QMic F1 20121009 ut - cb.uu.secarolina/QMicht2012/QMic_F1_20121009.pdf20121009 & 3...

9
20121009 1 PhDlevel course in Quan6ta6ve Microscopy In collabora6on between the Centre for Image Analysis, SciLifeLab and BioVis OctoberDecember 2012 Introduc6on and overview 2012 10 09 Carolina Wählby Associate professor in Quan6ta6ve Microscopy at SciLifeLab/the Centre for Image Analysis, Uppsala University and Principal Inves6gator, the Imaging PlaNorm of the Broad Ins6tute of Harvard and MIT, Cambridge; MA Carolina Wählby, [email protected] PreTy pictures, or measurements? Fundamental steps when extracting information from image data Image acquisition knowledge about the applica6on Carolina Wählby, [email protected] Preprocessing, filtering Making measurements, feature extraction LENGTH, WHIDTH, CURVATURE, TEXTURE… Object classification, interpretation, recognition Result WORMS ARE ALIVE Object detection, segmentation (including 3D and tracking over time) Course ‘themes’ A beTer communica6on between experts of from different fields can greatly improve the scien6fic value of an experiment. Carolina Wählby, [email protected] Knowledge about the image forma6on, possibili6es and limita6ons, can greatly improve the scien6fic value of an experiment. A beTer understanding of digital image processing, possibili6es and limita6ons, can greatly improve the scien6fic value of an experiment. Today: Some basic concepts in digital image processing and analysis and quan6ta6ve microscopy. An overview of the course. Introduc6on of course par6cipants. Ini6al discussion of projects. Carolina Wählby, [email protected] Carolina Wählby, [email protected] Pixels and image dimensions 2dimensional (2D) gray scale image. f(x,y)= f(1,1) f(1,2) f(1,3)…. f(2,1) f(2,2) f(2,3)… where f(1,1)=17, f(1,2)=16, f(1,3)=16 etc The value (or intensity) of a pixel is a func6on of the posi6on (x,y). f(1,1)

Transcript of QMic F1 20121009 ut - cb.uu.secarolina/QMicht2012/QMic_F1_20121009.pdf20121009 & 3...

Page 1: QMic F1 20121009 ut - cb.uu.secarolina/QMicht2012/QMic_F1_20121009.pdf20121009 & 3 Temporal&proper6es&& • A&digital&image&is&characterized&by&&Is&itpossible&to&image&live&samples,&and&whatare&the&

2012-­‐10-­‐09  

1  

PhD-­‐level  course  in    Quan6ta6ve  Microscopy  

In  collabora6on  between  the  Centre  for  Image  Analysis,  SciLifeLab  and  BioVis  October-­‐December  2012  

Introduc6on  and  overview  2012  10  09  

Carolina  Wählby  Associate  professor  in  Quan6ta6ve  Microscopy  at  SciLifeLab/the  Centre  for  Image  Analysis,  Uppsala  University  

and  Principal  Inves6gator,  the  Imaging  PlaNorm  of  the  Broad  Ins6tute  of  Harvard  and  MIT,  Cambridge;  MA  

Carolina  Wählby,  [email protected]  

PreTy  pictures,  or  measurements?   Fundamental steps when extracting information from image data

Image acquisition

knowledge  about  the  applica6on  

Carolina  Wählby,  [email protected]  

Preprocessing, filtering

Making measurements, feature extraction

LENGTH,  WHIDTH,  CURVATURE,  TEXTURE…  

Object classification, interpretation, recognition

Result WORMS  ARE  ALIVE  

Object detection, segmentation (including 3D and tracking over time)

Course  ‘themes’  

A  beTer  communica6on  between  experts  of  from  different  fields  can  greatly  improve  the  scien6fic  value  of  an  experiment.  

Carolina  Wählby,  [email protected]  

Knowledge  about  the  image  forma6on,  possibili6es  and  limita6ons,  can  greatly  improve  the  scien6fic  value  of  an  experiment.    

A  beTer  understanding  of  digital  image  processing,  possibili6es  and  limita6ons,  can  greatly  improve  the  scien6fic  value  of  an  experiment.    

