ofthe

      largestspace in the computerized scannedimage of the abdominal area, and related compoundswhichbeen numberedin the previous step.

      ThusLiver area has been specified. Reviewing of theareas of connected regionsinthe image has shown that the largest area has the 51071 elements. Elements ofallthe other regions have been given a value of zero, thusthe elements ofthe largestarea haveremained as shown inFigure 4.

      It could be noticed thatthis imagecontains theregion ofliver, as well as many parts ofother organs, and thereason for this isthat the values ofthe graylevels ofthese parts areclose to thevaluesof graylevels of the liver area components.

      Fig. 4. Elements of the Largest Area in the Image.

   Fourth Step:

   A series of Morphological Operations are being applied through this step, exactly the Closing and Opening operations to obtain the Liver Area, and to get rid of parts of other organs.

   Opening function has the duty of smoothing borders of the body,also itcuts smallbarriersbetweenconnectedpartsas well asdeletessimpleoutcropswith smallthicknesses. While Closing function tendstosmoothen the borders ofbody, but thisprocess( asthe oppositeof thefirst one ) integrates and mergesthe narrowintersections, gets rid of theholes, andfills in the blanks located within the borders ofthe body. An appropriateconfigurationelement which will be used in both functions should be chosen [8].

   Initially, Closing function isperformedtofill in the small blanksin the body ofthe liverusing a configuration element has the shape of disk and a diameterof3elements.

   configuration element has the shape of disk and a diameterof13elements.

   Fig. 6. Resulted Image of applying Opening function to separate the regions of different organs.

   Fig. 7. Resulted Image of applying Closing function on the image resulted from Fig6.

   It could be noticed that more than one region has formed (the region of Liver in addition to another small region). To get rid of all additional regions, one additional opening function is being performed, but this time by using a configuration element a diameterof25elements ( Which is an estimated value to the areas of non-desired regions, those could appear along with the area of Liver ).

   Fig. 8. Resulted Image of applying Opening function to get rid of non- desired regions

   3. Fifth Step:

   Original valuesof the components ofthe liver area (derived fromthe original image) are being retraced, according to theequation (2):

   Fig. 5. Resulted Image of applying Closing function to fill in small blanks in the Liver.

   After filling small blanks in the area of Liver, both Opening and Closing functions are being applied respectively

   , = 0 , , > 0

   0

   Where:

   0

   , : Derived image of the area of Liver,

   , :

   to separate the regions of different organs by using a

   Original image,

   , : Image of Fig 8.

   Fig. 9. Extracted Liver area of the computerized scanned image of the abdomen area.

   1. Second Phase (Extraction of Suspicious Regions of Liver Area):

      The Image of Liver extracted of the Abdomen Image includes the suspicious regions which should be extracted. This is considered a hard process due to the nature of these suspicious regions in terms of shape, census and distribution. Therefore, it is necessary to find a flexible algorithms, capable to deal with this diversity in various pathological conditions.

      Considering Image items as a group of data, classified into minor groups, each has items of the same properties ( such as grey levels ), algorithmscan be used tocollect datafor thecompilation ofpicture elementswithin groups, so thatelementsof one set are similarwitheach other, and differ withelements ofthe other groups.

      FCMalgorithmisone ofthe most importantdata collectionalgorithmsthat are usedin the field ofmedicalimage processing. It doesnt rely onfixed valuesof thelevelsof gray,butdeal with cases on the basis of one by one, which givesgood results, although there is a differenceinvaluesof gray levels in medicalimagesresult ofmany factors, such asthe parameters ofthe imaging processandphysicalspecificationsfor the patient. To apply theFCMalgorithmon the image ofthe liver area, fcm function has been used of libraryofimage processingtoolsinMatlab [9].

      [center,U,obj_fcn] = fcm(data,cluster_n) [center,U,obj_fcn] = fcm(data,cluster_n,options)

      Where:

      • Data is the Matrix Data, which is a data group should be divided into groups so that each line of this matrix confirms one data point (in our case, an element of the image).

