image analysis using r - londonr · image processing libraries in cran biops image processing and...
TRANSCRIPT
Chris Campbell
LondonR - 13th July 2010
Image Analysis Using R
Steps to image analysis • Image capture
• Clean image/reduce noise
• Extract information
• Analyze information
Image Capture
Light
Photography
Light microscopy
Fluorescence microscopy
gels
cells
tissue samples
http:// ... western blot
http:// ... cells
Image Capture
Light
Photography
Light microscopy
Fluorescence microscopy
gels
cells
tissue samples
X-ray Radiography
Computed tomography (CT)
bones
tumours
http:// ... x-ray
http:// ... cat scan
Image Capture
Light
Photography
Light microscopy
Fluorescence microscopy
gels
cells
tissue samples
X-ray Radiography
Computed tomography
bones
tumours
Magnetism Magnetic resonance imaging (MRI) patients
http:// ... MRI
Image Capture
Light
Photography
Light microscopy
Fluorescence microscopy
gels
cells
tissue samples
X-ray Radiography
Computed tomography
bones
tumours
Magnetism Magnetic resonance imaging patients
Electrons Scanning electron microscopy
Transmission electron microscopy
insects
viruses
http:// ... SEM insect
http:// ... TEM virus
Image Capture
Light
Photography
Light microscopy
Fluorescence microscopy
gels
cells
tissue samples
X-ray Radiography
Computed tomography
bones
tumours
Magnetism Magnetic resonance imaging patients
Electrons Scanning electron microscopy
Transmission electron microscopy
insects
viruses
Positrons Positron emission tomography
(PET)
tumours
http:// ... positron emission tomography
Image Capture
Light
Photography
Light microscopy
Fluorescence microscopy
gels
cells
tissue samples
X-ray Radiography
Computed tomography
bones
tumours
Magnetism Magnetic resonance imaging patients
Electrons Scanning electron microscopy
Transmission electron microscopy
insects
viruses
Positrons Positron emission tomography
(PET)
tumours
Intermolecular
forces
Atomic force microscopy inorganic surfaces http://pico.iis.u-tokyo.ac.jp/media/16/20060621-QuenchedSi-AFM.jpg
Generally… • Use large numbers of images
• Use all images
• Use whole image, not crop
• Random selection not "typical region"
• i.e. avoid subjectivity
Image Processing Libraries in CRAN biOps Image processing and analysis
dcemri A Package for Medical Image Analysis
dpmixsim Dirichlet Process Mixture model simulation for clustering & image segmentation
edci Edge Detection and Clustering in Images
epsi Edge Preserving Smoothing for Images
FITSio FITS (Flexible Image Transport System) utilities
PET Simulation and Reconstruction of PET Images
R4dfp 4dfp MRI Image Read & Write Routines
rimage Image Processing Module for R
RImageJ R bindings for ImageJ
ripa R Image Processing & Analysis
tractor.base A package for reading, manipulating & visualising magnetic resonance images
adimpro Adaptive Smoothing of Digital Images
Libraries in CRAN biOps Image processing and analysis
dcemri A Package for Medical Image Analysis
dpmixsim Dirichlet Process Mixture model simulation for clustering & image segmentation
edci Edge Detection and Clustering in Images
epsi Edge Preserving Smoothing for Images
FITSio FITS (Flexible Image Transport System) utilities
PET Simulation and Reconstruction of PET Images
R4dfp 4dfp MRI Image Read & Write Routines
rimage Image Processing Module for R
RImageJ R bindings for ImageJ
ripa R Image Processing & Analysis
tractor.base A package for reading, manipulating & visualising magnetic resonance images
adimpro Adaptive Smoothing of Digital Images
• Open source
• Java
• Image analysis software http://rsbweb.nih.gov/ij/
package:RImageJ • Authors: Romain Francois & Philippe Grosjean
• Bindings between R and ImageJ
Subjectivity vs. Objectivity • Hypothesis: blue blobs are always larger than yellow blobs
Subjectivity • Hypothesis: blue blobs are always larger than yellow blobs
Manual
measurements
Subjectivity • Hypothesis: blue blobs are always larger than yellow blobs
It’s easy to accept
manual
measurements
when they make
sense, but it’s
tempting to
repeat them if
they seem wrong
Subjectivity • Hypothesis: blue blobs are always larger than yellow blobs
Subjective
observer accepts
expected
hypothesis
Objectivity • Hypothesis: blue blobs are always larger than yellow blobs
Automatically
threshold
Objectivity • Hypothesis: blue blobs are always larger than yellow blobs
Objective observer
automates analysis
and rejects
hypothesis
Automate Procedures • Identify objects without making subjective decisions
Run ImageJ from R • Open connection to
an image
• Use IJ$run() to
access macros
• Great potential for
automating image
processing from R
Run ImageJ from R • However, some key macros not yet implemented
(e.g. setAutoThreshold, imageCalculator)
package:rimage • Author: Nikon
• Reads jpegs into
RGB arrays
• Plot function defined
for objects of class
"imagematrix"
Analyze information • Plots and statistical summaries of particles from image
Single image
Multiple images
Conclusions • Images available?
• Ensure quality/validate method
• Choose useful measures
• Use analysis to make predictions
Acknowledgements • Mango Solutions www.mango-solutions.com
• L. R. Contreras-Rojas, R. H. Guy
http://www.bath.ac.uk/pharmacy/staff/rhg.html
• NAPOLEON http://www.ehu.es/napoleon/