Today:  

•  Some  basic  concepts  in  digital  image  processing  and  analysis  and  quan6ta6ve  microscopy.  

•  An  overview  of  the  course.  •  Introduc6on  of  course  par6cipants.  •  Ini6al  discussion  of  projects.  

Carolina  Wählby,  [email protected]   Carolina  Wählby,  [email protected]  

Pixels  and  image  dimensions  

2-­‐dimensional  (2D)  gray  scale  image.  

f(x,y)=    f(1,1)  f(1,2)  f(1,3)….      f(2,1)  f(2,2)  f(2,3)…      …  

where  f(1,1)=17,  f(1,2)=16,  f(1,3)=16  etc  

The  value  (or  intensity)  of  a  pixel  is  a  func6on  of  the  posi6on  (x,y).  

f(1,1)  

Page 2: QMic F1 20121009 ut - cb.uu.secarolina/QMicht2012/QMic_F1_20121009.pdf20121009 & 3 Temporal&proper6es&& • A&digital&image&is&characterized&by&&Is&itpossible&to&image&live&samples,&and&whatare&the&

2012-­‐10-­‐09  

2  

Resolu6on  -­‐a  word  with  many  defini6ons  

Resolu6on  of  a  screen:  the  number  of  pixels  that  fit  in  the  x-­‐  and  y-­‐dimension.  It  is  also  common  to  say  that  the  resolu6on  of  an  image  is  its  size  in  pixels  (NOT  correct  defini6on  during  this  course).    

Carolina  Wählby,  [email protected]  

Resolu6on  of  an  image  is  the  shortest  distance  (in  μm)  between  two  points  on  a  specimen  (observed  as  Airy  disks)  that  can  s6ll  be  dis6nguished  (in  the  digital  image)  as  separate  en66es.    

Illustra6on  borrowed  from  hTp://www.microscopyu.com  

Pixel  size  is  the  distance  in  the  specimen  (in  μm)  corresponding  to  the  distance  between  two  pixels  in  the  digital  image.    Note  that  the  pixel  size  of  the  digital  camera  has  to  be  small  enough  not  to  limit  the  resolu6on  of  the  op6cal  system.  However,  a  smaller  pixel  size  can  not  increase  the  resolu6on  beyond  the  resolu6on  limit  of  the  op6cal  system.   Carolina  Wählby,  [email protected]  

A  2D  digital  image  with  3  color  channels  

=  

Red  channel,  λ=1   Green  channel,  λ=2   Blue  channel,  λ=3  

f(x,y,λ)=    f(1,1,1)  f(1,2,1)  f(1,3,1)….      f(2,1,1)  f(2,2,1)  f(2,3,1)…      …  

where  f(1,1,1)=17,  f(1,2,1)=16,  f(1,3,1)=16    while    f(1,1,2)=0,  f(1,2,2)=0  etc  

The  value  (or  intensity)  of  a  pixel  is  a  func6on  of  the  posi6on  x,y  and  spectral  channel  λ.  

Spectral  informa6on:  color  images.  Densitometric  and  spectral  proper6es  

•  What  physical  property  of  the  specimen  is  translated  to  image  intensity,  and  how  precise  is  this  transla6on?    

•  How  can  sample  prepara6on  affect  the  densitometric  proper6es,  and  what  are  the  sources  of  variability  and  noise?    

•  Is  it  possible  to  op6mize  signal/noise  proper6es  by  adjus6ng  exposure  6me  or  similar,  and  what  are  the  drawbacks?    

•  Can  image  data  be  normalized  or  the  system  be  calibrated?    •  What  are  the  spectral  proper6es  of  the  imaging  system  (if  any)?    •  What  physical  proper6es  can  be  quan6fied  using  the  spectral  

informa6on,  and  what  are  the  sources  of  varia6on  (sample  treatment,  exposure  6mes,  temperature  etc).    

Carolina  Wählby,  [email protected]  

Carolina  Wählby,  [email protected]  

3-­‐dimensional  (3D)  images.  

