Transcript Undecided still

Problems in Biological Imaging:
Opportunities for Signal Processing
Jelena Kovačević
bimagicLab
Center for Bioimage Informatics
Department of Biomedical Engineering
Department of Electrical and Computer Engineering
Carnegie Mellon University
Cast of Characters
The Roadmap
Issues
Revolution in biology
Tools
Framework
What can we do?
Tasks
Revolution in Biology

Focus in biology



Vertical to horizontal approach
“Omics”: genomics, proteomics, …
Fluorescence microscopy




Hugely successful
Allows for live-cell imaging
Fluorescent markers, starting with GFP
Allows for collection of high-dimensional data sets




2D images and 3D volumes
At multiple time instants
Multiple channels
Analysis and interpretation

Cumbersome, nonreproducible, error prone
Goal
PSF h

Imaging in systems biology

Use informatics to


Leads to


acquire, store, manipulate
and share large
bioimaging databases
automated, efficient and
robust processing
Need

Host of sophisticated tools
from many areas
A/D
Restoration
Denoising +
Deconvolution
Denoising
Deconvolution
Registration
Mosaicing
Segmentation
Tracking
Analysis
Modeling
The Roadmap
Issues
Revolution in biology
Noise levels and types
Lack of ground truth
Large deviations
Low definition and contrast
Wide range of time-frequency features
Noise Levels and Types

Shift towards noninvasive





Data collected farther from the source
Signals typically corrupted by
high levels of noise
Weak biosignals
Standard SP techniques not used
but even those will not work well
with such signals
Types of noise


Electrical, neuronal, …
Modeling of noise a problem
Lack of Ground Truth

Shift towards noninvasive

No access to ground truth
Large Deviations

Humans and/or animals as ``customers'‘

Wide range of states considered ``normal'‘

Looking for is a range rather than a single state

Large deviations from the range of normal states may
characterize what we are looking for
normal
delayed
abnormal
Low Definition and Contrast

Images typically have low contrast
and are poorly defined

Lack of consistent edges
Wide Range of Time- and FrequencyLocalized Features

Bioimages

Global behaviors together with spikes and transients

Puts time-frequency tools to the test

“Speckled” nature---stochastic representation
The Roadmap
Issues
Revolution in biology
Framework
Continuous-domain image processing
From continuous to discrete domain
Discrete-domain image processing
Continuous-Domain Image Processing
PSF h

Specimen (object) vs
image of it (projection)
A/D
Restoration
Denoising +
Deconvolution
Denoising
Deconvolution
Registration
Mosaicing

LSI systems


Impulse response of the
microscope: PSF
Fourier view

FT or FS
Segmentation
Tracking
Analysis
Modeling
From Continuous to Discrete
PSF h

Resolution in microscopy
A/D
Restoration
Denoising +
Deconvolution

Filtering before sampling

Sources of uncertainty
Denoising
Deconvolution
Registration
Mosaicing
Segmentation
Tracking
Analysis
Modeling
Discrete-Domain Image Processing
PSF h

LSI system, digital
filtering

Consider the signal as



Infinite signal with finite
number of nonzero
coefficients
Finite signal
Fourier view

DTFT

DFT
A/D
Restoration
Denoising +
Deconvolution
Denoising
Deconvolution
Registration
Mosaicing
Segmentation
Tracking
Analysis
Modeling
The Roadmap
Issues
Revolution in biology
Tools
Framework
Signal and image representations
Fourier analysis
Gabor analysis
Multiresolution analysis
Data-driven representation and analysis
Signal Representations
ER
FT
WP
WT
Dirac
STFT
basis
f

“Holy Grail” of signal
analysis/processing

Actin

t
Understand the “blob”-like
structure of the energy
distribution in the timefrequency space
Design a representation
reflecting that
Data Driven Representation & Analysis

Use representations based on training data and
automated learning approaches

Wavelet packets

PCA & variations

ICA

…
Estimation Framework

Random variations introduced by system noise,
artifacts, uncertainty originating from the biological
phenomena lead to statistical methods

Seek the solution optimal in some probabilistic
sense

Optimality criterion

MSE, often depends on unknown parameters

Bayesian framework, MAP estimators
The Roadmap
Issues
Revolution in biology
Tools
Framework
Tasks
Acquisition
Deblurring, denoising, restoration
Registration and mosaicing
Segmentation, tracing and tracking
Classification and clustering
Modeling
Acquisition


Issues in acquisition of
fluorescence microscope images
Increase resolution




Acquire for longer periods




Total data acquisition is reduced, speeding up image acquisition
Allows a higher frame rate (increased temporal resolution)
Allows us to spend more time acquiring the regions of interest (which gives increased spatial
resolution)
Acquisition process damages both the signal (photobleaching) and the cell (phototoxicity)
Efficient acquisition reduces the total amount of data acquired, thus reducing damage to the cell
This allows us to observe cellular processes for longer periods
Intelligent acquisition


Acquire only where and when needed  adaptivity
Model driven (microscope model & data model)
Model-Driven Acquisition

