emg decomposition

Upload: niranjan-rao-kakarla

Post on 08-Apr-2018

226 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/6/2019 Emg Decomposition

    1/42

    EMG SIGNAL DECOMPOSITIONUSING PATTERN CLASSIFICATION

    TECHNIQUES

    Under the guidance ofDr. VINOD

    KUMARPROFESSOR &HEAD

    By

    niranjana rao kakarla

    M.Tech.(M&I) 08528010 I.I.T. Roorkee

  • 8/6/2019 Emg Decomposition

    2/42

    Topics to be discussed:-

    EMG signal composition

    Steps involved in decomposition

    Single classifier approachesMultiple classifier approaches

    Comparative study

  • 8/6/2019 Emg Decomposition

    3/42

    EMG SIGNAL COMPOSITION Electromyography is the detection of muscle activity

    associated with muscle contraction

    MOTOR UNIT: a motor unit is an -motoneuron and all themuscle fibres innervated by its axon

    MUP waveform: the summation of the spatially andtemporally dispersed potentials, created by impulses

    propagating along the individual muscle fibers of a motorunit

    MUPT: is the collection of MUP waveforms generated by onemotor unit separated by their inter-discharge intervals

  • 8/6/2019 Emg Decomposition

    4/42

    Fig. 1. MUPT with MUPs separated by their inter-dischargeintervals[1]

    The superposition of the MUPTs of all recruited motor unitsand background noise comprises EMG signal

    Fig. 2. A 1-s epoch of raw EMG signal[2]

  • 8/6/2019 Emg Decomposition

    5/42

    E M G S IG N A L D E C O M P O S IT IO Nprocess of resolving a composite EMG signal in to its

    constituent MUPTs

    Figure.3. Schematic representation of the detection anddecomposition of the intramuscular EMG signal[3].

  • 8/6/2019 Emg Decomposition

    6/42

    Applications of decompositionThe shapes and occurrence times of MUPs provide an

    important source of information to assist in the diagnosis of

    neuromuscular disorders.

    reflect the structural and physiological changes of a MU

    central nervous system recruitment and control of MUs can

    be

    studied.

  • 8/6/2019 Emg Decomposition

    7/42

    Major steps involved in EMG signal decomposition are

    1. segmentation

    2. feature extraction

    3. clustering

    4. supervised classification

  • 8/6/2019 Emg Decomposition

    8/42

    Segmentation of the composite signal

    Segmentation refers to the detection of all MUAPsgenerated by MUs active during signal acquisition.

    Fig. 4. (Top) Raw EMG segment recorded from biceps brachii using

    concentric needle electrode-a typical "mixed interference pattern" in which

    the individual MUAP's are difficult to identify and characterize because of

    superimpositions. (bottom) Same segment after filtering using the second-

    order differentiating filter[4].

  • 8/6/2019 Emg Decomposition

    9/42

    Fe a tu re e xtra ctio nFeature

    The detected MUAPs must be represented using a vector for

    pattern recognition.

    The characteristics of the MUAPs used for this representation

    are called features.

    Feature spaces

    The multi-dimensional space composed of all possible feature

    values is called the feature space.

  • 8/6/2019 Emg Decomposition

    10/42

    feature spaces used to represent MUAPs for signal

    decomposition are:

    1.peak to peak voltage, number of phases, duration, number of

    turns, etc.

    2.Fourier transformation coefficients

    3.Wavelet coefficients

    4.time samples of the band-pass filtered signal

    5.time samples of the low-pass differentiated signal.

  • 8/6/2019 Emg Decomposition

    11/42

    CLUSTERING OF DETECTED MUAPs

    partitioning detected MUAPs into a number of groups orclusters.

  • 8/6/2019 Emg Decomposition

    12/42

  • 8/6/2019 Emg Decomposition

    13/42

    SINGLE CLASSIFIER APPROACHESCERTAINITY CLASSIFIER

    It is a non-parametric template matching classifier

    Uses certainty based approach for assigning MUPs to MUPTs

    For a set ofM MUPT class labels = {1,2,..,M} the decision

    functions for the assignment of MUP mj are evaluated for only

    the two MUPTs with the most similar templates to MUP mj

    It determines two types of decision functions

    1. based on shape information

    2. based on firing pattern information

  • 8/6/2019 Emg Decomposition

    14/42

    The shape information decision functions include:

    1 .Normalized shape certainty CND : represents the distance

    from a candidate MUP mjto the template of a MUPT i1.