3D  informa6on  can  be  collected  using  different  focal  depths  (e.g.  confocal  and  two-­‐photon  microscopy)  or  a  tomographic  imaging  system.      The  intensity  is  now  a  func6on    f(x,y,z,λ).    Note  that  the  resolu6on  in  the  z-­‐dimension  is  rarely  the  same  as  the  resolu6on  in  x-­‐  and  y.  A  3D  pixel  is  onen  called  a  voxel.  

Geometric  proper6es  

•  How  are  the  three  spa6al  dimensions  of  the  real  world  transferred  to  the  dimensions  of  the  digital  image?    

•  Is  it  a  projec6on,  a  slice  through  the  object  or  a  surface?    

•  How  exact  is  the  imaging?    •  Can  geometric  proper6es  of  the  imaged  object  be  distorted  affec6ng  quan6ta6ve  measurements?    

•  What  resolu6on  can  be  achieved,  and  what  physical  proper6es  limit  the  resolu6on?    

•  Is  the  resolu6on  the  same  in  all  spa6al  dimensions?    

Carolina  Wählby,  [email protected]  

Time-­‐lapse  images  (movies)  t=0   t=1   t=2  

If  we  follow  dynamic  events,  we  add  a  6me  dimension:  f(x,y,z,λ,t)  

Example:  6me  series  (movie)  of  dividing  cell.  Spa6al  informa6on:  x,y,z  (z  always=1).  Spectral  informa6on:  brighNield  +  fluorescent  channel  Temporal  informa6on:  images  captured  every  10  minutes  (temporal  resolu6on  =  10  minutes)      

Page 3: QMic F1 20121009 ut - cb.uu.secarolina/QMicht2012/QMic_F1_20121009.pdf20121009 & 3 Temporal&proper6es&& • A&digital&image&is&characterized&by&&Is&itpossible&to&image&live&samples,&and&whatare&the&

2012-­‐10-­‐09  

3  

Temporal  proper6es    

•  Is  it  possible  to  image  live  samples,  and  what  are  the  limita6ons  when  it  comes  to  sample  prepara6on  and  handling?    

•  What  is  the  temporal  resolu6on  in  rela6on  to  signal  and  noise?  

 

Carolina  Wählby,  [email protected]  

image  forma6on  

A  digital  image  is  characterized  by    •  geometric  proper6es  and  resolu6ons  (x,y,z)  •  densitometric  and  spectral  proper6es  and  resolu6ons  (λ)  

•  temporal  proper6es  and  resolu6ons  (t)  of  the  imaging  system.    An  image  is  also  characterized  by  the  prepara6on  of  the  imaged  sample.  

Carolina  Wählby,  [email protected]  

Lectures  on  image  forma6on,  possibili6es  and  limita6ons  

•  Bright  field  microscopy,  slide  scanners  and  staining  techniques.  Anna  Asplund,  IGP,  UU.    •  Fluorescence,  confocal  and  two-­‐photon  microscopy.  Fluorescence  labeling  and  staining.  

Dirk  Pacholsky,  BioVis,  IGP,  UU.  •  Spectral  aspects  in  microscopy;  the  rela6on  between  illumina6on,  absorp6on,  emission  

and  reflec6on  in  rela6on  to  imaging  sensors  features.  Ewert  Bengtsson,  CBA,  UU.    •  Live  cell  imaging.  Göran  Månsson,  CLICK  facility,  KI.  •  Transmission  Electron  Microscopy  (TEM)  and  biological  sample  prepara6on.  Anders  

Ahlander,  BioVis,  UU.    •  Scanning  Electron  Microscopy  (SEM),  Atomic  Force  Microscopy  (AFM)  and  sample  

prepara6on  in  material  science.  Åsa  Kassman  Rudolphpi,  Dept.  Eng.  Science,  UU.    •  S6mulated  emission  deple6on  microscopy  (STED).  Daniel  Rönnlund,  Exp.  Biomolecular  

Physics,  KTH.    •  X-­‐ray  microtomography  and  op6cal  tomography.  Alexandra  Pacureanu,  CBA,  UU.    •  High-­‐throughput  imaging  systems  and  quan6ta6ve  microscopy  in  pharma-­‐industry.  Alan  

Sabirsh,  AstraZeneca.    

