feature extraction and segmentation of ct lungs images for better assessments

Upload: journal-of-computing

Post on 04-Apr-2018

217 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/30/2019 Feature Extraction and Segmentation of CT Lungs Images for better Assessments

    1/5

    Feature Extraction and Segmentation of CTLungs Images for better Assessments

    S. Yasin1, M. S. Naweed1, M. Rehman1

    Abstract Computed Tomography scan is widely being used for several medical diagnoses. CT imaging has already shown its

    practical impact in diagnosis of brain and lungs diseases. CT scan has brought to the seen; which was unseen before with the

    naked eye; and its native support as a digital imaging modality has invited the researchers to exploit the digital image

    processing techniques to better process the information available in the CT scan images as compared to the information

    processed by the naked eye. It is obvious that human vision has many constraints and limitations, and hence cannot process

    complete information available in the CT images; where as digital images can be manipulated by computer at pixel level. The

    objective of this research is to develop a segmentation technique in the field of medical imaging to help better visual inspection

    of certain lungs features which lead to better diagnosis of lungs diseases. In this research paper we have proposed a method by

    which the lungs are segmented using iterative threshold, morphological operators, and flood-fill algorithm; as well as basic lungs

    anatomy is labeled through eight-neighbor connectivity technique. Also, we have segmented different regions of the lungs

    anatomy by applying specific gray level filtering; through which the filtered anatomy of the lungs is not only separated but better

    examined as compared to previous implemented techniques. The overall accuracy of our results is almost 94%.

    Keywords: Computed Tomography (CT), Lungs Segmentation, Lungs Anatomy, Lungs Features Extraction, Image

    Processing

    1 INTRODUCTION

    ungs abnormalities in human cause different diseasesespecially in third world countries like Pakistanwhere health is major subject on which work is to be

    done. There is lack of latest medical technologies in Pakis-tan which is one of the major factors due to which these

    diseases are not diagnosed well in time and cause severedamages. Human eyes could not detect the abnormalitiesin human anatomy directly. Most of the time X-rays andRadio therapy is used to diagnose the diseases. But as thelast decision is to be made by the medical practitioner;human eye error factor could not be ignored. Delay indiagnosis of the disease causes the death in extreme cases.Due to all these factors segmentation in the field of imageprocessing is at high credit . Segmentation of digi tal im-ages is the main area of image processing.

    Segmentation focuses on the particular area which is tobe examined, highlights the region of interest and helpsthe medical practitioner to conclude some better diagno-sis of diseases. Manual analysis may take enough time toexamine this medical data. Automated analysis of suchhuge amount of medical data may lead to solution of thisproblem. Typically in medical imaging, the CT scan im-age data is in huge amount [1]. The iterative threshold [7]in combination to flood fill technique [5] is used to seg-ment basic lungs anatomy. The objective is to develop asegmentation technique in order to help the doctors andsurgeons to identify the abnormal portions of the lungs

    for the treatment of certain illness [9] in a better way. Theproposed method solves the problem using the betterobservation of human lungs anatomy. Lung nodules arebetter observed using our proposed methodology. Thispaper describes the iterative threshold based segmenta-

    tion procedure and multi-gray-level filtering technique,which is designed to identify the certain lungs regionsand extract the anatomical features of lungs. Interpreta-tion of medical images is often difficult and time consum-ing, even for expert physicians. There are a number ofalgori thms for segmentation [5]. There is a lot of segmen-tation techniques applied [6]; but no one is perfect in allmeans. We have performed the segmentation on the CTimages which were taken from Radiology Department,Bahawalpur Victoria Hospital, Bahawalpur, Punjab, Pa-kistan.

    2 MATERIALSANDMETHODSWe have proposed a method for anatomical features ex-traction of the lungs. This method consists of a step bystep procedure. In the first step we select the CT lungsimage. In second step, the selected CT image is convertedto binary image using iterative threshold technique. Inthi rd step, we apply erosion and di lation operators to fi llunwanted gaps in the background and foreground of thebinary image. Then, in fourth step, we obtain the originallungs image data by removing the background of lungsimage. In this step we have extracted the background bycombining 8-neighboor pixel connectivity and flood fill

    technique with four seed points; and then we have in-verted the resultant i mage to obtain the mask and appl iedthis mask to extract the lungs and rest of the anatomy. Inthe fi fth step we have re-binarize the resultant image and

    S. Yasin D epartm ent of Compu ter Science, Islamia U ni versity of Bawalpur M . S. N aweed D epartm ent of Compu ter Science, Islamia U niv ersit y of

    Bawalpur M . Rehman D epartm ent of Comput er Science, Islamia U niv ersit y of Ba-

    walpur

    JOURNAL OF COMPUTING, VOLUME 4, ISSUE 11, NOVEMBER 2012, ISSN (Online) 2151-9617

    https://sites.google.com/site/journalofcomputing

    WWW.JOURNALOFCOMPUTING.ORG 8

  • 7/30/2019 Feature Extraction and Segmentation of CT Lungs Images for better Assessments

