transform based fusion algorithm for medical images_updated

Upload: sneha3026

Post on 02-Jun-2018

218 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    1/29

    TRANSFORM BASED FUSION

    ALGORITHM FOR MEDICAL IMAGES

    Presented By---

    Sneha Singh

    M.Tech. 2ndyear

    (12528023)

    Department of Electrical Engineering

    Indian Institute of Technology Roorkee, Roorkee

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    2/29

    TOPICS TO BE DISCUSSED

    Medical imaging

    Fusion algorithms

    Process flow

    Image datasets

    Performance measures

    Results and Discussions

    Comparative Analysis

    Conclusion

    References

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    3/29

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    4/29

    LITERATURE REVIEW

    S. No. Authors Refs. Year Fusion Techniques

    1 H. Li et al. [3] 1995 This fusion scheme is based on discrete wavelet transform (DWT).

    2 S. Li et al. [6] 2002 This paper presented the combination of DWT and artificial neural

    networks (ANN) for pixel level multifocus image fusion.

    3 G. Piella [8] 2003 This paper proposed a general framework on pixel to pixel fusion

    schemes.

    4. B. Xu et al. [53] 2004 This paper presented fusion algorithm based on pulse coupled neural

    network (PCNN).

    5 W. Li and X

    Zhu

    [11] 2005 In this paper the algorithm is based on wavelet packet analysis (WPA)

    and pulse-coupled neural networks (PCNN).

    6 W. Li and X-F

    Zhu

    [10] 2005 This technique represents medical image fusionbased on pulse-coupled

    neural networks (PCNN).

    7 Q. Miao and

    B. Wang

    [13] 2006 Contourlet transform is used in this paper for image fusion .

    8 A. Wang et al. [14] 2006 This method demonstrate the application of wavelet transformation to

    multi-modality medicalimage fusion.

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    5/29

    LITERATURE REVIEW

    S. No. Authors Refs. Year Fusion Techniques

    9 B. Yang et al. [15] 2007 This technique presented the use of nonsubsampled contourlet transform

    (NSCT) for image fusion.

    10 D. Agrawal

    and J. Singhai

    [21] 2010 In this method a modifiedapproach of PCNNis proposed by reducing the

    processing time and computational complexity.

    11 Q.-g. Miao etal.

    [45] 2011 In this paper image fusion by using Shearlet transform is presented.

    12 C. Yuan et al. [49] 2011 This paper proposed image fusion by using Nonsubsampled shearlet

    transform (NSST).

    13 S. Das and

    M.K. Kundu

    [30] 2012 This paper proposed NSCTand the PCNNwith modified spatial frequency

    (MSF) in order to obtain better fusion results .

    14 P. Geng et.al. [31] 2012 Image fusion with PCNN in shearlet domain is presented.

    15 X. Sun et al. [34] 2013 This paper presented image fusion based on NSCT domain with improved

    contrast.

    16 P. Ganasala

    and V. Kumar

    [54] 2014 This paper presented CT and MR image fusion scheme in NSCT domain

    with improved image contrast.

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    6/29

    PROCESS FLOW

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    7/29

    CT images

    MR images

    IMAGE DATASETS

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    8/29

    Block representation of fusion technique in transform domain

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    9/29

    PERFORMANCE MEASURES

    S.NO. Performance Measures Mathematical Expression

    1 Entropy (EN) )

    =

    2 Standard deviation (STD)

    ,

    =

    =

    3 Mutual Information (MI) ; ; where ; ,(,) , (,)()()

    =

    =

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    10/29

    PERFORMANCE MEASURES

    S.NO.

    Performance Measures Mathematical Expression

    4 Spatial Frequency (SF) where

    1

    (1) ( , 1 (, ))

    =

    =

    1

    (1)

    ( 1, (, ))

    =

    =

    5 Image Quality Index (IQI) 2

    2 and

    0 , (, )

    2

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    11/29

    PERFORMANCE MEASURES

    S.NO. Performance Measures Mathematical Expression

    6 Edge Strength ,

    , (,)== , (, , (,)==where

    +( ,) and 1 ( , )

    , , , ; , > , , , ;

    , 1 , ,

    /2

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    12/29

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    13/29

    SUBJECTIVE ANALYSIS OF VARIOUS METHODS

    WT AVG MAX (M1) NSCT AVG MAX (M2)

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    14/29

    WT-AVG-MAX (M1)