      • Cluster n: Number of groups ( >1).

      • Center: Matrix of the final groups centers, so that each line represents the coordinates of the center of a group of them.

      • U: Matrix of random distribution.

      • Obj_fcn: Matrix of function values by repeating the algorithm.

      • Options: Matrix of options by which the parameters of the control and are as follows:

        • The first option: Fuzziness Degree. The default value : 2.

        • The second option: The maximum number of iterations of the steps allowed. The default value: 100.

        • The third option: Minimal value of the improvement of the outcome allowed for the continued implementation of the repetition of steps and, of course, within the limits of the maximum number of iterations. The default value: 0.00001

        • The fourth option: Showing values and cost improvement through the process implementation. The default value : 1.

   The second phaseconsistsoffivestepsas follows:

   1. The first step:

   In this step, we continued to apply the fcm function on the image of the liver derived by using the default options matrix and determine the number of groups within 3 groups, so that one of these groups includes the elements of the image that lies in the background of the image, while the second group includes picture elements in suspicious regions, and therefore the last group includes picture elements innatural areas. Table 1 shows the results of FCM application on the liver image extracted in figure 9.

   Table 1: Shows that the value of improvement at iteration 25 has recorded a value less than the allowable minimum value, therefor the function will stop at this iteration.

   Iterati on Count	Cost Function(obj_fcn)	Improvement Value (_)
   1	14323.457117
   2	11094.385136	3229.071981
   3	11093.741025	0.6441111
   4	11083.445149	10.295876
   5	10921.407518	162.037631
   6	8910.165077	2011.242441
   7	2928.070674	5982.094403
   8	1615.482088	1314.588586
   9	1282.112222	333.369866
   10	718.559898	561.552324
   11	563.817394	156.742504
   12	531.893084	31.92431
   13	522.099320	7.793764
   14	521.991893	2.107427
   15	521.389860	0.602033
   16	521.212805	0.177055
   17	521.159923	0.052882
   18	521.143996	0.015927
   19	521.139177	0.004819
   20	521.137716	0.001461
   21	521.137272	0.000444
   22	521.137137	0.000135
   23	521.137096	0.000041
   24	521.137084	0.000012
   25	521.137080	0.000004

   Resultingfrom the applicationof the fcm functionthree sets ofelements, the re-formation of the elements ofeach groupseparatelyto form aspecific part of theoriginal image, Figure10 shows theimagesresulting fromthe formation ofte elements ofeachof the resulting three groups:

   (A)

   (B)

   (C)

   Fig. 10. Imagesresulting fromthe formation ofthe elements ofthe three groups: ( A)Elements ofthe backgroundimageandincludesa very small numberofelementsthat do not belongto the background, (B)Imageof healthyelementsin the liver, (C) Image of suspiciousareasin the liver.

   1. The Second Step:

      For each oneof the picturesfromthe previous step, the corresponding elementsin theimageareaof the liverin Figure 9 should be retrieved,where theboycott of each image elements with the imageof the liverformsthreepartial images, if been collectedtogether, theresult would be theoriginal image which represents the entireliverareathat was drawnin the first phase.

      (A)

      (B)

      (C)

      Fig. 11. Photosof the threegroupsafterretrievingelementsofthe imageareaof the liverresulting fromthe first phase: (A)Image of elements ofthe backgroundandincludevery fewof elementsthat do not belongto the background, (B) Imageof healthyelementsin the liver, (C) Image of suspiciousareasin the liver.

   2. The Third Step:

      In this step, three groups resulting from the previous step are examined and the group that represents the cells of suspicious areas is selected, this is because the fcm function does not give images of the three groups in the same order each time it is executed (Even when using the same data for income). So it is not correct to always assume that the first group will represent the normal cells in the liver, and that the second group will represent suspicious cells regions in the liver area, and the third group will represent cells that belong to the background.

