1. Improved Techniques for Fast Sliding Thin-Slab Volume Visualization
Hello
Janice Turlington…What to say about me???
How to introduce Bill??? relationship (advisor), title,etc.
Improved Techniques for Fast Sliding Thin-Slab Volume Visualization……………………..
Not the correction…NEW techniques!!
Janice Z Turlington*, William E. Higgins*
** Electrical Engineering, **The BioEngineering Program,
and *Computer Science and Engineering
The Pennsylvania State University, University Park, PA 16802
SPIE Medical Imaging 2000, San Diego CA, 13 February

2. Motivation Objectives Show “true” information
Volumetric Computed Tomography (VCT)
The Motivation:
to find small structures in a volume of CT data…
mainly in order to find lymph nodes in the lungs.
Today, I will be showing you the two techniques that came out of my research.
A depth perspective rendering process and traversal-gradient range process.
Though I will not be talking about lymph nodes here…I will say that results that are coming from these techniques are quite exciting and…strongly indicate they are locating the lymph nodes in the lungs.
Starting out, one highly important thing is to build a process that will be used in the medical community …so they need information they feel confident and justified to use.
Meaning to me….that no matter the quality or quantity that …
Processing result in information that is truly in the CT data.
So to do this…data integrity has to be safeguarded…
Use the CT data directly…do no pre-processing.
Use functions that did not bias output.
3. Retain full calculation values throughout procedures (I.e., rounding)
Lastly, techniques are for real-time use in the Virtual Navigator system developed by Dr. Higgins research group. This is being presented at this conference.
…NEED formal name, Bill???, our research group is developing called??? ….need day
Motivation
Locate small structures in 3D CT volume
Objectives
Show “true” information
Confident results for professional use
Fast techniques for real-time use

3. Existing Visualization Techniques
volume/surface rendering4
multi-planar reconstruction2
Looking at the issue of small structures…I looked at existing techniques
Present techniques
projection imaging1
2D image resembling a chest X-ray 1{Hohne87,Napel92}…
Good global view
oblique slice viewing2 (multi-planar reconstruction MPR)
2D cross-sectional data at arbitrary orientations 2{Robb1988,Remy96,McGuinness97}
curved-section reformatting3 (“tube” view):
2D straightened arbitrary path 3{Robb1988,Hara96,Ramaswamy99}
volume or surface rendering4
global 3D view of structures in a 3D image 4{Ney90,Drebin88,Tiede90}
virtual endoscopic rendering5
3D rendering of interior endoluminal structures 5{Vining94,Ramaswamy1999}
projection imaging1
curved-section reformatting3
virtual endoscopic rendering5

4. Existing Limitations dependence on a priori knowledge
Present Visual Limitations
dependence on a prior knowledge
voxel intensity decisions
thresholding
intensity mapping
gradient mapping
partial-volume artifacts
blending artifacts
spatial resolution
voxel boundaries (vs. true image interfaces)
dependence on a priori knowledge
voxel intensity decisions
partial-volume artifacts
voxel boundaries
1{Hohne87,Napel92}
2{Robb1988,Remy96,McGuinness97}
3{Robb1988,Hara96,Ramaswamy99}
4{Ney90,Drebin88,Tiede90} 5{Vining94,Ramaswamy99}

5. Sliding Thin-Slab Visualization
(STS-MIP)
Technique that caught my eye was:
Sandy Napel: Sliding Thin-slab visualization
Because it removed obstructing tissues….
And extended the length of vessels
STS-MIP recognized by Remy-Jardin etc.
Clinical Study of the Value of Sliding-thin-slab Maximum Intensity Projection CT scans in the detection of Mild micronodular patterns Radiology 1996
Evaluated STS MIP projection reconstructions in the assessment of micronodular patterns of low profusion in diffuse infiltrative lung disease.
Due to increased conspicuity of tiny lesions:
1. expected to improve the understanding of lung changes
2. enabled a more precise characterization of distribution of lung changes than conventional CT scans
“Because of the ability to increase the conspicuity of tiny lesions, sliding thin-slab MIP is expected to improve our understanding of lung changes....we found sliding thin-slab MIPs enabled a more precise characterization of distribution of lung changes than conventional CT scans.”
Martine Remy-Jardin, Jacques Remy, Dominique Artaud, Franck Deschildre, Alain Duhamel
Diffuse Infiltrative Lung Disease: Clinical Value of Sliding-thin-slab Maximum Intensity Projection CT scans in the detection of Mild micronodular patterns Radiology 1996
{Napel 92}
improve understanding of lung changes
more precise characterization of distribution changes
{Remy-Jardin et al. 96}

