4. Motivation Image segmentation Machine Learning methods Quantification
Motivation
Automated high resolution scanning microscopes digitize large sets of histological samples, and
access anatomical features of cells and tissues from the mm range down to a resolution of 230
nm.
Different Levels of resolution from a conventional tissue scaner
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 4/41
5. Motivation Image segmentation Machine Learning methods Quantification
Motivation
The high quality of the scans allows to collect quantitative morpho-topological features of cells
and tissue from different samples which can be coupled to functional information through
concomitant immunostaining or fluorescent protein.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 5/41
7. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Meaning
Image segmentation is:
The process of partitioning an image into multiple segments (e.g. in raster images, sets of
pixels).
The goal of the segmentation is to simplify and/or change the representation of an image
into something that is more meaningful and easier to analyze.
Sample of assited segmentation using minimal surfaces
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 7/41
8. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Results in medicine
Segmentation of Cerb2 in breast tissue..
Average continuity of the membrane stain, as a measure about the Immunostaining
expresion.
The geometric and spatial constraints of the antibody stain.
etc.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 8/41
9. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Results in medicine
Segmentation of Cerb2 in breast tissue..
Average continuity of the membrane stain, as a measure about the Immunostaining
expresion.
The geometric and spatial constraints of the antibody stain.
etc.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 8/41
10. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Results in medicine
Segmentation of Cerb2 in breast tissue..
Average continuity of the membrane stain, as a measure about the Immunostaining
expresion.
The geometric and spatial constraints of the antibody stain.
etc.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 8/41
11. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Results in medicine
Segmentation of Cerb2 in breast tissue..
Average continuity of the membrane stain, as a measure about the Immunostaining
expresion.
The geometric and spatial constraints of the antibody stain.
etc.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 8/41
12. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Results in biology
Segmentation of dendrites and somacell from SNpc of mice as a Parkinson’s disease model (PD).
Number of survival soma cells.
Size and morfological characteristics.
Size of dendrite projections.
etc.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 9/41
13. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Results in biology
Segmentation of dendrites and somacell from SNpc of mice as a Parkinson’s disease model (PD).
Number of survival soma cells.
Size and morfological characteristics.
Size of dendrite projections.
etc.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 9/41
14. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Results in biology
Segmentation of dendrites and somacell from SNpc of mice as a Parkinson’s disease model (PD).
Number of survival soma cells.
Size and morfological characteristics.
Size of dendrite projections.
etc.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 9/41
15. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Results in biology
Segmentation of dendrites and somacell from SNpc of mice as a Parkinson’s disease model (PD).
Number of survival soma cells.
Size and morfological characteristics.
Size of dendrite projections.
etc.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 9/41
16. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Results in biology
Segmentation of dendrites and somacell from SNpc of mice as a Parkinson’s disease model (PD).
Number of survival soma cells.
Size and morfological characteristics.
Size of dendrite projections.
etc.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 9/41
17. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Thresholding method
Thresholding:The simplest method of image segmentation is called the thresholding method.
This method is based on a clip-level (or a threshold value) to turn a gray-scale image into a
binary image. The key of this method is to select the threshold value (or values when
multiple-levels are selected).
Snow segmentation, by treshold over the pixel intensyties.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 10/41
18. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Histogram-based methods
Histogram-based methods: In this technique, a histogram is computed from all of the pixels in
the image, and the peaks and valleys in the histogram are used to locate the clusters in the
image.Color or intensity can be used as the measure.
Segmentation by histogram of colors.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 11/41
19. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Levelset methods
Levelset methods: The central idea behind such an approach is to evolve a curve towards the
lowest potential of a cost function, where its definition reflects the task to be addressed and
imposes certain smoothness constraints. Lagrangian techniques are based on parameterizing the
contour according to some sampling strategy and then evolve each element according to image
and internal terms.
Curve evolution of active contours and region growing method.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 12/41
20. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Methods
Compression-based methods
Edge detection
Split-and-merge methods
Partial differential equation-based methods
Graph partitioning methods
Watershed transformation
Model based segmentation
Multi-scale segmentation
Enough?
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 13/41
21. Motivation Image segmentation Machine Learning methods Quantification
Segmentation
Methods
The basis for robust and accurate quantification of structural and functional features is the
segmentation of regions of interest (ROIs) which define different elements within the scans.
