Automated Slide Analysis Uli Klingbeil MetaSystems

advertisement
Automated Slide Analysis
Uli Klingbeil
MetaSystems
Applications of Automatic Scanning
Platform Metafer
• MetaCyte: automatic interphase FISH scoring
morphometric/intensity measurements
ploidy analysis
• RCDetect: rare cell detection (isolated tumor cells in
bone marrow and peripheral blood, fetal cells)
• MSearch: automatic metaphase finding in brightfield
and fluorescence
• Comet scan: automated comet assay analysis
• Micronuclei: toxicology
Manual interphase FISH scoring
Limitations:
• tedious and error-prone task
• result depends on skills and experience of operator
• lack of documentation
Desirable:
• quantitative analysis
• analysis of more cells for better sensitivity
• documentation of each cell for review and verification
Metafer - concept
CCD Camera
Metafer
Motorized
stage (x,y)
Motorized microscope (z)
real-time image
analysis and
microscope control
PC Win 9x
NT / 2000 / XP
Metafer Applications
MetaCyte - interphase FISH scoring
Step 1
set up slide, classifier and seach window
- Start unattended operation Step 2
determine focus within search window
Step 3
capture DAPI image and find suitable nuclei
Step 4
capture single color images in several focus
planes and create extended focus images
Step 5
find spots, measure distances and count spots
Step 6
review results
Determine focus
Step 2 Determine focus:
• scan search window in
coarse grid
• measure plane of best
focus at several grid
positions
• reconstruct complete
slide surface by bilinear
interpolation
• follow focus surface
during scan
Scan slide
Step 3 Scan slide
• microscope slide is scanned field by field at lowest
possible magnification without gaps
• objects of interest are automatically identified; gallery
images, object features and positions are recorded
Find suitable nuclei
• reject clusters
(objects with concavities)
• reject irregularly shaped
(e.g. too excentric) nuclei
• reject nuclei that are too
small / too large
Create extended focus image
Step 4 Capture extended forcus image of probe channels
• capture Z-stack
of images
• detect plane of
best focus for
each pixel
• combine pixels
of best focus
planes to extended focus image
Measure feature
Step 5 measure features
• morphometric and intensity features are measured
during the scan
• scanning of unlimited
number of color channels
(fluorochrome labels)
• classifiers combining
selected features can be
used to detect specific cells
Count spots
Step 6 Review results
• identify spots using
appropriate
morphometric and
intensity criteria
• create gallery and
histogram
• spot counts can be
corrected interactively
Display results
• histogram or scatter plot
display of measured
features
• gating for selection of
subpopulations
• relocation of cells detected
in a previous scan after
re-staining
Review results
• on screen review of analyzed cells:
• marked cells
• rejected cells
• deleted cells
- green frame
- blue cross
- red cross
• cells not suitable for analysis are
rejected or deleted
• preselected objects can be automatically relocated at 63x to
capture high resolution images
Features: Cell Morphology
Quantitate irregularity
size ...
Multi-color Probes
• identify spots in up
to six color channels
• create gallery and
histogram
Vysis UroVysion
High Throughput
spot counting in sperm cells hybridized with AneuVysion kit
(chr. 18, X, Y; 20x lens; 1000 cells scored in 10 min)
High throughput - slide feeder
Fusion Detection
• spot position is calculated in relation
to the focus level
and the spot area
• the 3D distance
between two spots
from different color
channels is calculated
Fusion Detection
•
Normal situation:
2R/2G
•
Translocation:
1R / 1G / 1G / 1RG
•
Short distance between
1R and 1G
No. of green spots
No. of red spots
Min. distance between
red and green spots
Fusion Detection
Minimum
spot separation r-g
Spot count green
Ratio Analysis
• isolated cells not
always available for
scoring (eg. detection
of gene amplification
in tissue sections)
• fluorescence ratio
analysis per area
provides information
on amplifications
tissue scanning: HER-2/neu
Feasibility Study:
Automatic HER-2/neu assessment in human
breast cancer tissue specimen
– HER-2/neu (c-erb2) amplification is a prognostic factor
in critical stage II, node-positive patients
(chemotherapy, Herceptin® treatment: yes/no)
– HER-2/neu probe: Vysis PathVysion™ Kit
– Detect FISH signals in formalin-fixed,
paraffin-embedded breast cancer
tissue sections
HER-2/neu– Anticipated Results
HER-2/neu : CEP17 ratio:  2
not amplified
2
HER-2/neu : CEP17 ratio: >
amplified
Crititical: HER-2/neu : CEP17 ratio between 1.8-2.2
tissue scanning: HER-2/neu
Problematic issues in scoring FISH signals in tissues:
• Tissue sections are thicker than nuclei and nuclei are not 'round'
 Automatic identification of nuclei is less robust
• Tumor cells are not evenly distributed
 Suitable cells are found in clusters
• Scoring highly amplified genes
 Individual spots cannot be counted
Alternate two-step approach:
• Tumor areas are defined manually
• Fluorescence ratios between her2 and CEP17 (ref.)
are measured per area not per cell!