    DatasetPERFORMANCE MEASURES

    EN MI STD SF Q0 QXY/F

    # 1 5.2127 2.5634 65.3796 5.1216 0.2822 0.2769

    # 2 4.8940 2.5240 59.7501 6.6219 0.3052 0.2586

    # 3 4.4542 2.3354 66.8901 5.5161 0.3290 0.2592

    # 4 5.2507 2.1770 57.0081 7.1815 0.3244 0.2502

    # 5 4.8650 2.3399 65.7486 6.5576 0.2964 0.2405

    # 6 4.1145 2.6323 62.4138 5.5684 0.2883 0.2669

    # 7 4.0108 2.4357 69.7617 6.3112 0.3655 0.2758

    # 8 4.7939 2.3288 62.6742 5.8714 0.3781 0.2929

    # 9 4.5105 2.4310 61.4288 6.0779 0.2856 0.2395

    NSCT-AVG-MAX (M2)

    Datase

    t

    PERFORMANCE MEASURES

    EN MI STD SF Q0 QXY/F

    # 1 5.1201 3.0783 63.8424 4.8754 0.3525 0.3784

    # 2 4.8601 2.7564 57.1828 6.2705 0.3208 0.3460

    # 3 4.5035 2.4772 65.1020 5.6107 0.3224 0.3443

    # 4 5.2960 2.4050 60.2343 7.1192 0.3894 0.3401

    # 5 4.7970 2.5112 63.1778 6.2101 0.3595 0.3328

    # 6 4.2301 2.7863 60.4923 5.3852 0.3306 0.3456

    # 7 4.0645 2.5961 68.9085 5.8846 0.3389 0.4086

    # 8 4.8013 2.5628 60.3235 6.0536 0.3573 0.3930

    # 9 4.6494 2.6086 59.0970 6.2537 0.2988 0.3118NSST-AVG-MAX (M3)

    Datase

    t

    PERFORMANCE MEASURES

    EN MI STD SF Q0 QXY/F

    # 1 5.2057 3.2190 67.5122 5.2468 0.4204 0.4380

    # 2 4.8638 2.9961 61.8963 6.5585 0.4097 0.4592# 3 4.6208 2.6345 68.3783 6.0983 0.3997 0.4059

    # 4 5.2286 2.7054 63.3878 7.5095 0.4502 0.4963

    # 5 4.8337 2.7793 67.7438 6.7447 0.4336 0.4847

    # 6 4.2791 3.1108 63.5787 5.5623 0.4872 0.4736

    # 7 4.1262 2.8338 71.4323 5.9743 0.5218 0.5163

    # 8 4.8554 2.8409 65.3852 6.1661 0.4452 0.5448

    # 9 4.6864 2.8091 62.9468 6.5367 0.4295 0.4102

    NSCT-MAX-SF (M4)

    Datase

    t

    PERFORMANCE MEASURES

    EN MI STD SF Q0 QXY/F

    # 1 5.1121 3.1058 86.1278 5.3696 0.4415 0.4299

    # 2 4.9348 3.0139 81.3054 6.3475 0.4128 0.4636# 3 4.6876 2.9970 91.6373 5.6933 0.4173 0.4172

    # 4 5.2871 2.7727 83.2657 7.2480 0.4278 0.4692

    # 5 4.8924 2.9413 89.1206 6.3264 0.4225 0.4723

    # 6 4.2841 3.1173 85.3086 5.4073 0.5131 0.4957

    # 7 4.1826 3.0357 95.0274 6.0055 0.5681 0.4805

    # 8 4.8586 2.8499 84.3038 6.1553 0.4662 0.5490

    # 9 4.6937 3.0268 83.9519 6.5796 0.4333 0.4164

    NSCT-MAX-MSF-PCNN (M5)Datase

    t

    PERFORMANCE MEASURES

    EN MI STD SF Q0 QXY/F

    # 1 5.2727 3.1384 86.2748 5.4210 0.4542 0.4375

    # 2 4.9361 3.0489 81.5871 6.6242 0.4346 0.5088

    # 3 4.6935 3.0389 91.6671 5.8879 0.4354 0.4234

    # 4 5.3184 2.8223 83.3683 7.3292 0.4365 0.4866

    # 5 4.9108 2.9532 89.4888 6.6292 0.4303 0.4729

    # 6 4.3675 3.1358 85.6309 5.5829 0.5304 0.4975

    # 7 4.1897 3.0390 95.4015 6.2845 0.5702 0.4825

    # 8 4.8604 2.8607 84.3626 6.2186 0.4722 0.5772

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    15/29

    NSST-MAX-SF-PCNN (M6)