      Screening process has executed and the group has been identified, depending on the maximum value of gray levels in each group because:

      gL> gS >gB (3)

      Where :

      gL: The maximum value of grey levels in the group of healthy cells in the liver area.

      gS:The maximum value of grey levels in the group of suspicious regions cells in the liver area.

      gB:The maximum value of grey levels in the group of image background elements.

      Therefore, the maximum valueof thegreylevelsis calculatedineach image of the three images resulted fromthe previous step,andthe previous termis being applied to select the imagethat representssuspiciousareasin the region ofthe liver.

   3. The Fourth Step:

      In this step, regions of very small areas is being neglected and deleted, comparing to the areas of suspicious areas which required to be extracted.

      This happens as a result of bwareaopen function application, which is one of the available morphological operations at MATLAB image processing library, so that all connected regions with thenumber of elementsless than orequal toa specified number are deleted. (100 elements, which is estimated valueofthese smallareas).

      Fig. 12. Deleting connected regions which have very small areas comparing to the areas of suspicious regions.

   4. The Fifth Step:

   The original values is being retrieved ofthe suspiciousregionselements of theimage of Liver by using the equation (4):

   depends on comparison of their properties (Shape,Area, proximity to the periphery,) with the standard values of these properties of Tumor Blocks.

3. RESULTS AND DISCUSSION

   Table2: Results of applying developed method on images presented as percentages

   Studied Case	Qty .	Extractio n Accuracy %	Remarks
   Primaryhepat ocellulartumo rs	6	95	Most of Tumors regions have been extracted
   Transitional secondary hepatic tumors.	2	80	Most of Secondary Tumors regions have been extracted
   Healthy Liver	1	100	No wrong regions have been extracted
   Cyst in the Liver Area	1	100	Cyst region has been extracted completely.
   Suspicious regions in the Liver Area	2	100	Suspicious Regions have been extracted completely.

   Accuracy ofextraction has been determine througha rough calculationof theratio ofsuspiciousareasderivedfrom the livertosuspiciousareas determined bymedical diagnosis.

   Almost allthe entireareasofsuspiciousregionsin the used imagesof the liver were extracted, no additional regions have

   , =

   0 , , > 0

   (4)

   been extracted as negative or wrong results(with the exception of parts ofborders of theliver, which could be

   Where:

   0

   disposedofduring thesubsequent treatment).

   A small portion ofthe suspiciousregioninoneof the used images has not been extracted, this is due that the section of

   , : Images of suspicious regions extracted of Liver area.

   0 , : Images of Liver resulted from the first phase (Liver Area Extraction)

   , : Combined Image, fig 12.

   Fig. 13. Suspicious Regions extracted of Liver Area.

   It is noticed from the figure 13 the existence of undesired regions in addition to the suspicious regions due to the rapprochement in the values of their grey levels and levels of grey of the suspicious regions. Getting rid of these parts

   theliver areathat contains thisparthas not beenextractedin the first stagebecause of the highcellsgray levelsof this section, which led todelete themduring the threshold step. Principally this problemcould beto overcome through the application ofappropriatepretreatment for usedimages [10].

   Almost secondary tumors areas have been extracted when developed method have been applied on secondary tumors images, the detection and extraction of these tumors require an additional special treatment due its nature in terms ofproliferation andspace.

   Correct results have been obtained when developed method has been applied on healthy cases, where cells of liver area been collected in one group except some parts of liver because of the liver is free of any suspicious region. These results showthe proper performanceof the developedmethodin bothhealthy and diseasedcases.

   Values of gray levels of the Liver borders cells are close to the values of gray levels of suspicious regions when imaging the Liver by using colored media. This has led to show parts of Liver borders is most of the resulted images when developed method has been applied. However, this doesnt affect the clarity of extracted suspicious regions.

4. CONCLUSION

Showthat the use CCLalgorithm in the process ofextractingthe liver areahelpsto develop an automated

method, because itis not affected bymanyvariable factors, such asdifferingimagingconditionsandspaceof the liver.

