Transcript Automation of Vessel Counting - University of California

Automation of Vessel Counting
Jessica DeQuach
Christman Lab
Boris Babenko
Belongie Lab
Myocardial Infarction and
Heart Failure
Heart Attack
Death of cells
Degradation of ECM
Wall thinning
Negative left ventricular remodeling
Heart Failure
Myocardial Infarction and
Heart Failure
RV
LV
6 weeks later….
Research With Injectable Biomaterials
• Want to develop an injectable biomaterial to
help regenerate/salvage damaged tissue
▫ Skeletal muscle
▫ Heart
• Injectable materials
▫ Minimally invasive (less surgery)
▫ Contours to shape of wound
Animal Surgeries
• Surgeries performed by Pam
or Rebecca
• Animals anesthetized
▫ First surgery: injury (infarct,
ischemia)
▫ Second surgery: injection, ~1
week later
• Animals kept in vivarium
until time point
http://kentsimmons.uwinnipeg.ca/16cm05/16labman05/lb8pg12.htm
Animal Studies (weeks to months)
• To get statistical significance, usually need n >= 4
per group
▫ Need to account for animal death, and variability
• Need to compare to a control group
• Note* need to include different experimental groups
on each day
Analysis
• Various methods to determine whether the
material is helping the damaged area
▫ Function
▫ Neovascularization
▫ Wall thickness
▫ Cellular infiltration
MRI
Neovascularization
• Formation of new blood vessels with red blood
perfusion
http://www.omegagenesis.com/html/Research.html
Animal Studies (weeks)
• Harvest tissue, freeze in OCT
• Section the sample to 10 um slices
Injection Region
• Identify the injection region
• Stain five evenly spaced slides through
the injection region
H&E
Vessel Stain
• Antibody stain for vessels
▫ Antibodies can identify
specific proteins
• Smooth muscle actin (red)
▫ Stains for smooth muscle
cells
• Isolectin (green)
▫ Stains for endothelial cells
http://www.siumed.edu/~dking2/crr/cvguide.htm
Animal Studies (weeks)
• Take images of vessel stain
• Need to quantify!
▫ The annotator should be trained to be
consistent
▫ Blinded so that no bias
▫ Only one person can quantify a set of data
Vessel Quantification
• Criteria:
▫ Within infarct or injection region
▫ Has a lumen
▫ Average diameter > 10 um
Vessel Quantification
Identify injection region
858_82_16
Vessel Quantification
Identify injection region
854_123_70
Vessel Quantification
Identify injection region
879_192_17
Vessel Quantification
Draw long and short axis
Vessel Quantification
Not all pictures will be that pretty
872_112_21
Vessel Quantification
• Vessel long and short axis are averaged
• Only count vessels > 10 um
• Hundreds of images, takes hours
• Only 1 person can quantify
Automation
• Automation of the process would be beneficial
• Need to determine “human error”
• This is where you come in!
• Quantify 150 images of skeletal muscle and heart.
Will need MATLAB, can do this from home
• Need data from several people
Matlab Program (designed by Boris et al)
Computer Vision (Belongie Lab)
• Combination of Artificial Intelligence, Machine
Learning, Image Processing, 3D Geometry (and
sometimes Cognitive Science)
• Examples in Industry:
▫ Face detection (i.e. most P&S cameras today)
▫ Google Goggles / TinEye
Computer Vision (Belongie Lab)
• Pedestrian Detection
• Reading text “in the wild”
Computer Vision and Bio
• Smart Vivarium (Piotr Dollar et al.)
• Coral Classification (Oscar Beijbom et al.)
Training
• We will first label 10 images together
• Then you will label 10 image on your own
▫ After someone double checks the answer you will
be okay to quantify on your own
Commitment
• Quantify 150 images of cardiac and skeletal
muscle, only in the Region of Interest (ROI)
• The Matlab program will automatically save
your data even if it crashes
• Can pace yourself, measure 10 or so a day
• Opportunity to participate in Bioengineering
Research, and can come to lab meetings, etc.
Contacts
• Jessica DeQuach
• [email protected]
• Boris Babenko
• [email protected]
• Anthony Monteforte
• [email protected]