6. Depth-Weighted Maximum (DWmax) Depth-Weighted Maximum (DWmax)
Our New STS Techniques
Depth-Weighted Maximum (DWmax)
Extreme Gradient (EG)
So we now have two new STS rendering techniques
Depth-Weighted Maximum (DWmax)
Extreme Gradient (EG)
Both Fast general-purpose algorithms
Plusssss: you can …Dynamic sequence visualization
Dwmax: a striking and accurate 3D impression
EG: a 4th dimension of information….the extent of change present within a slab
fast general-purpose algorithms
dynamic sequence visualization
accurate 3D impression
slab change volume evolution

7. Sliding Thin-Slab Movement
The sliding thin-slab method
windows a shallow (arbitrarily oriented) sub-volume of data
Window processed then slid forward
Each voxel along a ray is processed individually…
according to user defined mappings of
voxel intensity
voxel gradient
Which then defines voxel opacity
One of three Functions then composite these opacities along a ray:
maximum intensity (MIP)
maximum opacity (MOP)
classic volume rendering (CVR)
volume window column point slab

8. Fast STS Algorithms temporal coherence1 only 2 slices differ
Fast methods apply for ALL methods: Dwmax, EG, Max, Min
Reason: incorporate temporal coherence.
Concept: if one's point of view varies slowly, then consecutive views change little.
With Window processing of sub-volumes of data
2.) windowed volume of data for a slab differs only by two slices from the windowed volume of data for the slab that follows.
3.) many of the computations performed for a slab can be applied to next slab
Two techniques were developed…Depth-Weighted Maximum And EG
Lets look at Dwmax modeling…
After this, say you’ll now discuss Dwmax and EG, and fast algorithms will come out here
temporal coherence1
adjacent views change little
only 2 slices differ
slabi window calculations apply to slab i+1
1{Ramaswamy1999}

9. Depth-Weighted Maximum (DWmax)
Was there other information in the windowed data???
Some changes:
1.) to have the best chance to get small structures
work directly with actual CT scan data….and keep it clean and unaltered
So must: remove intensity biasing…that happens in
thresholding,
intensity mapping,
gradient mapping…
all takes out small detail small structures!!!
Which meant
A. take human preprocessing decisions out of the loop
B. Determine ray values using general decisions based on relative perspective
This went right along with the modeling of DWmax
Model to achieve goalsEnvision sight as a group of progressive planes… natural visual characteristics
Closest plane is most most intense…
Remaining planes fade into the distance till perception is lost.
Vision is defined by maximum opaque object encountered, before vision is lost to distance.
Global factor, such as the side of building, may stop planes shy of farthest visible plane.
For a small distance, projection of view is parallel (like ideal X-ray directed normal to viewing plane).
Now to take it to Dwmax…
Planes = Image slices
Closet plane = Base slice
first slice in processing window (full intensity)
Distance perception is lost = Depth of Vision (dv)
Distance building is encountered = Field of View (ds)
depth of processing window.
Fading = Depth-Weighting
function of dv and ds
Plane at distance of building = End slice in window = last slice processed
Therefore all weighting is done at the slice level and voxels do not need to be individually handled…main factor for fast processing
wi = (depth of view – distancei from base)

10. Depth-Weighting: 1D process
Depth weighting…so at each voxel…weighting determined by slice level is blindly applied
For each voxel in the base slice
Column of window data is processed
From base out
Base slice given full intensity…weight =1
Slices away from base, decreased by number of slice units they are away from base
All normalized by depth of vision
Only data up to the field of view is processed
Maximum weighted voxel values becomes slab point value
Temporal coherence:
Move window up ….new weighted end slice is larger than previous slab, becomes present slab value
If previous base slice was previous slab max, must brute force present slab
If neither end defines process….up-weight previous slab to present slab
Window defines slab data
Depth weights
not dependent on voxel intensity nor neighbors
Weighting is a slice level process….a function of slice’s location in window
weight independent of (x,y) voxel location in slice….same weight for every voxels in slice…
SPEED ADVATAGE Only calculate once then apply to all windows
Set of depth-weights apply to all windows
Why Depth of vision??? Advantages of depth of vision
A. Extending vision beyond the field of view provides a relative distance that effectively varies like-tissue intensities on the different slice to produce depth perceptive view.
B. allows end slice in window to retain values that add legitimate definition to slab..without last slice looses all its ablity to add to information to slab.