    Datase

    t

    PERFORMANCE MEASURES

    EN MI STD SF Q0 QXY/F

    # 1 5.2989 3.2226 86.6951 5.5828 0.4709 0.4494

    # 2 4.9914 3.0644 81.7370 6.6959 0.4482 0.5134

    # 3 4.7247 3.0310 92.1739 6.1315 0.4396 0.4261

    # 4 5.3486 2.8432 83.4918 7.6077 0.4553 0.5122

    # 5 4.9528 2.9722 89.6534 6.8515 0.4470 0.4919

    # 6 4.4379 3.1570 85.9482 5.7063 0.5482 0.5112

    # 7 4.5431 3.0913 95.7893 6.3453 0.5756 0.5231

    # 8 4.8970 2.8813 84.6587 6.4386 0.5023 0.5797

    # 9 4.7826 3.0502 84.4332 6.8568 0.4518 0.4244

    NSST-MAX-MSF-PCNN (M7)

    Datase

    t

    PERFORMANCE MEASURES

    EN MI STD SF Q0 QXY/F

    # 1 5.3401 3.2462 86.8412 5.6992 0.4764 0.4631

    # 2 4.9941 3.0792 81.9542 6.8099 0.4526 0.5164

    # 3 4.7301 3.0594 92.3936 6.1480 0.4490 0.4347

    # 4 5.3937 2.8469 83.7206 7.6801 0.4597 0.5143

    # 5 4.9768 2.9865 89.8980 6.9478 0.4583 0.4964

    # 6 4.5661 3.1730 86.1057 5.7910 0.5495 0.5283

    # 7 4.6669 3.0978 95.7932 6.3712 0.5825 0.5399

    # 8 4.9093 2.8877 84.6746 6.5236 0.5058 0.5856

    # 9 4.7990 3.0569 84.7801 6.8796 0.4527 0.4295

    NSST-RE-NSML-PCNN (M9)

    Datase

    t

    PERFORMANCE MEASURES

    EN MI STD SF Q0 QXY/F

    # 1 5.3430 3.2956 87.1860 5.7153 0.4839 0.4685

    # 25.0335 3.0853 82.0347 6.8577 0.4627 0.5496

    # 3 4.8078 3.0724 92.6860 6.2712 0.4833 0.5002

    # 4 5.4243 2.8509 83.8141 8.8396 0.4649 0.5255

    # 5 4.9927 2.9886 89.9084 7.7340 0.4602 0.5188

    # 6 4.6794 3.2026 86.1525 5.8769 0.5592 0.5514

    # 7 4.7696 3.1514 96.1604 6.4666 0.5857 0.5970

    # 8 4.9426 2.8933 85.0326 6.5816 0.5153 0.5930

    # 9 4.9018 3.0973 84.8521 6.9213 0.4557 0.4969

    NSCT-RE-NSML-PCNN (M8)