Acquisition

Grid acquisition

MR adaptive acquisition

Markov Random Fields

Example-based enhancement
Efficient Acquisition
Reconstruction
Reconstruction

Simple interpolation methods

Wavelet reconstruction

Model-based reconstruction
Knowledge Extraction
Modeling

MR Acquisition
[Merryman & Kovačević, 2005]

Problem



Measure of success



Why acquire in areas of
low fluorescence?
Acquire only when and
where needed
Accuracy
Problem dependent
Here:
Strive to maintain the
achieved classification
accuracy
Approach

Mimic “Battleship”
Compression Ratio
Efficient Acquisition and Learning of
Fluorescence Microscope Data Models
2.
Intelligent Acquisition
No
1.
Model Building
Model
satisfactory?
Yes
Model
Develop a mathematical framework and algorithms
to build accurate models of fluorescence microscope data sets
as well as design intelligent acquisition systems based on those models
1. Use all the input from the
microscope to model the
data set
2. Choose acquisition
regions that allow us to
construct an accurate model
in the shortest amount of time
Efficient Acquisition and Learning of
Fluorescence Microscope Data Models
[Jackson, Murphy & Kovačević, 2007]


Predict the distribution of fluorescence in the subsequent
frame and acquire accordingly

Predict likelihood of object moving to any given position

Acquire those positions with the highest likelihood

Too small an acquisition region may not find the object

Too large an acquisition region is inefficient
Motion models

Three motion models commonly observed in practice

Random walk

Constant velocity

Constant acceleration
Efficient Acquisition and Learning of
Fluorescence Microscope Data Models

Learning the motion model

Prediction: Based on current beliefs about motion model,
find likelihood of each object appearing at any given pixel
in the subsequent frame

Acquisition: Acquire the pixels that have the highest
overall likelihood of containing an object

Observation: Observe the actual location of each object,
if found

Update: Use this information to update our beliefs about
the motion models for each object
Efficient Acquisition and Learning of
Fluorescence Microscope Data Models

Known motion model

Single object, random walk of known variance

Probability distribution of it appearing in any given location
in the subsequent frame

Acquisition regions capture the locations where the object
is expected with the highest probabilities
Efficient Acquisition and Learning of
Fluorescence Microscope Data Models

Known motion model

If the object is detected, repeat, centering the new
acquisition region at the object’s most recent location

If the object is not detected, estimate where it is

Probability distribution given that the object was not in the
acquisition region
Efficient Acquisition and Learning of
Fluorescence Microscope Data Models

Known motion model

Predict this object’s location in the next frame

Probability distribution

1D case: choose two disconnected acquisition regions

2D case: choose to acquire between the two black circles
Deblurring, Denoising & Restoration


Microscope images contain artifacts

Blurring caused by a PSF

Noise from the electronics of digitization
Deblurring/deconvolution

Widefield microscopy

Effect of depth

Denoising

Deconvolution + Denoising = Restoration
Registration & Mosaicing

Registration


Find spatial relationship and alignment between images
Mosaicing

Used when fine resolution is needed within a global view

Stitching together pieces of an image

Usually requires registration, given overlapping pieces
Segmentation, Tracing & Tracking



Segmentation

Methods used: thresholding and watershed

Edge-based, region-based, combination

Active contours
Tracing

Mostly tracing of axons

Typical, path following approaches

Fail in the presence of noise
Tracking

Molecular dynamics and cell migration

Tracking of objects over time
Segmentation


Separate objects of interest
from each other and the
background

Fundamental step in
microscopy

Hand segmentation

Not reproducible

Not tight

Piecewise linear

Cannot compute statistics

Time-consuming
Current standard

Watershed segmentation
Active Contour Segmentation


Active contour algorithms

Contour comparable to an elastic string

Moved under external and internal forces

External: derived from the image (edges)

Internal: geometric properties of the contour (curvature)
Level Set method: A way to track the contour as it evolves



Positive inside the contour
(mountain)
Negative outside the contour
(valley)
Zero on the contour,
C embedded at its zero (sea) level
Fc < 0
<0
>0
=0
Fc > 0
n
STACS

Combines energy minimization approach with statistical modeling


Model matching
 Pixels inside and outside the contour follow different statistical
models
Modified STACs for fluorescence microscopy images
 No edge information


No obvious shape information
Segmentation driven by statistics of the image and contour
smoothness

MSTACS: Our level-set evolution equation

Topology needs to be preserved  TPSTACS
TPSTACS: Results
[Coulot, Kirschner, Chebira, Moura, Kovačević, Osuna & Murphy, 2006]

Successful

Problem
Hand-segmented


Solution

TPTACS
Extremely slow
MRSTACS
MRSTACS


Decompose image
to L levels
Smoothing renders cell
easier to discern

Detect cells using
morphological operations

Get coarse version of
contour (TPSTACS)

Refine contour iteratively
 faster
segmentation

Coarse result < 3 sec

Fine result < 30 min
h
↓2
g
↓2
h
↓2
↓2
g
↓2
horizontal
2D Filter bank
Level 1 decomposition
h
g
↓2
vertical
37
A Critical Review of Active Contours