11. What you can see it is… because of the slice-level weighting
DWmax Algorithms
volume
sub-volume
Here is the complete code for Dwmax…. Paper has complete detail
Too much to go into now…
it is addressed in the conference proceedings, and
I will be glad to discuss it after the presentation
What you can see it is… because of the slice-level weighting
Quite Condensed
Simplified
Functions are not 3D…
volume level: 1D direction of movement,
window level: 2D slice plane,
column level (voxel level) 1D direction of movement
3. Point Functions only have one decision statement
column .

12. STS-MIP DWmax CT image slice {Napel 1992} {Turlington 2000}
Healthy Human
3D electron-beam CT scan
(no contrast agent)
3mm/0.781mm/0.781mm
Transvere plane CT slice view
STSMIP
Extension of vessels BUT:
NO impression of depth or 3D
overlapping causes limited use of additional structural length information
Only usable as extra or secondary tool
DWmax
3D local structure
Strong impression of Depth
Endoluminal information
STS-MIP
DWmax
{Napel 1992}
{Turlington 2000}

13. DWmax: Transverse Sequence Views
Healthy human: 0.781/0.781mm
Effective view of physical structure of interior tissues,
clear visual perspective into thoracic cavity
External and internal relationship of structures
Peer into airways and see endoluminal structure
s=43
s=49

14. coronal detail DWmax: coronal sequence views projection image
No preprocessing, no segmentation, no contrast agent.
Note soft tissue definition and tracheal detail…extension of vessels…clarity and extension of information 3mm/.0781mm resolution
vessels, airways, soft tissue visible, traversal through front to back
Ds=19 Dv=30
projection image
coronal sequence views
single slice

15. DWmax: sagittal detail
*Can see very helpful in particular cases for close detailed study but especially great for general viewing purposes…initial look at CT
*Depth perception without computation and memory cost
*Fast, real-time worthy…slice level weighting means low computation and minimual memory required)
*Packages large quantiy of CT images for dynamic viewing
*No knowlege of image required…general process
*Only 2 slab size input parameters (unrelated to intensity characteristics)
*Slabs contain the entire full range of image data…so only single run per slab size is necessary
Dwmax: Easy & General
depth perception and interior information
fast real-time processing1
general process
no specific data type required
no knowledge of image needed
only 2 input parameters
Practical>>>efficient packaging for large quantity of data
New helical multi-detector scanner highly increases number of images
highly effective when dynamically viewed
Striking or abnormal changes “jump out” like beacons…grab one’s immediate attention
Quick and effective recognition
entire full range of image data is available
Only one run for all needed information
Bonus similarity to cross-sectional imaging
minimal learning curve + direct integration of experience
(Note Naidich comments)
base slice
base slice

17. Mapping of max tissue-density changes in slab
Extreme Gradient (EG)
Mapping of max tissue-density changes in slab
Extreme Gradient
A mapping of the maximum changes in tissue density occurring within the depth of a slab.
An EG point = magnitude of the difference between maximum - minimum intensity in the slices within a STS window.
EG a function of image extremes, structure size, and slab size
s(x,y) = max x,y{ } – max x,y{ }

18. volume sub-volume column EG Algorithms EG Characteristics
Tissues fall into well-separated unique bins (inherent from distinct separation between tissue intensities).
tissue value are consistent over a large range of slab sizes (because function of extremes).
boarders outline pattern change
structural size change (slab level)
shape changes within a slab (slab level)
evolutional change in a volume.
Shape/intensity change proportional to value/pattern
So Radical change in structure or intensity easily seen in EG value and pattern change
Entire range of image change is contained in a slab
column

