This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the National Strategic Reference Framework (NSRF 2007-
2013)
Region segmentation and classification in Region segmentation and classification in digital images of liver tissuedigital images of liver tissue
Markos G. Tsipouras, Zoi Tsianou, Nikolaos Giannakeas,
Alexandros T. Tzallas,
Pinelopi Manousou, Epameinondas V. Tsianos,
This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the
National Strategic Reference Framework (NSRF 2007-2013)
• Fibrosis assessment in liver biopsies is currently based on semiquantitative staging scores i.e. Ishak.
• Collagen proportional area (CPA) assessment through digital image analysis have proven to be more accurate than semiquantitative scores since it can provide quantification of collagen.
• However, there are no methods presented in the literature for automated digital image analysis for CPA assessment.
IntroductionIntroduction
This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the
National Strategic Reference Framework (NSRF 2007-2013)
• Development of an automated robust CPA assessment methodology through digital image analysis
• Three stage methodologya. Tissue/background separation (image segmentation)
b. Tissue characterization (region classification)• tissue• muscle tissue, blood clots, structural collagen, stain,
artifacts, fat
c. CPA assessment
AimAim
This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the
National Strategic Reference Framework (NSRF 2007-2013)
• 94 liver biopsies obtained from different patients • picroSirious red stained • photographed with a digital camera • the images included several tissue and non-tissue
areas, annotated by an expert pathologist
DatasetDataset
This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the
National Strategic Reference Framework (NSRF 2007-2013)
This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the
National Strategic Reference Framework (NSRF 2007-2013)
• Tissue detection:a. 3x3 pixels window
b. average value for RGB
c. clustering (K-means algorithm)
• Results to image regions
Methodology Methodology [1/2][1/2]
This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the
National Strategic Reference Framework (NSRF 2007-2013)
This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the
National Strategic Reference Framework (NSRF 2007-2013)
This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the
National Strategic Reference Framework (NSRF 2007-2013)
• Regions Classification:a. Calculation of several characteristics for each region
(based on pixel values/color and region shape) Region Shape: {Area, Eccentricity, Diameter, Euler Number,
Extent, Major & Minor Axis Length, Perimeter, Solidity}Pixel Color: {Mean/min/max Intensity for R/G/B channels}
b. Region classification into several classes(decision tree algorithm)
Classes: {Tissue, Muscle Tissue, Blood Clot, Structural Collagen, Dye, Artifact}
Methodology Methodology [2/2][2/2]
This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the
National Strategic Reference Framework (NSRF 2007-2013)
Structural Collagen
Stain
Blood Clot
Artifact
This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the
National Strategic Reference Framework (NSRF 2007-2013)
ResultsResults
Tissue Muscle Blood Str. Col. Stain Artifact Total %
Tissue 403 68 25 4 2 0 502 80.3
Muscle 1 42 0 0 0 0 43 97.7
Blood 0 0 24 0 0 0 24 100
Str. Col 1 1 0 35 0 0 37 94.6
Stain 0 0 0 0 3 0 3 100
Artifact 0 0 0 0 0 7 7 100
Total Classification Accuracy: 83.5%
This research project has been co-financed by the European Union (European Regional Development Fund- ERDF) and Greek national funds through the Operational Program “THESSALY- MAINLAND GREECE AND EPIRUS-2007-2013” of the
National Strategic Reference Framework (NSRF 2007-2013)
• Although CPA assessment through DIA has been introduced for more than two decades, and proven to be superior to traditional semiquantitative scores, it has not yet reached the everyday clinical practice.
• Development of simple to use and robust methodologies for all stages of image analysis can lead to wider spread of CPA assessment through DIA.
ConclusionsConclusions