HER-2/neu– Location of tumour regions
1. Tumour regions cannot be detected
automatically; they must be selected
based on cell morphology.
Slide area
Solution :
Position list scanning
• Interactive selection of
tumour areas on the slide
or
• Transfer coordinates of
tumor regions from
parallel IHC stained slide
Regions
of interest
tissue scanning: finding cells
2. In most cases it is not
possible to find isolated
single cells.
Solution :
Tile sampling.
• FISH signals are
measured in “tiles”,
not in cells.
• ROI are overlaid with
equi-sized squares
(“tiles”), placed
automatically according
to the counterstain
intensity (lower cut-off)
tissue: Inhomogenous Hybridization
3. Hybridization signals
exhibit a high degree of
variability and artifacts.
Solution :
Automatic rejection of
tiles, field of view and
specimen.
Example: Automatic rejection of an entire field-of-view
due to numerous spot-like artifacts
HER-2 image channel, negative counterstain mask.
tissue scanning: few Tumor Cells
4. Tumour cell areas are
surrounded by normal
cells (90 % of population).
Solution :
Interactive or automatic
definition of tumour
regions.
Green line encloses tumor cell cluster;
red line excludes normal cell within
tissue scanning: HER-2/neu - HSR
5. Spot clusters (HSRs)
cannot be „spot counted“.
Solution :
Automatic classification:
HER-2
Manual
• Non-HSR: use spot
counting algorithm
• HSR: estimate count
from area measurement
after image processing
Auto
CEP-17
HER-2/neu
- Results Non-amplified
HER-2/neu
- Results Amplified
tissue scanning: HER-2/neu
Her2 / HSR
12
Automatic Sample Mean Count
Automatic Sample Mean Count
Her2 / Non-HSR
10
8
6
4
2
0
0
2
4
6
8
10
True Sample Mean Count
12
20
18
16
14
12
10
8
6
4
2
0
0
2
4
6
8
10 12 14 16 18 20
True Sample Mean Count
Good correlation between manual and machine count
Σ: 95 cases
tissue scanning: HER-2/neu
Region of Interest Analysis
12
Automatic HER-2/CEP17 Ratio
Automatic Sample Mean Count
CEP17
10
8
6
4
2
0
14
12
10
8
6
4
2
0
0
2
4
6
8
10
12
True Sample Mean Count
Very good correlation
manual vs. machine count
0
2
4
6
8
10
12
14
True HER-2/CEP17 Ratio
Overestimation of ratio in
manual vs. machine count
Σ: 95 cases
tissue scanning: HER-2/neu
• Semi-automatic assessment of FISH signal scoring is feasible
in tissue sections
• The combination of interactive definition of ROIs and “Tile
sampling” is a robust approach to overcome the problems
associated with FISH in tissue sections
• Clinical trials for FDA approval are on their way (finalized 2Q
2004).
• Preliminary results: overall error rate: approx. 2 %
• Systematic error: 0 %
• About 20 – 30 % of slides are automatically rejected due to
“poor” quality
Tissue Micro Arrays
• identify, map cores
• address, relocate cores
• analyze individual cores
Result presentation
• histograms
• scatterplots
• data tables
• summary data
• patient data
Classifier Training
Training feature provides easy adaptation to
new cell types / probe kits:
automatic capture
of training data set
interactive cell selection
and classification
automatic optimization
of parameters
best possible match of
human and automatic results
Classifier Training
Interactive pre-classification:
•cell selection (counterstain)
•spot counts (signal channels)
red channel
green channel
MetaCyte - Performance
scanning speed depends on application:
slide preparation
cell density
 # of fields of view
cell flatness  # of focal planes
probe
hybridization

size
 optical magnification
brightness
 capture time
# of labels
 # of color channels
quality
 # of cells needed
up to 200-300 cells per minute
MetaCyte - Specifications
• typical scanning speed of six to twenty minutes
per slide depending on the application
• capacity of eight slides per run
(80 slides with slide feeder)
• Software modules for several applications
• seamless integration of isis and ikaros
Download