Automated segmentation of blood cells in Giemsa stained...

19
© FIMM - Institiute for Molecular Medicine Finland www.fimm.fi dd.mm.yyyy © FIMM - Institiute for Molecular Medicine Finland www.fimm.fi 8 June 2012 Automated segmentation of blood cells in Giemsa stained digitized thin blood films Margarita Walliander Institute for Molecular Medicine Finland University of Helsinki

Transcript of Automated segmentation of blood cells in Giemsa stained...

Page 1: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

© FIMM - Institiute for Molecular Medicine Finland www.fimm.fi

dd.mm.yyyy

© FIMM - Institiute for Molecular Medicine Finland www.fimm.fi

8 June 2012

Automated segmentation of blood cells in Giemsa stained digitized thin blood films Margarita Walliander

Institute for Molecular Medicine Finland

University of Helsinki

Page 2: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi Figure by http://www.funsci.com 3

What is a Giemsa stained thin blood film?

8 June 2012

www.fimm.fi 4

What is a Giemsa stained thin blood film?

8 June 2012

Page 3: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 5

What is a Giemsa stained thin blood film?

8 June 2012

Metafer, MetaSystems, Altlussheim and Axio Imager Z2, Carl Zeiss MicroImaging, using a 63X objective (NA 1.4 oil immersion, pixel size 0.10µm)

fimm.webmicroscope.net/Research/Momic/tp2012 *

* Linder et al. Web-based virtual microscopy for parasitology: a novel tool for education and quality assurance. PLoS Negl Trop Dis. 2008;2(10)

www.fimm.fi 6

What is a Giemsa stained thin blood film?

8 June 2012 fimm.webmicroscope.net/Research/Momic/tp2012

Page 4: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 7

Why do we need a thin blood film? •  Analysis of erythrocytes and leucocytes in thin blood films is an

important task for accurate assessment of infectious disease diagnostics.

• Manual counting of cells is currently the only method available for quantifying parasitaemia in infected blood.

• Manual cell counting is time consuming and subject to variability*.

• We propose an automatized method for separating and counting cells.

8 June 2012

* O'Meara et al: Reader technique as a source of variability in determining malaria parasite density by microscopy. Malaria Journal 2006, 5:118

www.fimm.fi

Materials

•  500 fields of view from each blood film sample were captured.

•  6 of the samples were infected with Plasmodium falciparum,

•  4 were non-infected controls.

8

8 June 2012

Page 5: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi

Methods

•  Adaptive histogram thresholds

•  Background/foreground separation

•  Recognition of different objects that compose the foreground

•  Cell counting

8 June 2012

Original image

Foreground Background

Round Cells

Found Cells

Approx Cells

Total RBC

WBC

www.fimm.fi 10

Original image

8 June 2012

Page 6: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 11

Green channel

8 June 2012

www.fimm.fi 12

Adaptive histogram thresholds

The color distribution of a thin blood film presents a bimodal shape.

B is the threshold that separates background from foreground.

8 June 2012

Page 7: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi

Separation of background and foreground

8 June 2012

www.fimm.fi 14

Separation of background and foreground original

8 June 2012

Page 8: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 15

Round objects selection

8 June 2012 Roundness = 0.75

www.fimm.fi

RoundCells RoundCells = Objects with similar area (mean +/- σ)

8 June 2012

Page 9: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi

RoundCells

8 June 2012 AvgRBC= average ( RoundCells )

106 cells

www.fimm.fi 18

Detection of heavily stained objects

H is the threshold that will define all the heavily stained objects included in the foreground

• WBCs

•  parasites

•  platelets

•  debris

8 June 2012

Page 10: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 19

Detection of heavily stained objects

8 June 2012

www.fimm.fi 20

Heavily stained objects > AvgRBC

8 June 2012

Page 11: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 21

Heavily stained objects > AvgRBC

8 June 2012

WBCs

www.fimm.fi 22

Foreground - WBC - RoundCells

8 June 2012

Page 12: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 23

Hough transform

8 June 2012

www.fimm.fi 24

Hough transform

8 June 2012

Page 13: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 25

Hough transform

8 June 2012

www.fimm.fi 26

Detection of centers

8 June 2012

Page 14: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 27

Dilation of detected centers

8 June 2012

www.fimm.fi 28

Foundcells

8 June 2012

44 cells

AvgRBC2 = average ( FoundCells )

Page 15: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 29

Foreground-WBC-RoundCells-FoundCells

8 June 2012

www.fimm.fi 30

# ApproxCells

Area ( Background - RoundCells – WBCs – FoundCells)

Area (AvgRBC2)

8 June 2012

Page 16: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 31

# ApproxCells

Area ( Background - RoundCells – WBCs – FoundCells)

Area (AvgRBC2)

8 June 2012

79 cells

www.fimm.fi 32

Total cell counting

8 June 2012

# RoundCells

# FoundCells

# ApproxCells Σ"Total amount of

Red Blood Cells =

Page 17: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi

Example Image 1500x1500 pixels ~ 1.9 fields of view

Cells Counted 106

Cells Found 44

Cells Approximated 79

Total 229 RBC s

2 WBCs

www.fimm.fi 34

Validation results

8 June 2012

Red blood cells White blood cells

Sample Annotated Automated Error % Annotated Automated Error % I1 3145 3160 0,47 20 20 0,00 I2 4048 4058 0,24 34 34 0,00 I3 2796 2782 0,50 22 22 0,00 I4 2972 2958 0,47 30 32 6,66 I5 3047 3042 0,16 77 75 2,59 I6 3513 3514 0,02 71 72 1,38 C1 3396 3389 0,20 75 75 0,00 C2 3491 3482 0,25 55 56 1,81 C3 3093 3087 0,19 42 42 0,00 C4 3197 3206 0,28 49 50 2,04

TOTAL 32698 32678 0,06 476 477 0,21 % 100 99,9388 100 99,79

30 fields of view per sample

Page 18: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 35

Test results

8 June 2012

Red blood cells

sample! #RoundCells! #FoundCells! #ApproxCells! #TotalRBC!

I1! 17935! 40512! 886! 59333!I2! 16458! 52600! 1178! 70236!I3! 23068! 20475! 430! 43973!I4! 14237! 32438! 305! 46980!I5! 13670! 31918! 502! 46090!I6! 16656! 41098! 275! 58029!C1! 18379! 38993! 388! 57760!C2! 23704! 29669! 272! 53645!C3! 22290! 22669! 503! 45462!C4! 18108! 32674! 823! 51605!

TOTAL! 184505! 343046! 5562! 533113!

500 fields of view per sample

www.fimm.fi

Conclusions

•  An unsupervised tool for separating red blood cells and white blood cells in Giemsa stained thin blood films.

•  An automated red blood cell and white blood cell counting tool.

•  The segmentation of blood cells in thin blood films can be used as a pre-processing step to specify the regions of interest for a secondary algorithm.

36

8 June 2012

Page 19: Automated segmentation of blood cells in Giemsa stained ...mitel.dimi.uniud.it/tp2012/presentations/B3-Walliander.pdf · Automated segmentation of blood cells in Giemsa stained digitized

www.fimm.fi 37

Thanks!

8 June 2012