This algorithmisuseful inextractingsuspiciousareasofthe liver area, due to its reliability andnon-affecting by conditions changes ofthe imaging process, accurate results been obtained by this method as shown in Table 2.

The success of developed method in the research on optimized extraction for the liver area, this is due to the extraction of suspicious areas in the second phase of the developed method, so the suspicious areas in area of the liver has to be extracted within the area of the liver in the first stage of the fully developed method.The more clarity and separate of the liver area, the better extraction of the liver area, thebestextractionforsuspiciousareas could be obtained when the discrepancy betweenvalues ofthe graylevelsofnormal liver cellsand valuesofthe graylevels of cells of suspiciousareas.

Developedmethodcan be appliedtoa larger number ofimagesobtainedfordifferent casesofliverlesionsandpatientsin order to conducta clinical trialon the performance ofthe method, In addition to the adjustment of its steps, such as th optimal select of the volume and shape of used composition elementsof morphological operations used in this way.

ACKNOWLEDGMENT

This paper was supported by the internal project FEKT-S- 14-2429.

1. An Automatic Diagnostic System for CT Liver Image Classification. 1998, year. 45. DOI: http://theoncologist.alphamedpress.org/content/9/4.

2. GLETSOS, M., S.G. MOUGIAKAKOU, G.K. MATSOPOULOS,

   K.S. NIKITA, A.S. NIKITA a D. KELEKIS. Design and optimization of a neural network classifier. IEEE Transactions on Information Technology in Biomedicine. 2003, vol. 7, issue 3, s. 153-162. DOI: 10.1109/TITB.2003.813793. Available online: http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=1229 853

3. CHIEN-CHENG LEE, PAU-CHOO CHUNG a HONG-MING TSAI. Identifying multiple abdominal organs from CT image series using a multimodal contextual neural network and spatial fuzzy rules. IEEE Transactions on Information Technology in Biomedicine. 2003, vol. 7, issue 3. DOI: 10.1109/TITB.2003.813795. Available online: http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=1229 860

4. LAKSHMANA, B. a M. P. ARAKERI. Texture based characterization of liver tumor on computed tomography images. Proceedings of the International Conference. New York, New York, USA: ACM Press, 2011, s. 141-. DOI: 10.1145/1980022.1980052. Available online: http://portal.acm.org/citation.cfm?doid=1980022.1980052

5. GONZALEZ, Rafael C, Richard E WOODS a Steven L EDDINS. 2nd ed. Natick: Gates mark Publishing, 2009, xviii, 826 s. ISBN 978-0- 9820854-0-0.

6. SIVANANDAM, S, S SUMATHI a S DEEPA. Free Preview Introduction to Fuzzy Logic using MATLAB. New York: Springer, c2007, xiv, 430 p. ISBN 35-403-5780-7.

7. Proceedings of the International Conference. New York, New York, USA: ACM Press [cit. 2014-04-14]. Available online:http://portal.acm.org/citation.cfm?doid=1980022.1980052

REFERENCES:

1. SAHANI. Imaging the Liver. The Oncologist. 2004-07-01, vol. 9, issue 4, s. 385-397. DOI: 10.1634/theoncologist.9-4-385. Available online: http://theoncologist.alphamedpress.org/cgi/doi/10.1634/theoncologist. 9-4-385

2. ULAGAMUTHALVI, V. a D. SRIDHARAN. Development of Diagnostic Classifier for Ultrasound Liver Lesion Images. International Journal of Computer Applications. 2012-08-30, vol. 52, issue 18, s. 12-15. DOI: 10.5120/8300-1681. Available online: http://research.ijcaonline.org/volume52/number18/pxc3881681.pdf

3. WITHEY, D.J. a Z.J. KOLES. Medical Image Segmentation: Methods and Software. 2007 Joint Meeting of the 6th International Symposium on Noninvasive Functional Source Imaging of the Brain and Heart and the International Conference on Functional Biomedical Imaging. IEEE, 2007, s. 140-143. DOI: 10.1109/NFSI-ICFBI.2007.4387709. Available online:

http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=4387 709