Carolina  Wählby,  [email protected]  

fundamental steps in image processing and analysis

Image acquisition

knowledge  about  the  applica6on  

Carolina  Wählby,  [email protected]  

Preprocessing, filtering

Making measurements, feature extraction

LENGTH,  WHIDTH,  CURVATURE,  TEXTURE…  

Object classification, interpretation, recognition

Result WORMS  ARE  ALIVE  

Object detection, segmentation (including 3D and tracking over time)

Carolina  Wählby,  [email protected]  Carolina  Wählby,  [email protected]  

Image  segmenta6on  -­‐to  par66on  an  image  into  regions  of  interest  

2-­‐dimensional  (2D)  gray  scale  image.  

Thresholding:  define  an  object  as  connected  pixels  brighter  than  a  fixed  intensity  threshold  f(x,y)>T  

T=20  

Carolina  Wählby,  [email protected]  

Image  segmenta6on  -­‐to  par66on  an  image  into  regions  of  interest  (ROI)  

What  is  the  best  intensity  threshold  value  for  dividing  the  intensity  histogram  into  foreground  and  background  pixels?  

Here?  

Or  here?  

Pixel  intensity  

Freq

uency  

raw  input    image  

binary  image:  0=background  1=objects  

labeled  objects:  each  connected  component  is  a  ROI,  represented  by  a  random  color.  

Page 4: QMic F1 20121009 ut - cb.uu.secarolina/QMicht2012/QMic_F1_20121009.pdf20121009 & 3 Temporal&proper6es&& • A&digital&image&is&characterized&by&&Is&itpossible&to&image&live&samples,&and&whatare&the&

2012-­‐10-­‐09  

4  

Clustered  and  overlapping  objects  may  be  further  separated  using  more  advanced  segmenta6on  methods  that  make  use  of  more  informa6on,  such  as  shape,  edges  and  intensity  distribu6ons  within  the  regions  of  interest.    

Input:   segmented  blobs  (from  red)  segmented  nuclei  (from  blue)  

Example:  Coun6ng  ‘blobs’  per  cell  by  using  the  segmenta6on  result  from  one  image  channel  to  extract  informa6on  from  another  channel.  

Outline  of  ‘cytoplasms’   blobs  assigned  to  ‘cytoplasm’  

summary:  Image  segmenta6on  Segmenta6on  is  onen  the  most  difficult  problem  to  solve  in  image  analysis.    No  universal  solu6on  exist,  and  this  is  a  large  and  ac6ve  field  of  research.    The  result  of  an  analysis  task  relies  heavily  on  the  robustness  of  the  image  segmenta6on  (including  3D  segmenta6on  and  tracking).    The  problem  can  be  simplified  by  choice/op6mizing  the  sample  prepara6on  and  seyngs  of  the  imaging  system  •  choice  of  sensor  •  Illumina6on  •  background  

Carolina  Wählby,  [email protected]  

Once the image is segmented and the ROIs have been identified: Making measurements; feature extraction

•  Topological features •  Size and shape features

–  Based on a binary mask of object/objects •  Intensity features

–  Based on histogram of object/objects and color/multiple fluorescent channels

•  Texture features –  Statistics and spatial gray scale variations

•  Relational/structural features –  How objects (parts of objects) relate to one another

•  Counts –  How many sub-object (dots) per main object

Note that image segmentation may not always be necessary: features can also be measured on a per-image basis.

Carolina  Wählby,  [email protected]  

Object classification, interpretation and recognition

Carolina  Wählby,  [email protected]  

 Example:  Define  nuclei  by  intensity  thresholding  of  blue  image  channel.    Define  a  cytoplasms  as  a  region  surrounding  a  nucleus  at  a  fixed  distance.    x1=mean  pixel  intensity  in  green  channel  within  cytoplasm.  x2=mean  pixel  intensity  in  green  channel  within  nucleus.      

Using  the  measured  features,  objects  can  be  grouped  or  iden6fied  as  belonging  to  a  par6cular  class  or  recognized  as  having  a  given  phenotype.  

x1  and  x2  are  ‘features’,  or  measurements  made  on  the  objects.  Each  object  is  represented  by  a  symbol  in  ‘feature  space’.    