    2.Relative shape certainty CRD :represents the distance from a

    candidate MUP mjto the template of the closest MUPT relative to

    the distance from the same MUP to the second closest MUPT.

    1.

    1.

  • 8/6/2019 Emg Decomposition

    15/42

    The firing pattern information is presented by the firing certainty

    based function CFC with respect to the established firing pattern of

    the MUPT

    The decision of assigning a MUP mjto MUPT i is based on the value

    of overall certainty

    where i=1,2.

    MUPT mj is assigned to MUPT i if overall certainty is the

    greatest and if it is greater than the minimal certainty threshold

    (Cm) for which a classification is to be made.

    Other wise MUP mj is left unassigned.

    j

    iC

    . . j j j ji NDi RDi FCiC C C C =

  • 8/6/2019 Emg Decomposition

    16/42

    FUZZY k-NN CLASSIFIER The fuzzy k-NN classifier uses a fuzzy non-parametric classification

    procedure based on the nearest neighbor classification rule.

    By passes probability estimation and goes directly to decision

    functions.

    It determines for each candidate MUP mj a MUPT i membershipi

    (mj) representing the shape based strength of membership of mj

    in MUPT class i

    Also determines firing assertion decision function assesingthe time

    of occurrence of MUPT mj with respect to the established firing

    pattern of MUPT class i

    j

    FAiA

  • 8/6/2019 Emg Decomposition

    17/42

    The overall assertion value for assigning MUP mj to MUPT class

    i is defined as:

    MUP mj is assigned to the MUPT class i with the highest assertion

    value and if this value is above the minimum assertion value

    threshold (Am) of the MUPT i to which a classification is to be

    made.

    Otherwise MUP mjis left unassigned.

    j

    iA

    ( ).i

    j ji j FAi A m A=

  • 8/6/2019 Emg Decomposition

    18/42

    MATCHED TEMPLATE FILTER CLASSIFIERThis method uses correlation measure as an estimate of the degree

    of similarity between MUP and MUPT templates.

    Two matched template filters are used for supervised MUP

    classification

    1. normalized cross correlation

    2. pseudo- correlation

    The MTF classifier also determines for MUP mja firing time similarity

    decision function with respect to the established firing pattern

    of the MUPT

    jFSi

    S

  • 8/6/2019 Emg Decomposition

    19/42

    The decision of assigning a MUP to a MUPT is based on the overall

    similarity function

    Where

    MUPT mj is assigned to MUPT iif the value of is greatest and it is

    greater than minimal similarity threshold (sm) for which a

    classification is to be made.

    Otherwise MUP mjis left unassigned.

    ( ).i

    j j j

    i FSiS x S

    =

    j

    iS

    j

    iS

  • 8/6/2019 Emg Decomposition

    20/42

    Multiple Classifier Approaches

    Fig.6.Classifier fusion system basicarchitecture[6]

    Classifier fusion system architecture belongs to theparallel category of combining classifiers.

  • 8/6/2019 Emg Decomposition

    21/42

    The decision aggregation module in a classifier fusion system

    combines the base classifier outputs to achieve a group

    consensus.

    Majority Voting Aggregation

    A MUP x is classified to belong to MUPT class i if over half ofthe classifiers say x i .

    Average rule aggregation

    Combines the set of decision confidences

    1

    ( )

    ( )

    K

    ik

    k

    i

    Cf x

    Q xK

    ==

    1( ) argmax ( ( ))M

    i i x Q x= =

  • 8/6/2019 Emg Decomposition

    22/42

    One-Stage Classifier Fusion Does not contain the ensemble members selection module

    Uses a fixed set of base classifiers

    Choosing base classifiers can be performed directly through

    exhaustive search with the performance of the fusion being

    the objective function.

    As the number of base classifiers increases, this approach

    becomes computationally too expensive.

  • 8/6/2019 Emg Decomposition

    23/42

    Diversity-Based One-Stage Classifier Fusion Contains an ensemble members selection module

    The ensemble choice module selects the subsets of classifiers

    that can be combined to achieve better accuracy

    The kappa statistic is used to select base classifiers having an

    excellent level of agreement to form ensembles having

    satisfactory classification performance.