    Datase

    t

    PERFORMANCE MEASURES

    EN MI STD SF Q0 QXY/F

    # 1 5.2839

    3.1822

    86.5459

    5.5582

    0.4616

    0.4480

    # 2 4.9362

    3.0604

    81.5441

    6.3185

    0.4401

    0.5098

    # 3 4.7231

    3.0454

    92.1624

    5.8938

    0.4387

    0.4236

    # 4 5.3255

    2.8377

    83.3934

    7.4228

    0.4423

    0.4962

    # 5 4.9413

    2.9641

    89.6431

    6.6472

    0.4388

    0.4843

    # 6 4.4298

    3.1544

    85.6424

    5.6298

    0.5422

    0.5008

    # 7 4.2629

    3.0902

    95.4214

    6.2919

    0.5746

    0.5211

    # 8 4.8878

    2.8624

    84.6310

    6.3458

    0.4855

    0.5784

    # 9 4.7392

    3.0475

    84.3938

    6.7025

    0.4511

    0.4211

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    16/29

    Comparative analysis of the averaged performance measures obtained by the

    fusion methods

    Fusion Methods EN MI STD SF Q0 QXY/F

    WT-AVG-MAX 4.6785 0.44012.4186 0.140463.4506 3.89916.0920 0.64510.3172 0.03510.2623 0.0177

    NSCT-AVG-MAX 4.7024 0.39412.6424 0.204462.0401 3.56995.9626 0.63560.3411 0.02680.3556 0.0311

    NSST-AVG-MAX 4.7444 0.37072.8810 0.190865.8068 3.15016.2664 0.67020.4441 0.03870.4699 0.0468

    NSCT-MAX-SF-PCNN 4.7703 0.35852.9845 0.113386.6721 4.41636.1258 0.59640.4558 0.05240.4660 0.0421

    NSCT-MAX-NMSF-PCNN 4.8081 0.37133.0086 0.110286.9047 4.43416.2976 0.60170.4679 0.04940.4785 0.0490

    NSST-MAX-SF-PCNN 4.8863 0.30733.0348 0.121887.1756 4.52856.4685 0.62670.4821 0.04930.4924 0.0508

    NSST-MAX-NMSF-PCNN 4.9307 0.2848 3.0482 0.126787.3512 4.48716.5389 0.62300.4874 0.04860.5009 0.0510

    NSCT-RE-NSML-PCNN 4.8366 0.34993.0271 0.118987.0419 4.49686.3123 0.58190.4750 0.05030.4870 0.0503

    NSST-RE-NSML-PCNN 4.9883 0.25083.0708 0.142387.5363 4.55736.8071 0.96930.4968 0.04700.5334 0.0435

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    17/29

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    18/29

    CONCLUSIONS

    Several image fusion techniques using different transformation methods are

    discussed.

    DWT capable to preserve the spectral information but unable to express spatial

    characteristics efficiently.

    NSCT improve the performance and overcome the limitations of the wavelet .

    Pixel-level spatial domain fusion such as averaging usually lead to contrast

    reduction while improve the overall frequency of pixels.

    PCNN gives good decision output as it biologically inspired by human visual system

    hence better contrast.

    Spatial and modified spatial frequency is become good focus indicator as total

    activity and clarity level measure.

    NSST improves the shortcoming of NSCT and gives better directionality hence

    ensures good image quality.

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    19/29

    CONCLUSIONS

    Three fusion schemes using NSST are presented and compared with another rules

    based on same fusion combination but using WT and NSCT.

    WT based on AVG-MAX gives high entropy and spectral information but lacks of

    contrast and localization.

    NSCT improve the performance with high degree of localization and directionality

    but due to AVG it degraded with low standard deviation.

    NSST outperforms the two methods and gives better evaluation measures.

    Compared with other approaches PCNN-based fusion are of higher clarity and

    contrast.

    The sum-modified-Laplacian (SML) can well reflects the feature information of

    the edges of the image.

    The energy of image reflects the brightness level of scene.

    The fusion scheme based on NSST using regional energy and NSML motivated

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    20/29

    REFERENCES

    References.docx

    http://localhost/var/www/apps/conversion/tmp/scratch_2/References.docxhttp://localhost/var/www/apps/conversion/tmp/scratch_2/References.docx
  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    21/29

    Activity Level Measures

    S.No.

    Activity Measures Mathematical Expression

    1 Pixel Averaging (AVG) (, ) ( , (, ))

    2 Maximum Coefficientselection (MAX) , (, ), (, ) (, )(, ), otherwise

    3 Regional Energy (RE)

    , , (,)==

    where (,)is weight template of size 33

    (,) 1

    9

    1 1 11 1 11 1 1

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    22/29

    Activity Level Measures

    S.No. Activity Measures Mathematical Expression

    4 Novel sum-modified-laplacian (NSML)

    , 2 , 1, 1, 2 , , 1 , 1

    , , . , where (,)is weight template of size 33 ,

    1/15 2/15 1/152/15 3/15 2/151/15 2/15 1/15

    5 Modified SpatialFrequency(MSF)

    1(1)(1) , 1, 1

    =

    =

    1(1)(1) 1, , 1 ==

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    23/29

    Wavelet Transform (WT)

    Wavelet transform for image fusion

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    24/29

    Nonsubsampled Contourlet Transform(NSCT)

    Block diagram Nonsubsampled contourlet transform based imagedecomposition

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    25/29

    Nonsubsampled Contourlet Transform(NSCT)

    Block diagram Nonsubsampled contourlet transform based imagedecomposition

    Nonsubsampled Shearlet T f

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    26/29

    Nonsubsampled Shearlet Transform(NSST)

    Block diagram Nonsubsampled shearlet transform based imagedecomposition

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    27/29

    Pulse Coupled Neural Network (PCNN)

    PCNN Neuron Model

    28

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    28/29

    Mathematical equations supports PCNN neuron model:

    28

    , , , 1 ,,,,[ 1 ],

    ,

    ,

    1

    ,,,

    ,

    [ 1 ],

    , , 1 , , , 1 ,[]

    , 1, , > ,0,

  • 8/11/2019 Transform Based Fusion Algorithm for Medical Images_updated

    29/29