Carolina  Wählby,  [email protected]  

Example:  a  simple  adjustment  of  the  microscope  parameters  (focus)  made  cell  segmenta6on  much  easier  to  automate,  though  the  visual  difference  may  not  be  so  large  between  the  two  images.  

Lectures  on  digital  image  processing  and  analysis  

•  Extrac6ng  quan6ta6ve  informa6on  from  microscopy  data.  Cris  Luengo  and  Ida-­‐Maria  Sintorn,  CBA,  UU/SLU.    

•  Introduc6on  to  sonware  for  digital  image  processing  focused  on  quan6ta6ve  microscopy.  Carolina  Wählby  CBA    

•  Independent  work  on  projects  in  pairs  or  groups  of  3,  with  individually  scheduled  guidance/discussions  with  Carolina  Wählby,  CBA,  UU.    

•  Oral  presenta6on  and  discussion  of  projects.  

Carolina  Wählby,  [email protected]  

Page 5: QMic F1 20121009 ut - cb.uu.secarolina/QMicht2012/QMic_F1_20121009.pdf20121009 & 3 Temporal&proper6es&& • A&digital&image&is&characterized&by&&Is&itpossible&to&image&live&samples,&and&whatare&the&

2012-­‐10-­‐09  

5  

What  are  the  colors  of  the  cells?  

Why  don’t  we  just  look  at  the  images?  

Same  brightness…  

A   B  Which  nucleus  is  brighter?  

A   B  

Which  nucleus  is  larger?  

Same  size…  

Same  color…  

Why  quan6ta6ve  microscopy?  •  Reproducible  measurements:  by  sharing  our  ‘analysis  

pipeline’  anyone  can  reproduce  the  results  on  our  image  data.  •  Measurements  from  large  volumes  of  image  data  can  be  

automated,  leyng  us  focus  on  more  challenging  problems.  •  Measurements  from  larger  numbers  of  images  will  increase  

the  significance  of  our  results.  •  The  measurements  will  not  be  biased  by  the  observer.  

Carolina  Wählby,  [email protected]  

-­‐With  the  aid  of  informa=cs,  microscopy  is  in  the  midst  of  a  crucial  evolu=on  into  a  more  quan=ta=ve  and  powerful  technique.    

             Daniel  Evanko,  editor  of  Nature  Methods,  June  2012  

Introduc6on  of  course  par6cipants  

Carolina  Wählby,  [email protected]  

Johanna  Näslund  <[email protected]>  

Carolina  Wählby,  [email protected]  

Sultana  Jahan    <[email protected]>  

Carolina  Wählby,  [email protected]  

Diana  Telessemian  <[email protected]>  

Carolina  Wählby,  [email protected]  

Page 6: QMic F1 20121009 ut - cb.uu.secarolina/QMicht2012/QMic_F1_20121009.pdf20121009 & 3 Temporal&proper6es&& • A&digital&image&is&characterized&by&&Is&itpossible&to&image&live&samples,&and&whatare&the&

2012-­‐10-­‐09  

6  

Omer  Ishaq    <[email protected]>  

Carolina  Wählby,  [email protected]  

Arash  Sanamrad    <[email protected]>  

Carolina  Wählby,  [email protected]  

Kris6na  Lidayová    <[email protected]>  

Carolina  Wählby,  [email protected]  

Yinghua  Zha    <[email protected]>  

Carolina  Wählby,  [email protected]  

Amir  Motevakel    <[email protected]>  

Carolina  Wählby,  [email protected]  

Yu  Zhang    <[email protected]>  

Carolina  Wählby,  [email protected]  

Page 7: QMic F1 20121009 ut - cb.uu.secarolina/QMicht2012/QMic_F1_20121009.pdf20121009 & 3 Temporal&proper6es&& • A&digital&image&is&characterized&by&&Is&itpossible&to&image&live&samples,&and&whatare&the&

2012-­‐10-­‐09  

7  

Amol  Bhandage    <[email protected]>  

Carolina  Wählby,  [email protected]  

Haixia  Liu    <[email protected]>  

Carolina  Wählby,  [email protected]  

Benjamin  Holmgren  <[email protected]>  

Carolina  Wählby,  [email protected]  