Due to the diversity of possible targets, segmentation strategies need to be highly flexible in
order to define the ROIs for consecutive feature extraction.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 14/41
24. Motivation Image segmentation Machine Learning methods Quantification
Machine learning
Machine learning is a scientific discipline concerned with the design and development of
algorithms that allow computers to evolve behaviors based on empirical data, such as from
sensor data or databases.
The problem of learning can be viewed as a problem of estimating some unknown phenomenon
from the observed data.
In the literature, several learning algorithms have been propose,
artificial neural networks, supervised.
decision and regression trees,semisupervised.
connectionist networks, unsupervised.
probabilistic networks unsupervised.
and other statistical models,
fuzzy inference systems,
genetic algorithms,
genetic programming,
inductive logic programming,
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 17/41
25. Motivation Image segmentation Machine Learning methods Quantification
Machine learning
Preliminary Concepts
The sample generator generates the sample S, the supervisor or target operator establishes the
structure and relations that exist in the sample space Z, and the learning machine constructs an
approximation of the supervisor’s operator.
Supervised learning model,
h∗
: Unknown concept we want to approximate, hs : Approximation of the concept by imitation or
identification.
Unupervised learning model,
h∗
: Unknown concept we want to approximate, hs : Approximation of the concept by imitation or
identification.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 18/41
26. Motivation Image segmentation Machine Learning methods Quantification
Machine learning
Preliminary Concepts
Empirical Risk Minimization To select the best possible hyphotesis an indirect functional must
be minimized, given l : HxZ −→ +
0 , emp(h) = 1
n
n
i=1 l(h; zi )
Structural risk minimization seeks to prevent overfitting by incorporating a regularization
penalty into the optimization. The regularization penalty can be viewed as implementing a form
of Occam’s razor that prefers simpler functions over more complex ones.
Generalization Error Is defined as the absolute difference between the observed error rate
emp(hs ) and the expected error (hs ) of the hypothesis hS
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 19/41
27. Motivation Image segmentation Machine Learning methods Quantification
Machine learning
Preliminary Concepts
Empirical Risk Minimization To select the best possible hyphotesis an indirect functional must
be minimized, given l : HxZ −→ +
0 , emp(h) = 1
n
n
i=1 l(h; zi )
Structural risk minimization seeks to prevent overfitting by incorporating a regularization
penalty into the optimization. The regularization penalty can be viewed as implementing a form
of Occam’s razor that prefers simpler functions over more complex ones.
Generalization Error Is defined as the absolute difference between the observed error rate
emp(hs ) and the expected error (hs ) of the hypothesis hS
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 19/41
28. Motivation Image segmentation Machine Learning methods Quantification
Machine learning
Preliminary Concepts
Empirical Risk Minimization To select the best possible hyphotesis an indirect functional must
be minimized, given l : HxZ −→ +
0 , emp(h) = 1
n
n
i=1 l(h; zi )
Structural risk minimization seeks to prevent overfitting by incorporating a regularization
penalty into the optimization. The regularization penalty can be viewed as implementing a form
of Occam’s razor that prefers simpler functions over more complex ones.
Generalization Error Is defined as the absolute difference between the observed error rate
emp(hs ) and the expected error (hs ) of the hypothesis hS
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 19/41
29. Motivation Image segmentation Machine Learning methods Quantification
Machine learning
Preliminary Concepts
Empirical Risk Minimization To select the best possible hyphotesis an indirect functional must
be minimized, given l : HxZ −→ +
0 , emp(h) = 1
n
n
i=1 l(h; zi )
Structural risk minimization seeks to prevent overfitting by incorporating a regularization
penalty into the optimization. The regularization penalty can be viewed as implementing a form
of Occam’s razor that prefers simpler functions over more complex ones.
Generalization Error Is defined as the absolute difference between the observed error rate
emp(hs ) and the expected error (hs ) of the hypothesis hS
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 19/41
30. Motivation Image segmentation Machine Learning methods Quantification
Machine learning
Support Vector Machine
Support Vector Machine SVM
A support vector machine (SVM) is a concept in statistics and computer science for a set of
related supervised learning methods that analyze data and recognize patterns. The standard
SVM takes a set of input data and predicts, for each given input, which of two possible classes
comprises the input, making the SVM a non-probabilistic binary linear classifier. Given a set of
training examples, each marked as belonging to one of two categories.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 20/41
31. Motivation Image segmentation Machine Learning methods Quantification
Machine learning
Support Vector Machine
Support Vector Machine SVM
An SVM training algorithm builds a model that assigns new examples into one category or the
other. An SVM model is a representation of the examples as points in space, mapped so that
the examples of the separate categories are divided by a clear gap that is as wide as possible.