Flexible

Can be tuned to be accurate

Adapt to topological changes in the image

But…






Tuning of parameters is involved
Updating the level set function – inefficient
What is the ‘contour’ in a digital image?
Discrete topological rules – external constraints can cause
abruptness
Multiresolution – how do we reconstruct the level set function?
 New math needed
Active Mask Framework: No Contours

Fluorescence microscope images speckled in nature
Estimate densities of bright pixels in local neighborhood
at different scales


Recast computation of force as a transformation


20
20
40
40
60
No need for the time consuming extension function
60
80
80
100
100
120
120
For image f, transform T is
140
160
140
160
20
40
60
80
100
120
140
160
180
200
20
40

60
80
100
120
140
160
180
200
A slight blur
Original Image
20
20
40
40
60
60
Windowing function  and scale factor a
80

100
120

140
80
Different conditions (cell lines, resolution, etc.)  Different  and a
100
120
TPSTACS: Rectangular , a = 1 and suitable operands
140
160
160
20
40
60
80
100
120
140
160
180
200
Enough to discern the cell
boundary
20
40
60
80
100
120
140
160
180
200
Too much blur – Edges
rounded
Active Masks: Results
HeLa cells – Total protein image

HeLa cells – Membrane protein image
Success

Initialization: Level set function is identically zero

Iterations: 3

Time taken: 6.5 sec per iteration
Active Masks

Pros






Cons


Framework suited to digital images
Can be made specific with the choice of suitable forces,
windows and scale factors
Performance not critically dependent on initialization
Easy and fast to compute
Translation, dilation and rotation invariance can be
preserved
Topology preservations hard
 Multiple active mask framework
Multiple Active Masks

Initialization

Random initialization with M»M0 masks,
where M0 = expected number of objects in the image

Evolution: driven by distributor functions

Can incorporate multiresolution/multiscale

Convergence

Experimentally

Working on a proof
Results of STACS on Different Modalities
Yeast DIC
Cardiac MRI: Endocardium and epicardium
Brain fMRI
Axial
Coronal
Saggital

True Positive

False Positive

False Negative
Classification Problems in Bioimaging

Determination of
protein subcellular location patterns
[Chebira, Barbotin, Jackson, Merryman, Srinivasa, Murphy & Kovačević, 2007]

Detection of developmental stages in
Drosophila embryos
[Kellogg, Chebira, Goyal, Cuadra, Zappe, Minden & Kovačević, 2007]

Classification of histological stem-cell teratomas
[Ozolek, Castro, Jenkinson, Chebira,, Kovačević, Navara, Sukhwani,
Orwig, Ben-Yehudah & Schatten, 2007]

Fingerprint recognition
[Hennings, Thornton, Kovačević & Kumar , 2005]
[Chebira, Coelho, Sandryhalia, Lin, Jenkinson, MacSleyne, Hoffman, Cuadra,
Jackson, Püschel & Kovačević , 2007]
Develop an automated system capable of
fast, robust and accurate classification
Multiresolution Classification
shorthand System
Generic Classification
MR

Classification
C
W
Weighting
Algorithm
Hypothesis: Better classification accuracy obtained if we use the spacefrequency information lying in the MR subspaces


Feature
FE
MR Extraction
Compute features in the MR-decomposed subspaces (subbands) instead
Would like to use wavelet packets

Do not have an obvious cost measure

Do it implicitly instead
MR Block
FE
MR
C
W

Grow full tree to L levels

Use all nodes

MR Bases


DWT

DFT

DCT

…
MR Frames

SWT

DT-CWT

DD-DWT

Our design: LTFT
Lapped Tight Frame Transforms

Build MR transforms for these problems

Not many nonredundant ones exist

Seed them from higher-dimensional bases
Feature Extraction and Classifier


Feature Extraction
MR

New Haralick texture features (T3, 26 features)

Morphological features (M, 16 features)

Zernike features (Z, 49 features)
Classifier

Neural networks

No hidden layers
MR
FE
C
W
FE
C
W
Weighting Procedure


Local decisions
MR

Decision vectors for each subband
of each training image containing C numbers

Goal: combine local decisions into a global one
Algorithms

Open form (iterative)

Closed form (analytical)


Per data set

Per class
Pruning criteria
FE
C
W
Determination of PSL Patterns:
Results

MR significantly
outperforms
NMR

MRF outperform
MRB

Per-Dataset CF
slightly
outperforms OF

Trend is flat
→
T3 set enough
Why Do MR Frames Work?

Looking into classes of signals where bases/frame
perform better

Simple example

Real plane

Two classes

Decision rule

Union of nonoverlaping parallelograms, bases,
otherwise, frames
Conclusions and Opportunities
Issues
Revolution in biology
Tools
Framework
What can we do?
Tasks
Conclusions & Opportunities
The “dream”:
automated, efficient and
reliable processing as well
as knowledge extraction
from large bioimage databases

Dig in!

Gaps to fill

Need tools adapted to
specific bioimaging
applications

Need to adapt state-of-theart techniques and/or
come up with new ones for
bioimaging tasks