Amanuel  Abraha    <[email protected]>  

Carolina  Wählby,  [email protected]  

Azadeh  Fakhrzadeh    <[email protected]>  

Carolina  Wählby,  [email protected]  

Ellinor  Spörndly-­‐Nees    <Ellinor.Sporndly-­‐[email protected]>  

Carolina  Wählby,  [email protected]  

Page 8: QMic F1 20121009 ut - cb.uu.secarolina/QMicht2012/QMic_F1_20121009.pdf20121009 & 3 Temporal&proper6es&& • A&digital&image&is&characterized&by&&Is&itpossible&to&image&live&samples,&and&whatare&the&

2012-­‐10-­‐09  

8  

Gustaf  Kylberg    <[email protected]>  

Carolina  Wählby,  [email protected]  

Beyna  Ryl    <[email protected]>  

Carolina  Wählby,  [email protected]  

Anna  Åsman    <[email protected]>  

Carolina  Wählby,  [email protected]  

Sanna  Hede    <[email protected]>  

Carolina  Wählby,  [email protected]  

Rui  Miao    <[email protected]>  

Carolina  Wählby,  [email protected]  

Hanneke  Marjolijn  Peele  <[email protected]>  

Carolina  Wählby,  [email protected]  

Page 9: QMic F1 20121009 ut - cb.uu.secarolina/QMicht2012/QMic_F1_20121009.pdf20121009 & 3 Temporal&proper6es&& • A&digital&image&is&characterized&by&&Is&itpossible&to&image&live&samples,&and&whatare&the&

2012-­‐10-­‐09  

9  

Carolina  Wählby,  [email protected]  

Andreas  Kårsnäs    <[email protected]>  

Carolina Wählby <[email protected]>

MSc  in  Molecular  Biotechnology  at  UU,  thesis  at  CMB,  KI:  Using  GFP  constructs  to  study  PML-­‐bodies  and  protein  targe6ng  

Cell  detec6on  and  quan6fica6on  of  ruffling  at  Rac1-­‐ac6va6on:  a  consultancy  study  sponsored  by  Amersham  Biosciences  for  the  first  version  of  INCell  Analyser.  

3h post plating 96h post plating Tacking  of  unstained  stem  cells  in  collabora6on  with  Chalmers  and  Department  of  Clinical  Neuroscience,  Göteborg  University  and  wavelet-­‐based  cell  classifica6on  for  Histogenics.   PostDoc,  Dept.  Gene6cs  and  Pathology,  UU.  Detec6on  and  classifica6on  

of  signals  from  different  variants  of  padlock  and  proximity  probes  in  cells,  6ssue  and  on  glass  surfaces.    

wt mutant

PhD  in  digital  image  analysis  at  the  Centre  for  Image  Analysis,  UU:  Collabora6on  with  CCK,  KI  on  quan6fica6on  of  cyclin  expression  in  normal  vs  cancer  6ssue.      

Associate  Professor  (docent)  in  Digital  Image  Processing  2009  

Principal  Inves6gator,  Imaging  PlaNorm,    Broad  Ins6tute  of  Harvard  and  MIT  (since  2009)  

Associate  Professor  (lektor)  in  Quan6ta6ve  Microscopy,  SciLifeLab  &  Centre  for  Image  Analysis,  Uppsala  University  

Projects  •  Op6onal  1-­‐5p;  the  number  of  points  will  be  decided  jointly  aner  

project  presenta6on.  •  Work  in  small  groups,  preferably  partners  with  different  

backgrounds,  with  the  aim  to  act  as  a  ‘teacher’  for  your  project  partner(s).  

•  Access  to  a  desired  imaging  system  will  be  organized  if  needed.  •  Get  started  discussing/working  on  projects  in  parallel  with  course  

work  –  projects  may  be  closely  related  to  your  current  research.  •  November  13-­‐December  10  Independent  work  on  projects,  with  

individually  scheduled  guidance/discussions  with  Carolina.    •  Tuesday  December  11    

10-­‐15  Oral  presenta6on  of  projects,  approximately  20  min/group,  and  handing  in  of  wriTen  reports  (template  will  be  provided).    

Carolina  Wählby,  [email protected]