New examples are then mapped into that same space and predicted to belong to a category
based on which side of the gap they fall on.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 21/41
32. Motivation Image segmentation Machine Learning methods Quantification
Machine learning
Support Vector Machine
Not linear separable problems
The idea is to gain linearly separation by mapping the data to a higher dimensional space
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 22/41
33. Motivation Image segmentation Machine Learning methods Quantification
Machine learning
Support Vector Machine
Kernel trick
Where φ is a function that maps into another space:
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 23/41
34. Motivation Image segmentation Machine Learning methods Quantification
Machine learning
Support Vector Machine
Are explicitly based on a theoretical model of learning
Come with theoretical guarantees about their performance
Have a modular design that allows one to separately implement and design their
components
Are not affected by local minima
Do not suffer from the curse of dimensionality
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 24/41
36. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Quantification of soma cells in IHC images
This study explore the possible impact of targeting XBP-1, one of thetranscriptional factors
involved in the UPR, in the survival of SNpc under basal andpathological conditions
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 26/41
37. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Parkinson’s disease (PD) is the second most common neurodegenerative disease, affecting
at least 1% of the population over 55 years old.
The major clinical symptom of PD is impairment of motor control as a result from
extensive dopaminergic neuron death in the substantia nigra pars compacta (SNpc)
The mechanism involved in dopaminergic neuron loss in PD remains speculative.
Many different molecular mechanisms are proposed to explain the loss of dopaminergic
neurons in Parkinson Disease (PD), including oxidative stress and mitochondrial damage.
Increasing evidence from genetic and toxicological models of PD suggest a possible
involvement of endoplasmic reticulum stress (ER) and the unfolded protein response (UPR)
in disease process.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 27/41
38. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Parkinson’s disease (PD) is the second most common neurodegenerative disease, affecting
at least 1% of the population over 55 years old.
The major clinical symptom of PD is impairment of motor control as a result from
extensive dopaminergic neuron death in the substantia nigra pars compacta (SNpc)
The mechanism involved in dopaminergic neuron loss in PD remains speculative.
Many different molecular mechanisms are proposed to explain the loss of dopaminergic
neurons in Parkinson Disease (PD), including oxidative stress and mitochondrial damage.
Increasing evidence from genetic and toxicological models of PD suggest a possible
involvement of endoplasmic reticulum stress (ER) and the unfolded protein response (UPR)
in disease process.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 27/41
39. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Parkinson’s disease (PD) is the second most common neurodegenerative disease, affecting
at least 1% of the population over 55 years old.
The major clinical symptom of PD is impairment of motor control as a result from
extensive dopaminergic neuron death in the substantia nigra pars compacta (SNpc)
The mechanism involved in dopaminergic neuron loss in PD remains speculative.
Many different molecular mechanisms are proposed to explain the loss of dopaminergic
neurons in Parkinson Disease (PD), including oxidative stress and mitochondrial damage.
Increasing evidence from genetic and toxicological models of PD suggest a possible
involvement of endoplasmic reticulum stress (ER) and the unfolded protein response (UPR)
in disease process.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 27/41
40. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Parkinson’s disease (PD) is the second most common neurodegenerative disease, affecting
at least 1% of the population over 55 years old.
The major clinical symptom of PD is impairment of motor control as a result from
extensive dopaminergic neuron death in the substantia nigra pars compacta (SNpc)
The mechanism involved in dopaminergic neuron loss in PD remains speculative.
Many different molecular mechanisms are proposed to explain the loss of dopaminergic
neurons in Parkinson Disease (PD), including oxidative stress and mitochondrial damage.
Increasing evidence from genetic and toxicological models of PD suggest a possible
involvement of endoplasmic reticulum stress (ER) and the unfolded protein response (UPR)
in disease process.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 27/41
41. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Parkinson’s disease (PD) is the second most common neurodegenerative disease, affecting
at least 1% of the population over 55 years old.
The major clinical symptom of PD is impairment of motor control as a result from
extensive dopaminergic neuron death in the substantia nigra pars compacta (SNpc)
The mechanism involved in dopaminergic neuron loss in PD remains speculative.
Many different molecular mechanisms are proposed to explain the loss of dopaminergic
neurons in Parkinson Disease (PD), including oxidative stress and mitochondrial damage.