  • 8/6/2019 Emg Decomposition

    24/42

    Hybrid Classifier Fusion Does not contain the ensemble members selection module

    Uses a fixed set of base classifiers

    It uses a hybrid aggregation module which is a combination of

    two stages of aggregation

    The first aggregator is based on the abstract level and the

    second is based on the measurement level

  • 8/6/2019 Emg Decomposition

    25/42

    The hybrid aggregation scheme works as follows:

    First stage:

    The outputs of the ensemble of classifiers are presented to

    the majority voting aggregator.

    If all classifiers state a decision that a MUP pattern is left

    unassigned, then it stays unassigned.

    If over half of the classifiers assign a MUP pattern to the sameMUPT class, then that MUP pattern is allocated to that MUPT

    class and no further assignment is processed.

  • 8/6/2019 Emg Decomposition

    26/42

    Second stage:

    This stage is activated for those MUP patterns for which only

    half or less than half of the ensemble of classifiers in the

    first stage specify a decision for a MUP pattern to be

    assigned to the same MUPT class.

    The outputs of the ensemble of classifiers are presented to

    the average rule aggregator, or the trainable aggregator

    represented by the Sugeno fuzzy integral.

  • 8/6/2019 Emg Decomposition

    27/42

    For each MUP pattern, the overall combined confidence

    values representing the degree of membership in eachMUPT class are determined

    MUP pattern is assigned to the MUPT class for which its

    determined overall combined confidence is the largest and

    if it is above the specified aggregator confidence threshold

    set for that MUPT class

    otherwise the MUP pattern is left unassigned

  • 8/6/2019 Emg Decomposition

    28/42

    Diversity-Based Hybrid Classifier FusionThe diversity-based hybrid classifier fusion scheme is a two-

    stage process

    consists of two aggregators with a pre-stage classifier

    selection module for each aggregator.

    The ensemble candidate classifiers selected for aggregationare decided through assessing the degree of agreement

    using the kappa statistic measure

  • 8/6/2019 Emg Decomposition

    29/42

    The diversity-based hybrid fusion scheme works as follows:

    First stage:

    classifiers selected for aggregation by the first aggregator are

    those having the maximum degree of agreement

    The outputs of the classifiers are presented to the majority

    voting aggregator.

    If all the classifiers state a decision that a MUP pattern is left

    unassigned it stays unassigned.

  • 8/6/2019 Emg Decomposition

    30/42

    Second stage:

    This stage is used for those MUP patterns for which only half

    or less than half of the ensemble classifiers in the first

    stage specify a decision for a MUP pattern to be assigned to

    the same MUPT class

    classifiers selected for aggregation at the second combiner

    are those having a minimum degree of agreement

    considering only the unassigned category

  • 8/6/2019 Emg Decomposition

    31/42

    The outputs of the classifiers are presented to the average

    rule aggregator or the trainable aggregator represented by

    Sugeno fuzzy integral aggregator.

    For each MUP pattern, the overall combined confidence

    values representing the degree of membership in each

    MUPT class are determined

    MUP pattern is assigned to the MUPT class for which its

    determined overall combined confidence is the largest and

    if its above the specified aggregator confidence threshold

    set for that MUPT class.

  • 8/6/2019 Emg Decomposition

    32/42

    Comparative Study The single classifier and multi-classifier approaches are

    compared in terms of the difference between the correctclassification rate CCr% and error rate Er%.

    number of MUPs erroneously classified

    100number of MUPs assignedrE=

    number of MUPs correctly classified100

    total number of MUPs detectedr

    CC =

  • 8/6/2019 Emg Decomposition

    33/42

    The base classifiers used for experimentation are

    1. four ACC classifiers e1,e2, e3, e4

    2. four AFNNC classifiers e5,e6,e7,e8

    3. four ANCCC classifiers e9,e10 ,e11 ,e12

    4. four ApCC classifiers e13 ,e14 ,e15 ,e16

  • 8/6/2019 Emg Decomposition

    34/42

    Classifiers e1, e5, e9, e13 werefed with time-domain

    first-order discrete derivative features

    Classifiers e2, e6, e10 , e14 were fed withtime-

    domain first-order discrete derivative features

    Classifiers e3, e7, e11 , e15 werefedwith wavelet-

    domain first order discrete derivative features

    Classifiers e4, e8, e12 , e16 were fed with wavelet-

    domain first-order discretederivative features

  • 8/6/2019 Emg Decomposition

    35/42

    Table .1 Mean and mean absolute deviation (MAD) of the difference betweencorrect classification rate CCr and error rate Er for the different single classifier

    approaches across the three EMG signal data sets[9]