    2/5

    applied the SEDM (split-erode-dilate-merge) process. The

    split part of the SEDM process breaks the image intomany parts according to the anatomy separated in re-binarized image. The erode part of the SEDM process isapplied 3-5 times on each part of the anatomy to growand fill the un-desired gaps. The dilate part of the SEDMprocess is applied exactly the same 3-5 times to shrinkeach part of the image back to original size. The merge

    part of the SEDM process is used to combine the parts ofthe image back to single image. The purpose of SEDMprocess is not only to fill the unwanted gaps in lungsanatomy but also keeping the parts of the anatomy sepa-rated to each other; see differences in Figure 2. (d) andFigure 2. (i) clearly. The resultant image of the SEDMprocess is used as mask to extract lungs area with maxi-mum accuracy obtained. In sixth step, the lungs image islabeled into different lungs regions. These regions aretreated as separate objects of the lungs; such as left lung,right lung, trachea, and any other disconnected lung re-gion (may be treated as noise). In seventh and last step we

    have extracted the anatomical features of lungs objects.These features are extracted using gray-level filtering andlabeled as blood vessels, soft tissues, lungs boundary etc.These features are labeled by applying statistical observa-tions over 20 CT scan data sets, each containing 20 to 30slices each. This method gives us overall 94% accurateresults.

    3 RESULTSANDDISCUSSIONS

    In this section we have shown the results of the different

    processes mentioned in the methodology explained inFigure 1.

    Original Lungs CT Image

    Covert to Binary Image

    (Iterative Threshold)

    Obtain Lung Mask(Erosion + Dilation)

    Apply SEDM Process

    (Split, Erode, Dilate, Merge)

    Label Objects in Lungs

    (Left Lung, Right Lung, Trachea, Others)

    Extract Anatomical Features(Filter: Blood vessel, Soft tissues, Alveoli,

    Lungs Boundary)

    Obtain Lungs Region Of Interest

    (Background Removal: Obtain Mask,

    Invert Mask, Apply Mask)

    Fig. 1. Step Wise Process of Features Extraction of the Lungs.

    a) Original Image b) Binary Image

    c) After Erosion d) After Dilation

    e) Obtain Mask f) Invert Mask

    g) Lungs ROI h) Re-Binarize

    JOURNAL OF COMPUTING, VOLUME 4, ISSUE 11, NOVEMBER 2012, ISSN (Online) 2151-9617

    https://sites.google.com/site/journalofcomputing

    WWW.JOURNALOFCOMPUTING.ORG 9

  • 7/30/2019 Feature Extraction and Segmentation of CT Lungs Images for better Assessments

    3/5

    Here is another set of results examined by implementingthe SEDM methodology to extract lungs and its features,with the help of software developed using C#.NET and

    windows operating system. First of all the original imageis shown which lies between gray level intensity rangefrom 0 to 255 in which complete lungs anatomy is visibleas shown in Figure 3.

    When we take gray level intensity range from 0 to 24then only trachea part is visible as Figure 4.

    As we increase the gray level intensity range from 24

    to 47 then soft tissue material including air sacks has beenobserved as in Figure 5.

    It is observed that gray level intensity range from 25 to38 and from 39 to 46 segments two regions which werecombined in Figure 5. The fir st range from 25 to 38 showsalveoli and second range from 39 to 46 which shows softti ssues and their boundaries, which clearly help the medi-cal practitioner about the anatomical structure of the

    lungs.

    i) SEDM Applied j) Lungs Only

    k) Left Lung l) Trachea

    m) Right Lung n) Blood Vessels

    o) Soft Tissues

    Fig. 2. Lungs Segmentation and Feature Extraction Process Results.

    Fig. 3. Complete Lung Anatomy by Gray Level Intensity Range from 0

    to 255

    Fig. 4. Only Trachea Part by Gray Level Intensity Range from 0 to 24

    Fig. 5. Soft Tissue with Air Sacks by Gray Level Intensity Range from

    25 to 46

    JOURNAL OF COMPUTING, VOLUME 4, ISSUE 11, NOVEMBER 2012, ISSN (Online) 2151-9617

    https://sites.google.com/site/journalofcomputing

    WWW.JOURNALOFCOMPUTING.ORG 10

  • 7/30/2019 Feature Extraction and Segmentation of CT Lungs Images for better Assessments

    4/5

    Furthermore as we move gray level intensity range from47 to 66 then lungs blood vessels with boundary areawith some soft tissues (noise) is visible in Figure 8.

    At last when we set the gray level intensity range from 67to 255 then blood vessels, their boundaries and lungboundaries are clearly vi sible as shown below.

