discuscrowds (image '14, dresden, selected slides).pdf

Upload: paul-h-artes

Post on 07-Jul-2018

215 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    1/29

    Silk Purses, Sow’s Ears

    IMAGE meeting

    12-14 March 2014, Schloss Eckberg, Dresden

    Paul H Artes, PhDOphthalmology and Visual Sciences, Dalhousie University

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    2/29

    How damaged are these discs?

    What is the likelihood of damage? (0-100%)What is the amount of damage (neural tissue, vision)?

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    3/29

    How good are these observers?

    ...distinguish between discs of healthy people & patients

    ...match decision criteria to prevalence.

     you? 

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    4/29

    Discus, GONE, others

    www.gone-project.com

    www.discusproject.blogspot.com

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    5/29

    V RI BILITY

    OF EXPERT OBSERVERS IN

    EV LU TING T

    OPTIC

    DISC

    BYPaul R.

    Lichter MD

    INTRODUCTION

    CORRECT

    EVALUATION

    OF

    THE

    OPTIC

    DISC

    IS FUND MENT L TO THE

    DIAGNOSIS

    and

    management

    of

    g l au co ma . Y e t

    this is sometimes

    an extremely

    difficult

    t sk

    Numerous attempts have been

    made to

    describe what is normal

    or

    abnormal about an optic disc

    yet

    no

    precise

    means is

    available

    to

    deter

    mine the

    status

    of an optic

    disc

    with

    reference to

    glaucoma.

    Descriptions

    and measurements

    used

    by

    some

    individuals

    have

    no t been utilized

    by

    others. With this in

    mind a

    study of

    optic disc evaluation

    was

    under

    taken. The

    study was

    particularly aimed at testing the

    validity

    of

    disc

    evaluation utilizing nonstereo

    and

    stereo

    photographs.

    The

    purpose

    of

    this

    report

    is

    to

    analyze the observer

    variability

    of

    16 glaucoma

    experts

    in

    evaluating the

    optic disc. This

    analysis includes

    the

    determination ofcup/disc ratio as well as

    the

    estimate of

    the

    probability

    of pathologic

    glaucomatous change

    in

    a

    series

    of optic

    discs.

    B CKGROUND

    Ever

    since the invention

    of

    the ophthalmoscope in

    1851

    the

    optic

    disc

    and

    its

    relationship

    to

    glaucoma

    has

    intrigued

    many

    an

    ophthalmologist.

    In

    1855 Weber

    stated

    that the optic disc

    was

    cupped in

    glaucoma

    rather

    than

    swollen or bulging as previously

    believed.

    Through the next

    several

    decades little

    was added to

    knowledge

    about th e disc

    in

    glaucoma.

    At

    the

    turn

    of the century Elschnig2 noted th e

    variable

    appearance

    of

    normal discs.

    He described five

    types

    of discs and

    this

    description

    h as g en er al ly s to od

    the test of

    time.

    However, this classification

    of

    normal discs

    is

    not widely used

    today.

     From the Department

    of

    Ophthalmology

    University

    of

    Michigan Medical School

    nn

    Arbor Michigan.

    TR.

    AM. OPHTH. Soc. vol.

    LXXIV, 1976

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    6/29

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    7/29

    positive rate in VF!negative group

      p  o  s   i   t   i  v  e

      r  a   t  e

       i  n

       V   F   !  p  o  s   i   t   i  v  e

      g  r  o  u  p

    0.0 0.2 0.4 0.6 0.8 1.0

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    A

    BC

    D

    AUC=0.8

    83%

    r=0.75

    mean.Diff=0.5

    criterion=!0.56

    A

     

    0

    5

    10

    15

    20

    25

    30

    response

      r  e  s  p  o  n  s  e

      a

      e  n  c  y

      s

    dh ph ns pd dd

    0

    5

    10

    15

    20

    25

    30

    40   24 2   20   14

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    8/29

    response time - ceiling at 20 secs.1026 submissions,674 observers*403 IP addresses* A single person could have generated several IDs. If several attempts were made under the sameID, those responses were averaged (per image).

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    9/29

    positive rate in VF!negative group

      p  o  s   i   t   i  v  e

      r  a   t  e

       i  n

       V   F

       !  p  o  s   i   t   i  v  e

      g  r  o  u  p

    0.0 0.2 0.4 0.6 0.8 1.0

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    A

    BC

    D

    AUC=0.5

    !1%

    r=0.3

    mean.Diff=0.96

    criterion=!0.04

    A

     

    0

    5

    10

    15

    20

    25

    30

    response

      r  e  s  p  o  n  s  e   l  a

       t  e  n  c  y   (  s   )

    dh ph ns pd dd

    0

    5

    10

    15

    20

    25

    30

    7 46   1   3 6 10

    positive rate in VF!negative group

      p  o  s   i   t   i  v  e

      r  a   t  e

       i  n

       V   F   !  p  o  s   i   t   i  v  e

      g  r  o  u  p

    0.0 0.2 0.4 0.6 0.8 1.0

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    A

    B

    C

    D

    AUC=0.85

    96%

    r=0.7

    mean.Diff=0.58

    criterion=!0.61

    A

     

    0

    5

    10

    15

    20

    25

    30

    response

      r  e  s  p  o  n  s  e

       l  a   t  e  n  c  y   (  s   )

    dh ph ns pd dd

    0

    5

    10

    15

    20

    25

    30

    37 23 15 14 11

    removed ~5% of data,2.5% from either extremes.

    Area under ROC curve

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    10/29

    positive rate in VF!negative group

      p

      o  s   i   t   i  v  e

      r  a   t  e

       i  n

       V   F   !  p  o  s   i   t   i  v  e

      g  r  o  u  p

    0.0 0.2 0.4 0.6 0.8 1.0

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    A

    B

    C

    D

    AUC=0.77

    73%

    r=0.55

    mean.Diff=0.5

    criterion=!1.2

    A

     

    11

    positive rate in VF!negative group

      p

      o  s   i   t   i  v  e  r  a   t  e   i  n

       V   F   !  p  o  s   i   t   i  v  e

      g  r  o  u  p

    0.0 0.2 0.4 0.6 0.8 1.0

    0.0

    0.2

    0.4

    0.6

    0.8

    1.0

    A

    B

    C

    D

    AUC=0.8

    82%

    r=0.66

    mean.Diff=0.69

    criterion=!0.04

    A

     

    11 1

       o     b

       s   e   r   v   e   r   s

    Two observers with similar “skill” (ie, discriminatoryperformance, but vastly different decision criteria…

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    11/29

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    12/29

    average (0.69)

    “voting” (0.82)   o     b   s   e   r   v   e   r   s

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    13/29

         i   m   a   g   e   s

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    14/29

         i   m   a   g   e   s

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    15/29

         i   m   a   g   e   s

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    16/29

         i   m   a   g   e   s

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    17/29

         i   m   a   g   e   s

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    18/29

         i   m   a   g   e   s

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    19/29

         i   m   a   g   e   s

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    20/29

         i   m   a   g   e   s

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    21/29

         i   m   a   g   e   s

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    22/29

         i   m   a   g   e   s

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    23/29

      C   E   N

       S  O   R

       E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    24/29

      C   E   N   S

      O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    25/29

      C   E   N   S

      O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    26/29

      C   E   N   S  O

       R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    27/29

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    28/29

      C   E   N

       S  O   R   E   D

  • 8/19/2019 DiscusCrowds (IMAGE '14, Dresden, selected slides).pdf

    29/29

      C   E   N

       S  O   R   E   D