Increasing evidence from genetic and toxicological models of PD suggest a possible
involvement of endoplasmic reticulum stress (ER) and the unfolded protein response (UPR)
in disease process.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 27/41
42. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Parkinson’s disease (PD) is the second most common neurodegenerative disease, affecting
at least 1% of the population over 55 years old.
The major clinical symptom of PD is impairment of motor control as a result from
extensive dopaminergic neuron death in the substantia nigra pars compacta (SNpc)
The mechanism involved in dopaminergic neuron loss in PD remains speculative.
Many different molecular mechanisms are proposed to explain the loss of dopaminergic
neurons in Parkinson Disease (PD), including oxidative stress and mitochondrial damage.
Increasing evidence from genetic and toxicological models of PD suggest a possible
involvement of endoplasmic reticulum stress (ER) and the unfolded protein response (UPR)
in disease process.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 27/41
43. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Experimental Setup
One of the most frequently used pharmacological PD modelin rodents is the unilateral
injection of 6-hydroxydopamine (6-OHDA) in the striatum(Dauer et al, 2003).
This toxin acts specifically in dopaminergic neurons, inducing aretrograde damage, which
eventually results in cell dysfunction and death (Blum et al,2001).
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 28/41
44. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Experimental Setup
One of the most frequently used pharmacological PD modelin rodents is the unilateral
injection of 6-hydroxydopamine (6-OHDA) in the striatum(Dauer et al, 2003).
This toxin acts specifically in dopaminergic neurons, inducing aretrograde damage, which
eventually results in cell dysfunction and death (Blum et al,2001).
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 28/41
45. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Experimental Setup
One of the most frequently used pharmacological PD modelin rodents is the unilateral
injection of 6-hydroxydopamine (6-OHDA) in the striatum(Dauer et al, 2003).
This toxin acts specifically in dopaminergic neurons, inducing aretrograde damage, which
eventually results in cell dysfunction and death (Blum et al,2001).
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 28/41
46. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Objective
Automate the acounting of neurons in order to calculate the ratio of surviving neurons in
the ipsalateral side in the SNpc.
Compute the ratio of dendrite projection between the Ipsilateral and Contralateral sides of
the SNpc.
This work propose a ROI segmentation method by use of a supervised statistical learning
classifier, Support Vector Machine (SVM), under this approach ROI’s segmentation requires:
to find an optimal set of features to represent the images in a multiparametric space.
train an SVM model and perform a robust classification.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 29/41
47. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Objective
Automate the acounting of neurons in order to calculate the ratio of surviving neurons in
the ipsalateral side in the SNpc.
Compute the ratio of dendrite projection between the Ipsilateral and Contralateral sides of
the SNpc.
This work propose a ROI segmentation method by use of a supervised statistical learning
classifier, Support Vector Machine (SVM), under this approach ROI’s segmentation requires:
to find an optimal set of features to represent the images in a multiparametric space.
train an SVM model and perform a robust classification.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 29/41
48. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Objective
Automate the acounting of neurons in order to calculate the ratio of surviving neurons in
the ipsalateral side in the SNpc.
Compute the ratio of dendrite projection between the Ipsilateral and Contralateral sides of
the SNpc.
This work propose a ROI segmentation method by use of a supervised statistical learning
classifier, Support Vector Machine (SVM), under this approach ROI’s segmentation requires:
to find an optimal set of features to represent the images in a multiparametric space.
train an SVM model and perform a robust classification.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 29/41
49. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Objective
Automate the acounting of neurons in order to calculate the ratio of surviving neurons in
the ipsalateral side in the SNpc.
Compute the ratio of dendrite projection between the Ipsilateral and Contralateral sides of
the SNpc.
This work propose a ROI segmentation method by use of a supervised statistical learning
classifier, Support Vector Machine (SVM), under this approach ROI’s segmentation requires:
to find an optimal set of features to represent the images in a multiparametric space.
train an SVM model and perform a robust classification.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 29/41
50. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Color space
The images were transform from de RGB color space to HSV table.
Hue channel : encode the color
0 = RGB(1, 0, 0)
60 = RGB(1, 1, 0)
120 = RGB(0, 1, 0)
180 = RGB(0, 1, 1)
240 = RGB(0, 0, 1)
300 = RGB(1, 0, 1)
360 = 0
Saturation: Encode the pureness intensity Distance to the black and white axis
Value: Encode the brightness
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 30/41
51. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Color space
The images were transform from de RGB color space to HSV table.