    C la ssifie r In d e p e n d e n t

    sim u la te d sig n a lsR e la te d

    sim u la te d sig n a lsR ea lsig n als

    e 1 . ( . )8 1 9 4 9 . ( . )7 5 0 2 5 . ( . )7 8 5 0 9

    e 2 . ( . )83 9 4 9 . ( . )76 5 1 8 . ( . )72 0 0 3

    e3

    . ( . )82 3 4 1 . ( . )75 6 1 6 . ( . )71 9 1 4

    e4 . ( . )84 7 4 0 . ( . )76 4 1 5 . ( . )67 1 1 4

    e5 . ( . )85 2 2 5 . ( . )73 7 1 0 . ( . )80 9 0 9

    Bestsingle classifier e6

    . ( . )90 4 1 7 . ( . )80 7 2 3 . ( . )77 5 2 6

    e7 . ( . )83 1 2 4 . ( . )73 3 0 5 . ( . )73 5 0 4

    e8 . ( . )88 9 1 5 . ( . )79 0 1 8 . ( . )73 4 2 4

    Average of 8 singleclassifiers

    . ( . )85 0 3 3 . ( . )76 2 1 6 . ( . )74 3 1 2

  • 8/6/2019 Emg Decomposition

    36/42

    Table .1 continues..e 9 . ( . )79 3 3 2 . ( . )59 2 0 4 . ( . )69 0 1 3

    e10 . ( . )80 6 3 1 . ( . )54 6 2 6 . ( . )63 0 0 7

    e11 . ( . )76 1 2 7 . ( . )59 4 1 0 . ( . )58 7 0 4

    e12 . ( . )77 7 2 7 . ( . )56 0 0 8 . ( . )49 5 0 3

    e13 . ( . )78 0 3 4 . ( . )62 6 0 7 . ( . )71 3 0 4

    e14 . ( . )77 8 2 8 . ( . )59 4 0 9 . ( . )66 1 2 7

    e15 . ( . )77 5 2 9 . ( . )62 2 0 1 . ( . )68 0 1 5

    e16 . ( . )77 6 2 2 . ( . )58 9 0 9 . ( . )56 6 3 6

    Average of 16 singleclassifiers

    . ( . )81 5 0 1 . ( . )67 7 0 4 . ( . )68 5 0 0

  • 8/6/2019 Emg Decomposition

    37/42

    Table .2 Mean and mean absolute deviation (MAD) of the difference between correctclassification rate CCr and error rate Er for the different single classifier and multi-classifier