    4 EXTRACTIONOFFEATURESFROMOTHERCTLUNGSIMAGES

    g. 6. Alveoli by Gray Level Intensity Range from 25 to 38

    g. 7. Soft Tissues with Boundaries by Gray Level Intensity Range

    m 39 to 46

    g. 8. Blood V essel Boun daries by G ray Level I nt ensity Range from 47 t o 66

    Fig. 9. Blood Vessels with Lungs Boundaries by Gray Level Intensity

    Range from 67 to 255

    Gray

    LevelsExtracted Objects Image

    0 24 Trachea

    24 47Rounded air sacksSoft tissue material

    39 46 Just alveoli

    47 255Blood vessels w ithboundaries and rest

    55 255

    Lung boundary +Blood vessels +Blood vessels boundary

    67 122Blood vessels boundaryonly

    67 255Thick boundary ofblood vessels

    98 255Shear Blood vesselsBoundary

    JOURNAL OF COMPUTING, VOLUME 4, ISSUE 11, NOVEMBER 2012, ISSN (Online) 2151-9617

    https://sites.google.com/site/journalofcomputing

    WWW.JOURNALOFCOMPUTING.ORG 11

  • 7/30/2019 Feature Extraction and Segmentation of CT Lungs Images for better Assessments

    5/5

    It is observed that traditionally the lungs examined bymedical practitioner is not so easy task but by the imageprocessing techniques, it helps the medical practitioner tonot only better examine the patient but also give better andefficient results.

    5 CONCLUSION&FUTUREWORKMedical imaging is already a grown-up field and beingvastly used in several medical fields. Many doctors andsurgeons in the world identify the abnormalities on thebehalf of the anatomical features of the human body. Toenhance and improve their experience with medical im-ages many software has been built and still research is onthe trails. The objective of our research is to extract andenhance the certain features of interest; presented inlungs images to provide better observation and examina-tion, which may otherwise be neglected from a humaneye. We have discussed a SEDM method which extracts

    the lungs region of interest (ROI), and anatomical featuresare extracted using digital image processing techniquesfor better v isual inspection by fi ltering and/ or enhancingcertain portions of the lungs. The accuracy measure isgood but it will be increased in future. In addition, ourmethod makes effective use of object oriented techniquesto extract certain portions of the lungs into labeled objectsand provides filtering and better observation of differentobjects of the lungs. Several processes have already beenapplied to achieve this current methodology and it is stillin progress, so that maximum accuracy can be achievedin future research.

    REFERENCES

    [1] T. Lin-Yu, and L. Huang, An Adaptive Thresholding Method for

    Automatic Lung Segmentation in CT Images, In Proc. of IEEE AFRI-

    CON2009, Nairobi, Kenya, p. 1-5.

    [2] U. Soumik and J.M. Reinhardt , Smoothing lung segmentati on

    surfaces in 3D x-ray CT images using anatomic guidance, In

    Proc SPIE 2004, p. 1066-1075.

    [3] A.G. Samuel, and W.F. Sensakovic, Automated lung segmen-

    tation for thoracic CT: impact on computer-aided diagnosis,

    Academic Radiology, Vol. 11, Issue. 9, pp. 1011-1021,Sep, 2004.

    [4] H. Shiying, E.A . Hoffman, and J.M. Reinhardt , Automatic

    Lung Segmentation for Accurate Quantitation of Volumetric X-

    Ray CT Images, IEEE TRANSACTIONS ON M EDICAL IM -

    AGING,Vol. 20, Issue. 6, pp. 490-498, June, 2001.

    [5] I. Lui s, W. Schroeder, L. N q, and J. Cates, The ITK Software

    Guide, 2nd Ed. The Insight Segmentation and Registration Tool-

    ki t version 1.4, 2004.

    [6] P. Regina, and K.D. Toennies, Segmentation of medical images

    using adaptive region growing In Proc SPIE 2001 M edical Im ag-

    in g,Vol. 4322, p.1337-3386.

    [7] Z. Vesna, and M. Bojic, Automatic detection of abnormalit ies

    in lung radiographs caused by planocellular lung cancer , i n

    Proc. Biomedical Engin eeri ng M ECBM E, 2011 1st M iddle

    East Conf. 21-24, Feb., p. 69-72.

    [8] A. Michela, B. Lazzerini, F. Marcell oni, Segmentation and

    reconstruction of the lung volume in CT images In Proc. of the

    A CM symposium on Appli ed computin g, 2005. NY, USA P. 255-259

    [9] G. Tristan, J. Montagnat and I .E.Magni n Creais, Texture based

    medical image indexing and retrieval: application to cardiac

    imaging In Proc. of the 6th ACM SIGGM Int. workshop on

    Mult imedia information retri eval P. 135 - 142 , NY, USA

    [10] A. El-Bazl, A .A. Farag, R. Falk , R. La Rocca. Automatic identif i-

    cation of lung abnormali ties in chest spiral CT scans In Proc.of

    ieee Inl .conf.ICASSP 03, 2003, Vol.2. p.261-264.

    JOURNAL OF COMPUTING, VOLUME 4, ISSUE 11, NOVEMBER 2012, ISSN (Online) 2151-9617

    https://sites.google.com/site/journalofcomputing

    WWW.JOURNALOFCOMPUTING.ORG 12