Hue channel : encode the color
0 = RGB(1, 0, 0)
60 = RGB(1, 1, 0)
120 = RGB(0, 1, 0)
180 = RGB(0, 1, 1)
240 = RGB(0, 0, 1)
300 = RGB(1, 0, 1)
360 = 0
Saturation: Encode the pureness intensity Distance to the black and white axis
Value: Encode the brightness
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 30/41
52. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Color space
The images were transform from de RGB color space to HSV table.
Hue channel : encode the color
0 = RGB(1, 0, 0)
60 = RGB(1, 1, 0)
120 = RGB(0, 1, 0)
180 = RGB(0, 1, 1)
240 = RGB(0, 0, 1)
300 = RGB(1, 0, 1)
360 = 0
Saturation: Encode the pureness intensity Distance to the black and white axis
Value: Encode the brightness
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 30/41
53. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Color space
The images were transform from de RGB color space to HSV table.
Hue channel : encode the color
0 = RGB(1, 0, 0)
60 = RGB(1, 1, 0)
120 = RGB(0, 1, 0)
180 = RGB(0, 1, 1)
240 = RGB(0, 0, 1)
300 = RGB(1, 0, 1)
360 = 0
Saturation: Encode the pureness intensity Distance to the black and white axis
Value: Encode the brightness
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 30/41
54. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Color space
The images were transform from de RGB color space to HSV table.
Hue channel : encode the color
0 = RGB(1, 0, 0)
60 = RGB(1, 1, 0)
120 = RGB(0, 1, 0)
180 = RGB(0, 1, 1)
240 = RGB(0, 0, 1)
300 = RGB(1, 0, 1)
360 = 0
Saturation: Encode the pureness intensity Distance to the black and white axis
Value: Encode the brightness
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 30/41
55. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Feature space
Increase the problem dimensionality by convolution with Gabor Wavelets.
Provide a projection basis comparable in some cases to the projection basis obtained by
PCA and solving the eigen values problem
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 31/41
56. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Setup SVM
SVM Engine: Libsvm
There is no gold standard for the neuron morphology
Exist a ground truth for the quantification
Gold standard arbitrarily defined by masks drew manually to achieve the best segmentation
possible.
Training based on pixels and his projected properties.
92% average accuracy in training.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 32/41
57. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 33/41
58. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 34/41
59. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Quantification of soma cells in IHC images
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 35/41
60. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Pixels Classification of IHC-stained Images using SVM
Pixels Classification of IHC-stained Images using SVM
This study is focused on digital image processing of human breast tissues, which present an
invasive ductal carcinoma and have been treated using an specific IHC technique that allows to
detect the overexpression of HER2 protein (c-erbB-2 oncoprotein). This detection is very
important for prognosis of breast cancer and for the patient treatment as well. A high level of
HER2 overexpression implies a poor prognosis and the development of cancer metastasis.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 36/41
61. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Pixels Classification of IHC-stained Images using SVM
The general objective of this study is to provide support to pathologists and contribute to the
digital IHC image analysis, using Support Vector Machines (SVMs) for pixel image classification.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 37/41
62. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Pixels Classification of IHC-stained Images using SVM
Feature Extraction.
in this study 10 features were extracted in order to provide useful information for the
classification: mean, standard deviation, entropy, dynamic range, Sobel gradient magnitude, Y
(from CMYK) channel pixel, Jensen-Shannon divergence (magnitude and orientation),
vesselness, and Gabor wavelet.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 38/41
63. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Pixels Classification of IHC-stained Images using SVM
Jensen Shannon Divergence Texture analisys maps regions where coherent patterns of
intensities are identified. Patterns results from physical properties such as roughness, oriented
strands or reflectance differences such as the color on a surface.
Jensen Shannon Divergence This feature is based on some filters that are used for the
enhancement of vessels structures –ducts or a tubes that contains or conveys a body fluid– in
order to grade the stenoses for the diagnosis of the severity of vascular disease.
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 39/41
64. Motivation Image segmentation Machine Learning methods Quantification
Quantification
Pixels Classification of IHC-stained Images using SVM
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 40/41
65. Motivation Image segmentation Machine Learning methods Quantification
Questions ?
Rodrigo Rojas Moraleda
rodrigo.rojas@postgrado.usm.cl
Rodrigo Rojas Moraleda — Segmentation and Quantification of IHC-stained Tissue Images by Machine Learning Methods 41/41