    approaches across the three EMG signal data sets

    cla ssifie r In de p en de nt sim u latedsig n a ls

    R e lated

    sim u la te d sig n a lsR e al sig n als

    S in g le cla ssifie rs

    W ea ke st of 8 Sin gleC la ssifie rs e 4

    . ( . )84 7 4 0 . ( . )76 4 1 5 . ( . )67 1 1 4

    Weakest of 16 Single

    Classifier e12. ( . )77 7 2 7 . ( . )56 0 0 8 . ( . )49 5 0 3

    Best Single Classifier e6 . ( . )90 4 1 7 . ( . )80 7 2 3 . ( . )77 5 2 6

    Average of 8 SingleClassifiers

    . ( . )85 0 3 3 . ( . )76 2 1 6 . ( . )74 3 1 2

    Average of 16 SingleClassifier

    . ( . )81 5 0 1 . ( . )67 7 0 4 . ( . )68 5 0 0

  • 8/6/2019 Emg Decomposition

    38/42

    Table .2 continues

    -O n e S ta g e C la ssifie r Fu sio n [ ]2

    (Majority Voting fixed of)8

    . ( . )86 0 4 6 . ( . )79 2 3 0 . ( . )77 3 4 3

    (Average Fixed Rule fixed)of 8

    . ( . )88 0 2 5 . ( . )82 0 0 7 . ( . )85 1 1 2

    Sugeno Fuzzy Integral

    ( )fixed of 8

    . ( . )82 3 2 7 . ( . )78 3 1 9 . ( . )80 9 2 9

    - - [ ]Diversity based One Stage Classifier Fusion 2

    /Majority Voting 6 8 . ( . )87 6 4 2 . ( . )80 1 2 7 . ( . )78 8 4 8

    /Average Fixed Rule 6 8 . ( . )88 5 2 2 . ( . )82 1 1 1 . ( . )84 9 0 8

    /Sugeno Fuzzy Integral 6 8 . ( . )84 6 2 4 . ( . )80 2 1 1 . ( . )82 0 0 8

  • 8/6/2019 Emg Decomposition

    39/42

    AMVAFR, ADMVAFR stands for Adaptive (or Diversity-based) Majority Voting

    with Average Fixed Rule hybrid classifier fusion scheme, respectively.AMVSFI, ADMVSFI stands for Adaptive (or Diversity-based) Majority Voting withSugeno Fuzzy Integral one-stage classifier fusion scheme, respectively.

    [ ]H y b rid C la ssifie r Fu sio n 7

    ( )A M V A F R fix e d o f 6 . ( . )9 1 8 1 8 . ( . )8 4 6 1 3 . ( . )8 2 7 2 5

    ( )A M VS FI fixed of 6 . ( . )9 1 8 1 8 . ( . )8 4 6 1 3 . ( . )8 2 5 1 7

    - [ ]D ive rsity b a se d H yb rid C la ssifie r Fu sio n 6

    /A D M V A FR 6 8 . ( . )9 1 6 1 8 . ( . )8 4 4 0 7 . ( . )8 5 5 0 9

    /A D M V S F I 6 8 . ( . )9 1 2 1 8 . ( . )8 4 0 0 8 . ( . )8 5 2 0 9

    /A D M V A FR 6 1 6 . ( . )9 0 0 3 2 . ( . )8 3 2 0 7 . ( . )8 3 7 0 6

    /A D M V S F I 6 1 6 . ( . )8 9 6 3 3 . ( . )8 2 5 0 8 . ( . )8 2 8 0 4

    Table .2 continues..

  • 8/6/2019 Emg Decomposition

    40/42

    Conclusion one-stage aggregator classifier fusion and its diversity-based

    variant schemes have performance better than the average

    performance of the base classifiers

    Better than the performance of the best base classifier except

    across the independent simulated signals.

    The hybrid classifier fusion and its diversity-based variant

    approaches have performances that not only exceed the

    performance of any of the base classifiers forming the

    ensemble but also reduced classification errors for all data

    sets studied

  • 8/6/2019 Emg Decomposition

    41/42

    References1.Rasheed S, Stashuk D and Kamel M (2008) Fusion of multiple

    classifiers for motor unit potential sorting. BiomedicalSignal Processing and Control, 3(3):229243

    2.Rasheed S, Stashuk D and Kamel M (2008) Diversity-based

    combination of non-parametric classifiers for EMG signaldecomposition. Pattern Analysis & Applications, 11:385408

    3.Stashuk D W (2001) EMG signal decomposition: how can it beaccomplished and used? Journal of

    Electromyography and Kinesiology, 11:1511734.McGill K C, Cummins K and Dorfman L J (1985) Automatic

    decomposition of the clinical electromyogram. IEEETransactions on Biomedical Engineering, 32(7):470477

    5.

    6.

  • 8/6/2019 Emg Decomposition

    42/42

    5. Nikolic M, Srensen J A, Dahl K et al. (1997) Detailed analysis ofmotor unit activity. In Proceedings of the 19th Annual

    International Conference of the IEEE Engineering in Medicineand Biology Society Conference, 12571260

    6.Sarbast Rasheed, Daniel W. Stashuk, Mohamed S.Kamel(2002) Integrating Heterogeneous ClassifierEnsembles for EMG Signal Decomposition Based onClassifier Agreement IEEE transactions on informationtechnology in biomedicine, 1(11):1-17

    7. Rasheed S, Stashuk D and Kamel M(2007) A hybrid classifierfusion approach for motor unit potential classification duringEMG signal decomposition. IEEE Transactions on BiomedicalEngineering, 54(9):17151721

    8. Roberto Merletti, Philip Parker (2004) Electromyography:Physiology, Engineering, and Non-Invasive Applications.Wiley-IEEE Press, 1st edition

    9. Amine Nait-ali (2009) Advanced Biosignal Processing . Springer,1st edition