19. Extreme Gradient large change EG not windowed Function of:
Extreme Gradient contains all change information
Depending on the tissue of interest or level of changes being studied
An EG slab can be windowed to show this
EG: 4th dimension of DWmax
study tissue dynamics
traverse volume consecutive slab (volume evolution)
hold base vary slab size (detect entrace of details)
pattern boundaries
structural size and shape changes within a slab
evolutional change in a volume.
detect smaller structures hidden or buried in DWmax
detect sudden or irregular grow
The top view shows EG slide viewed with normal gray value display map
Window for low values, interior lung tissue and steady bone is shown most intensely.
Window for moderate values, highlights the change in heart size and vessles introduced
Window for large values, captures the branching of a bronchus in this slab, rib bones curving
When viewing Dwmax, this EG feature would be useful to filter the Dwmax view
i.e., soft tissue if focus, or small vessle location is the focus…
EG: Consistent & distinct values
Practical applications:
focused-dynamic viewing (when viewing DWmax display points windowed using EG, hold all other points constant)
tissue identification
structure segmentation
image extremes
structure size
slab size
small change
moderate change
EG not windowed
large change

20. Endoluminal Border Detection
Pratical application for combination of techniques
Edge functionality of EG + DWmax depth
Parallel ray processing eliminates perspective concerns
Overlay EG onto Dwmax:…DETECT
Wall thickness
Tissue irregularities
Abnormal growth (seen here)
This slide:
Human lung caner case:
Vary size of EG and keep Constant Dwmax…
outline of EG show details of trachea changes at varying depths
Ds=10, Dv=20
Shallow EG = 4…same size EG=10
merging DWmax and EG

21. Computation Gains of New Techniques
inversely proportionally to slab-size (ds)

22. Current Reconstruction Techniques
{Naidich et al. 97}
cross-sectional imaging is CT gold standard
axial images alone enable remarkably accurate assessment
true paradigm shift requires comparison to traditional methods
David Naidich James Gruden Georgeann McGuinness
Dorothy McCauley Meenaskshi Bhalla
Journal of Thoracic Imaging 1997
Volumetric (Helical/spiral) CT (VCT) of the airways
purpose of the paper…review various reconstruction techniques, and based on current knowledge, place them in a appropriate clinical context.
“At present, cross-sectional imaging remains the CT ‘gold standard’ for assessing pathology of the airways and is clearly complimentary to fiberoptic bronchoscopy”
“Despite the limitations outlined herein, axial images alone enable remarkably accurate assessment of the central and peripheral airways…
Any attempt to achieve a true ‘paradigm shift’ will require comparison of these new techniques to the traditional methods of airway evaluation.”

23. currently under development
lymph node detection
S=39

24. DWmax
EG: soft tissue
lymph node?
EG: high density

25. Conclusion: New Techniques
fast, real-time
small structures visible
depth perception
true information
fast, real-time worthy
slice level weighting
standard axis directed processing + parallel processing
temporal coherence
confidence
similar to cross-sectional imaging
companions technique compliment slab information
justification
interior details
detection of non-visible structures
evolution information
visual depth perception
small structure detail
full relative view
PLUS: EG definition and detection
true information from safeguarded data
intensity-independent weighting
Or (x,y) independent weighting (location of voxel in slice)
No parallel ray processing + standard-axis directed processing

26. My thanks to: Dr. William E. Higgins The Whitaker Foundation
Dr. K. Kirk Shung
Dr. Cheng Dong
My Father
“For nothing is hidden, except to be revealed;
nor has been secret, but that it should come to light.”
Mark 4:22

27. Virtual Bronchoscopic Approach Combining 3D CT and Endoscopic Video
2:40 pm Today…California Room
A. J. Sherbondy
Virtual Bronchoscopic Approach
Combining 3D CT and Endoscopic Video
Physiology and Functions from Multidimensional Images
Conference 3978
5:30 to 7:30 pm Today…California Room
G. McLennan
The Place of Virtual Bronchoscopy in Clinical Practice: Barriers and Solutions

28. Occurs in less than 2%of points
DWmax error
NOT visible
Occurs in less than 2%of points
must occur simultaneously:
closer to base than voxel of previous slab’s max
depth-weighted voxel must differ:
< 5.5% of 16-bit previous-slab point (=0.7% 8-bit)
if error occurs:
errror < 5 HU (16-bit) or 0 HU (8-bit)